The Perl Toolchain Summit needs more sponsors. If your company depends on Perl, please support this very important event.
Changes 20440
Decoder.xs 110387
MANIFEST 239
META.json 22
META.yml 1616
Makefile.PL 122
author_tools/bench.pl 515
author_tools/different_sereal_docs.sh 038
author_tools/hobodecoder.pl 1038
author_tools/update_from_header.pl 1250
const-c.inc 81229
inc/Sereal/BuildTools.pm 31155
lib/Sereal/Decoder/Constants.pm 31163
lib/Sereal/Decoder.pm 32218
lib/Sereal/Performance.pm 0265
miniz.c 04922
miniz.h 07
snappy/csnappy_internal_userspace.h 1206
srl_common.h 055
srl_decoder.c 221581
srl_decoder.h 2075
srl_protocol.h 922
t/004_testset.t 036
t/020_incremental.t 58
t/030_looks_like_sereal.t 1248
t/040_special_vars.t 043
t/060_each.t 038
t/070_alias_options.t 058
t/080_set_readonly.t 085
t/100_roundtrip.t 380
t/101_roundtrip_v1.t 380
t/150_dec_exception.t 216
t/210_bulk_readonly.t 034
t/550_decode_into.t 034
t/700_roundtrip/v1/plain.t 028
t/700_roundtrip/v1/plain_canon.t 027
t/700_roundtrip/v1/snappy.t 027
t/700_roundtrip/v1/snappy_canon.t 027
t/700_roundtrip/v2/dedudep_strings.t 029
t/700_roundtrip/v2/freeze_thaw.t 029
t/700_roundtrip/v2/plain.t 028
t/700_roundtrip/v2/plain_canon.t 027
t/700_roundtrip/v2/readonly.t 029
t/700_roundtrip/v2/snappy.t 027
t/700_roundtrip/v2/snappy_canon.t 027
t/700_roundtrip/v2/snappy_incr.t 029
t/700_roundtrip/v2/snappy_incr_canon.t 027
t/700_roundtrip/v2/sort_keys.t 029
t/700_roundtrip/v3/dedudep_strings.t 029
t/700_roundtrip/v3/freeze_thaw.t 029
t/700_roundtrip/v3/plain.t 028
t/700_roundtrip/v3/plain_canon.t 027
t/700_roundtrip/v3/readonly.t 029
t/700_roundtrip/v3/snappy.t 027
t/700_roundtrip/v3/snappy_canon.t 027
t/700_roundtrip/v3/snappy_incr.t 029
t/700_roundtrip/v3/snappy_incr_canon.t 027
t/700_roundtrip/v3/sort_keys.t 029
t/700_roundtrip/v3/zlib.t 032
t/700_roundtrip/v3/zlib_force.t 033
t/902_bad_input.t 015
t/903_reentrancy.t 047
t/lib/Sereal/BulkTest.pm 1119
t/lib/Sereal/TestSet.pm 121428
typemap 31
65 files changed (This is a version diff) 96211191
@@ -1,209 +1,45 @@
 Revision history for Perl extension Sereal-Decoder
 
-* Warning: For a seamless upgrade, upgrade to version 2
-*          of the decoder before upgrading to version 2 of the
+* Warning: For a seamless upgrade, upgrade to version 3
+*          of the decoder before upgrading to version 3 of the
 *          encoder!
 
-2.03 Tue Jan  7 20:00 2014 (AMS time)
-  - (Hopefully) final fixes to FREEZE/THAW functionality:
-    => Add safe assertion to make sure that we don't segfault on invalid
-       data.
-    => Fix encoding/decoding of data structures with repeated references
-       to the same instance of a class that has FREEZE/THAW hooks.
-       Thanks to Christian Hansen for a test case.
-  - Distribution dependency fix.
+3.002 Aug 20 2014
+  * Summary of changes from 3.001 - 3.002
+  - Introduce "set_readonly" option to decoder
+  - Introduce "set_readonly_scalars" option to decoder
+
+  * Bug fixes
+  - Fix assertion fails in DEBUG perls (Brian Fraser)
+  - Fix segfault with "into" interface when decoding references and
+    strings into the same target SV. See new test t/550_decode_into.t
+
+  * Test Infra Changes
+  - Split up bulk tests to speed up testing and make it easier
+    to see when a failure is restricted to a specific option.
+
+  * Big-Endian Support
+  - Improved support for Big-Endian machines. We now build and pass test
+    on Sparc and HP-UX and other platforms with big-endian or strict
+    alignedness requirements. Much thanks to Jarkko Hietaniemi,
+    Gregor Herrmann, and H. Merijn Brand for for their assistance with
+    this.
+  - We still have issues with s390x (Z/Os) with Sereal. If someone wants
+    to help it would be appreciated.
+
+3.001
+  - Upgrade to version 3 of the protocol
+    * Add Zlib compression support to the protocol
+    * Add Zlib support to Encoder/Decoder
+    * Add CANONICAL_UNDEF tag to represent PL_sv_undef
+    * Change magic header so it is trivial to detect
+      and reject utf8 encoded Sereal packets.
+  - Change how we serialize variables which have more
+    than one of IOK, NOK, and POK set. In general we
+    favour the numeric form, with the exception of when
+    all three are set when we favour the string form
+    (this is to handle engineering notation like "0E0"
+    where numeric and string equivalence may differ)
+
+Full change history available at https://github.com/Sereal/Sereal
 
-2.02 Mon Jan  6 15:00 2014 (AMS time)
-  - Fundamental fixes for FREEZE/THAW support in previous Sereal v2
-    releases. If you plan to use FREEZE/THAW, make sure you have 2.02
-    or better (dito for the encoder).
-
-2.01 Tue Dec 31 08:15 2013 (AMS time)
-  - Promoting changes from 0.37 to 2.00_03 to a stable release.
-    (This being the first protocol v2 stable release.)
-  - Minor performance tweaks.
-
-2.00_03 Sun Dec 29 10:33 2013 (AMS time)
-  - FREEZE/THAW hooks for object serialization.
-  - Test improvements (allowing for partial parallel run)
-
-2.00_02 Mon Oct 28 19:23 2013 (AMS time)
-  - Fix Test::Warn dependency problem of 2.00_01.
-
-2.00_01 Tue Oct 1 07:34 2013 (AMS time)
-  - NEW PROTOCOL VERSION: V2
-  - User-data in header functionality: You may embed arbitrary
-    Sereal-serializable data in a document header. The document
-    header isn't compressed, so this is ideal for retrieving
-    small chunks of meta-data (eg. routing information) without
-    having to deserialize the entire document.
-  - Relocatable Sereal document bodies
-  - Encoder never emits non-incremental Snappy encoding for V2
-  - Offsets now 1-based in relocatable format, not 0
-  - Fixed VERY obscure (and rare) memory leak.
-  - Fixed VERY obscure crash on invalid data.
-  - Improved error messages
-  - Remove warning about Sereal not being production-grade
-    (because it IS).
-  - C89/Windows fixes (bulk88)
-  - 5.18 compat: Skip test failing due to hash-randomization (Zefram)
-
-0.37 Mon Sep 2 07:40 2013 (AMS time)
-  - Remove spurious SvREFCOUNT_dec, previously causing
-    memory management problems on some vary invalid
-    input data.
-  - Fixed a memory leak on exception.
-
-0.36 Tue May 7 12:00 2013 (AMS time)
-  [changelog for encoder and decoder both]
-  - Add "incremental" option to decoder for easier decoding of
-    multiple sereal documents in one buffer.
-  - Make snappy and snappy_incr options mutually exclusive.
-  - Feature: Implement aliasing for deduping (aliased_dedupe_strings)
-
-0.35 Mon Apr 1 11:50 2013 (AMS time)
-  - Add new no_bless_objects option from Simon Bertrang.
-
-0.34 Sat Mar 23 18:59:18 2013 (AMS time)
-  - Fixup Manifest
-
-0.33 Sun Feb 17 17:26 2013 (AMS time)
-  - dummy release to keep in sync with encoder.
-
-0.32 Sun Feb 17 15:06 2013 (AMS time)
-  - dummy release to keep in sync with encoder.
-
-0.31 Sun Feb 17 15:06 2013 (AMS time)
-  - Daniel Dragan <bulk88@hotmail.com> spent a bunch of time
-    digging into the weird problems we were having with Snappy
-    encoded data on Windows on certain builds. Turned out that
-    it was right broken, and worked sometimes purely by chance.
-    He kindly provided a patch.
-
-0.30 Wed Feb 09 06:24 2013 (AMS time)
-  - dummy release to keep in sync with encoder.
-
-0.29 Sat Feb 09 18:09 2013 (AMS time)
-  - Change a gcc'ism, #warning to #pragma message
-    (which also works on my machines gcc)
-
-0.28 Sat Feb 09 16:20 2013 (AMS time)
-  - More fixups for building on Win32/C89 compilers
-  - Fix up handling of regexps on perl 5.10.1
-
-0.27 Sat Feb 09 12:58 2013 (AMS time)
-  - Various fixups to improve building on Win32
-  - Fix C89 violations
-  - Fixed Changelog order (why would I ever have listed oldest first?)
-
-0.26 - unreleased
-
-0.25 Tue Jan 22 18:00 2013
-  - Various compatibility fixes with old versions of Perl.
-    Specifically, fixes to regular expression handling that should help
-    with 5.10 support, as well as fixes that should improve the status
-    quo on 5.8.
-  - Potential fix for a leak wrt. regular expression support.
-  - Fewer compiler warnings on 32bit/gcc.
-
-0.24 Thu Jan 10 08:42 2013
-  - Test logic fixes: These prevented builds if the encoder wasn't
-    installed (or updated) before.
-
-0.23 Tue Jan 08 07:23 2013
-  * Important bug fix release *
-  - Important: support for the incremental-parsing Snappy format that
-    is enabled with the encoder's (versions 0.23 and up)
-    'snappy_incr' option.
-
-0.22 - unreleased
-
-0.21 Wed Jan 02 08:00 2013
-  - Option 'max_num_hash_entries' to limit the number of accepted
-    hash entries.
-
-0.20 - unreleased
-
-0.19  Thu Nov 23 07:00 2012
-  - Add configurable recursion limit to the decoder
-  - Downgrade UTF8 Sereal serialized strings before decoding them
-
-0.18 - unreleased
-0.17 - unreleased
-0.16 - unreleased
-
-0.15  Wed Oct 17 13:00 2012
-  - Decoder option validate_utf8
-    With this option (turned off by default for performance reasons)
-    the decoder will emit an error on malformed UTF-8. This will
-    *not* check for invalid code points; this can be added later, but Perls
-    that are recent enough will warn on those anyway.
-  - Thread-safety fix on Perls >= 5.8.7. Sereal is still not thread-safe
-    on older Perls 
-
-0.14 - unreleased
-
-0.13  Tue Oct  2 15:00 2012
-  * Important bug fix, important gotcha! *
-  - Work around bug in perls <= 5.8.8:
-      Multiple references to the same object that has overloading were
-      dropping overloadedness on the second and following occurrence.
-    => The work-around for this old perl bug means that we have to
-       bless objects as we find them during deserialization. By default,
-       on newer perls, we only do the bless()ing once after being done
-       with serialization. We normally do this for security reasons
-       (destructors!). This means that for perls <= 5.8.8, we necessarily
-       have the same (very limited) security problem that Storable has
-       at all times. Sucks, but better to have it working on 5.8.8 at all.
-    => For newer perls, we DO NOT have that security problem!
-    => The work-around also slows down operation on 5.8.8 and older. Keep
-       that in mind when comparing serialization techniques. Thank you.
-
-0.12 - unreleased
-
-0.11  Tue Sep 18 13:00 2012
-  - 5.8.5 fixes.
-  - Fixes to other languages' reference data output.
-
-0.10  Mon Sep 17 13:40 20120
-  - New 'refuse_objects' option causes the decoder to throw an
-    exception when encountering objects in the input data.
-    See the "ROBUSTNESS" section in the documentation for why that
-    is a good idea.
-  - Fixes to error messages.
-  - Compiler warning fixes.
-
-0.09  Fri Sep 14 09:40 2012
-  - Removes use of variadic macros for compatibility with inferior
-    compilers.
-  - "looks_like_sereal" method/function for a quick way to check for
-    valid Sereal *headers*.
-  - More liberal set of encoder versions we run the full tests with.
-
-0.08  Thu Sep 13 17:00 2012
-  - Documentation fixes
-  - Better tests
-
-0.07  Tue Sep 11 14:00 2012
-  - Improved robustness to bad input data.
-  - Test fixes for threaded perls (likely working around a bug in Perl)
-  - Improved documentation.
-
-0.06  Mon Sep 10 11:00 2012
-  - First public release (CPAN).
-  - Beta quality software.
-
-0.05  Fri Sep  7 14:00 2012
-  - internal release.
-
-0.04  Thu Sep  6 16:00 2012
-  - internal release.
-
-0.03  Tue Sep  4 17:09 2012
-  - internal release.
-
-0.02  Tue Aug  8 17:09 2012
-  - internal release.
-
-0.01  Tue Aug  8 17:09 2012
-  - original version; internal release.
@@ -8,16 +8,399 @@
 
 #include "ppport.h"
 
+#include "srl_common.h"
 #include "srl_decoder.h"
 #include "srl_protocol.h"
 
 /* Generated code for exposing C constants to Perl */
 #include "const-c.inc"
 
+#ifndef GvCV_set
+# define GvCV_set(gv, cv) (GvCV(gv) = (cv))
+#endif
+
+#ifndef PERL_ARGS_ASSERT_CROAK_XS_USAGE
+#define PERL_ARGS_ASSERT_CROAK_XS_USAGE assert(cv); assert(params)
+
+/* prototype to pass -Wmissing-prototypes */
+STATIC void
+S_croak_xs_usage(pTHX_ const CV *const cv, const char *const params);
+
+STATIC void
+S_croak_xs_usage(pTHX_ const CV *const cv, const char *const params)
+{
+    const GV *const gv = CvGV(cv);
+
+    PERL_ARGS_ASSERT_CROAK_XS_USAGE;
+
+    if (gv) {
+        const char *const gvname = GvNAME(gv);
+        const HV *const stash = GvSTASH(gv);
+        const char *const hvname = stash ? HvNAME(stash) : NULL;
+
+        if (hvname)
+            Perl_croak_nocontext("Usage: %s::%s(%s)", hvname, gvname, params);
+        else
+            Perl_croak_nocontext("Usage: %s(%s)", gvname, params);
+    } else {
+        /* Pants. I don't think that it should be possible to get here. */
+        Perl_croak_nocontext("Usage: CODE(0x%"UVxf")(%s)", PTR2UV(cv), params);
+    }
+}
+
+#ifdef PERL_IMPLICIT_CONTEXT
+#define croak_xs_usage(a,b)     S_croak_xs_usage(aTHX_ a,b)
+#else
+#define croak_xs_usage          S_croak_xs_usage
+#endif
+
+#endif
+
+
+#if defined(cv_set_call_checker) && defined(XopENTRY_set)
+# define USE_CUSTOM_OPS 1
+#else
+# define USE_CUSTOM_OPS 0
+#endif
+
+#define OPOPT_DO_BODY       (1<<0)
+#define OPOPT_DO_HEADER     (1<<1)
+#define OPOPT_OFFSET        (1<<2)
+#define OPOPT_OUTARG_BODY   (1<<3)
+#define OPOPT_OUTARG_HEADER (1<<4)
+#define OPOPT_LOOKS_LIKE    (1<<5)
+
+#define pp1_sereal_decode(opopt) THX_pp1_sereal_decode(aTHX_ opopt)
+static void
+THX_pp1_sereal_decode(pTHX_ U8 opopt)
+{
+    bool need_retvalue = GIMME_V != G_VOID;
+    SV *decoder_ref_sv, *decoder_sv, *src_sv;
+    UV offset;
+    SV *body_into, *header_into;
+    srl_decoder_t *decoder;
+    char *stash_name;
+    dSP;
+
+    header_into = expect_false(opopt & OPOPT_OUTARG_HEADER)
+                  ? POPs
+                  : expect_false(opopt & OPOPT_DO_HEADER) ? sv_newmortal() : NULL;
+    body_into = expect_false(opopt & OPOPT_OUTARG_BODY)
+                ? POPs
+                : expect_true(opopt & OPOPT_DO_BODY) ? sv_newmortal() : NULL;
+
+    offset = expect_false(opopt & OPOPT_OFFSET) ? SvUVx(POPs) : 0;
+    src_sv = POPs;
+    decoder_ref_sv = POPs;
+    PUTBACK;
+
+    if (!expect_true(
+          decoder_ref_sv &&
+          SvROK(decoder_ref_sv) &&
+          (decoder_sv = SvRV(decoder_ref_sv)) &&
+          SvOBJECT(decoder_sv) &&
+          (stash_name = HvNAME(SvSTASH(decoder_sv))) &&
+          !strcmp(stash_name, "Sereal::Decoder")
+       ))
+    {
+        croak("handle is not a Sereal::Decoder handle");
+    }
+
+    decoder = (srl_decoder_t *)SvIV(decoder_sv);
+    if (expect_true(opopt & OPOPT_DO_BODY)) {
+        if (opopt & OPOPT_DO_HEADER) {
+             srl_decode_all_into(aTHX_ decoder, src_sv, header_into,
+                                 body_into, offset);
+        } else {
+            srl_decode_into(aTHX_ decoder, src_sv, body_into, offset);
+        }
+    } else {
+        srl_decode_header_into(aTHX_ decoder, src_sv, header_into, offset);
+    }
+
+    if (expect_true(need_retvalue)) {
+        SV *retvalue;
+        if (expect_true(opopt & OPOPT_DO_BODY)) {
+            if (opopt & OPOPT_DO_HEADER) {
+                AV *retav = newAV();
+                retvalue = newRV_noinc((SV*)retav);
+                sv_2mortal(retvalue);
+                av_extend(retav, 1);
+                av_store(retav, 0, SvREFCNT_inc(header_into));
+                av_store(retav, 1, SvREFCNT_inc(body_into));
+            } else {
+                retvalue = body_into;
+            }
+        } else {
+            retvalue = header_into;
+        }
+        SPAGAIN;
+        XPUSHs(retvalue);
+        PUTBACK;
+    }
+}
+
+#define pp1_looks_like_sereal() THX_pp1_looks_like_sereal(aTHX)
+static void
+THX_pp1_looks_like_sereal(pTHX)
+{
+    dSP;
+    SV *data= TOPs;
+    /* shoud this be SvPOK() maybe? */
+    if ( SvOK(data) ) {
+        STRLEN len;
+        char *strdata= SvPV(data, len);
+        IV ret= srl_validate_header_version_pv_len(aTHX_ strdata, len);
+        if ( ret < 0 ) {
+            SETs(&PL_sv_no);
+        } else {
+            SETs(newSViv(ret & SRL_PROTOCOL_VERSION_MASK));
+        }
+    } else {
+        SETs(&PL_sv_no);
+    }
+}
+
+#if USE_CUSTOM_OPS
+
+static OP *
+THX_pp_sereal_decode(pTHX)
+{
+    pp1_sereal_decode(PL_op->op_private);
+    return NORMAL;
+}
+
+static OP *
+THX_pp_looks_like_sereal(pTHX)
+{
+    pp1_looks_like_sereal();
+    return NORMAL;
+}
+
+static OP *
+THX_ck_entersub_args_sereal_decoder(pTHX_ OP *entersubop, GV *namegv, SV *ckobj)
+{
+
+   /* pull apart a standard entersub op tree */
+
+    CV *cv = (CV*)ckobj;
+    I32 cv_private = CvXSUBANY(cv).any_i32;
+    U8 opopt = cv_private & 0xff;
+    U8 min_arity = (cv_private >> 8) & 0xff;
+    U8 max_arity = (cv_private >> 16) & 0xff;
+    OP *pushop, *firstargop, *cvop, *lastargop, *argop, *newop;
+    int arity;
+
+    /* Walk the OP structure under the "entersub" to validate that we
+     * can use the custom OP implementation. */
+
+    entersubop = ck_entersub_args_proto(entersubop, namegv, (SV*)cv);
+    pushop = cUNOPx(entersubop)->op_first;
+    if ( ! pushop->op_sibling )
+        pushop = cUNOPx(pushop)->op_first;
+    firstargop = pushop->op_sibling;
+
+    for (cvop = firstargop; cvop->op_sibling; cvop = cvop->op_sibling) ;
+
+    lastargop = pushop;
+    for (
+        arity = 0, lastargop = pushop, argop = firstargop;
+        argop != cvop;
+        lastargop = argop, argop = argop->op_sibling
+    ){
+        arity++;
+    }
+
+    if (expect_false(arity < min_arity || arity > max_arity))
+        return entersubop;
+
+    /* If we get here, we can replace the entersub with a suitable
+     * custom OP. */
+
+    if (arity > min_arity && (opopt & OPOPT_DO_BODY)) {
+        opopt |= OPOPT_OUTARG_BODY;
+        min_arity++;
+    }
+
+    if (arity > min_arity)
+        opopt |= OPOPT_OUTARG_HEADER;
+
+    pushop->op_sibling = cvop;
+    lastargop->op_sibling = NULL;
+    op_free(entersubop);
+    newop = newUNOP(OP_NULL, 0, firstargop);
+    newop->op_type    = OP_CUSTOM;
+    newop->op_private = opopt;
+    newop->op_ppaddr = opopt & OPOPT_LOOKS_LIKE ? THX_pp_looks_like_sereal : THX_pp_sereal_decode;
+    return newop;
+}
+
+#endif /* USE_CUSTOM_OPS */
+
+static void
+THX_xsfunc_sereal_decode(pTHX_ CV *cv)
+{
+    dMARK;
+    dSP;
+    SSize_t arity = SP - MARK;
+    I32 cv_private = CvXSUBANY(cv).any_i32;
+    U8 opopt = cv_private & 0xff;
+    U8 min_arity = (cv_private >> 8) & 0xff;
+    U8 max_arity = (cv_private >> 16) & 0xff;
+
+    if (arity < min_arity || arity > max_arity)
+        croak("bad Sereal decoder usage");
+    if (arity > min_arity && (opopt & OPOPT_DO_BODY)) {
+        opopt |= OPOPT_OUTARG_BODY;
+        min_arity++;
+    }
+    if (arity > min_arity)
+        opopt |= OPOPT_OUTARG_HEADER;
+
+    pp1_sereal_decode(opopt);
+}
+
+static void
+THX_xsfunc_looks_like_sereal(pTHX_ CV *cv)
+{
+    dMARK;
+    dSP;
+    SSize_t arity = SP - MARK;
+    I32 cv_private = CvXSUBANY(cv).any_i32;
+    U8 max_arity = (cv_private >> 16) & 0xff;
+
+    if (arity < 1 || arity > max_arity)
+        croak_xs_usage(cv, max_arity == 1 ? "data" : "[invocant,] data");
+    if(arity == 2) {
+        SV *data = POPs;
+        SETs(data);
+        PUTBACK;
+    }
+    pp1_looks_like_sereal();
+}
+
+
 
 MODULE = Sereal::Decoder        PACKAGE = Sereal::Decoder
 PROTOTYPES: DISABLE
 
+BOOT:
+{
+    struct {
+        char const *name_suffix;
+        U8 opopt;
+    } const funcs_to_install[] = {
+        { "",                           OPOPT_DO_BODY },
+        { "_only_header",               OPOPT_DO_HEADER },
+        { "_with_header",               (OPOPT_DO_BODY|OPOPT_DO_HEADER) },
+        { "_with_offset",               (OPOPT_DO_BODY|OPOPT_OFFSET) },
+        { "_only_header_with_offset",   (OPOPT_DO_HEADER|OPOPT_OFFSET) },
+        { "_with_header_and_offset",    (OPOPT_DO_BODY|OPOPT_DO_HEADER|OPOPT_OFFSET) },
+         /*012345678901234567890123*/
+    }, *fti;
+    int i;
+#if USE_CUSTOM_OPS
+    {
+        XOP *xop;
+        Newxz(xop, 1, XOP);
+        XopENTRY_set(xop, xop_name, "sereal_decode_with_object");
+        XopENTRY_set(xop, xop_desc, "sereal_decode_with_object");
+        XopENTRY_set(xop, xop_class, OA_UNOP);
+        Perl_custom_op_register(aTHX_ THX_pp_sereal_decode, xop);
+    }
+#endif /* USE_CUSTOM_OPS */
+    for (i = sizeof(funcs_to_install)/sizeof(*fti); i--; ) {
+#       define LONG_CLASS_FMT "Sereal::Decoder::sereal_decode%s_with_object"
+        char name[sizeof(LONG_CLASS_FMT)+24];
+        char proto[7], *p = proto;
+        U8 opopt;
+        I32 cv_private;
+        GV *gv;
+        CV *cv;
+
+        fti = &funcs_to_install[i];
+        opopt = fti->opopt;
+        /*
+         * The cv_private value incorporates flags describing the operation to be
+         * performed by the sub and precomputed arity limits.  0x020200 corresponds
+         * to min_arity=2 and max_arity=2.  The various additions to cv_private
+         * increment one or both of these sub-values.
+
+         * The six subs created there share a single C body function, and are
+         * differentiated only by the option flags in cv_private.  The custom ops
+         * likewise share one op_ppaddr function, and the operations they perform
+         * are differentiated by the same flags, stored in op_private.
+         */
+        cv_private = opopt | 0x020200;
+
+        /* Yes, the subs have prototypes.  The protoypes have no effect when the
+         * subs are used as methods, so there's no break of compatibility for those
+         * using the documented API.  There is a change that could be detected by
+         * code such as "Sereal::Decoder::decode($dec, @v)", that uses the methods
+         * directly in an undocumented way.
+         *
+         * The prototype, specifically the putting of argument expressions into
+         * scalar context, is required in order to be able to resolve arity at
+         * compile time.  If this wasn't done, there would have to be a pushmark
+         * op preceding the argument ops, and pp_sereal_decode() would need the
+         * same code as xsfunc_sereal_decode() to check arity and resolve the
+         * optional-parameter flags.
+         */
+        *p++ = '$';
+        *p++ = '$';
+
+        if (opopt & OPOPT_OFFSET) {
+            *p++ = '$';
+            cv_private += 0x010100;
+        }
+        *p++ = ';';
+        if (opopt & OPOPT_DO_BODY) {
+            *p++ = '$';
+            cv_private += 0x010000;
+        }
+        if (opopt & OPOPT_DO_HEADER) {
+            *p++ = '$';
+            cv_private += 0x010000;
+        }
+        *p = 0;
+        /* setup the name of the sub */
+        sprintf(name, LONG_CLASS_FMT, fti->name_suffix);
+        cv = newXSproto_portable(name, THX_xsfunc_sereal_decode, __FILE__,
+                proto);
+        CvXSUBANY(cv).any_i32 = cv_private;
+#if USE_CUSTOM_OPS
+        cv_set_call_checker(cv, THX_ck_entersub_args_sereal_decoder, (SV*)cv);
+#endif /* USE_CUSTOM_OPS */
+        sprintf(name, "Sereal::Decoder::decode%s", fti->name_suffix);
+        gv = gv_fetchpv(name, GV_ADDMULTI, SVt_PVCV);
+        GvCV_set(gv, cv);
+    }
+}
+
+BOOT:
+{
+#if USE_CUSTOM_OPS
+    {
+        XOP *xop;
+        Newxz(xop, 1, XOP);
+        XopENTRY_set(xop, xop_name, "scalar_looks_like_sereal");
+        XopENTRY_set(xop, xop_desc, "scalar_looks_like_sereal");
+        XopENTRY_set(xop, xop_class, OA_UNOP);
+        Perl_custom_op_register(aTHX_ THX_pp_looks_like_sereal, xop);
+    }
+#endif /* USE_CUSTOM_OPS */
+    {
+        CV *cv;
+        cv = newXSproto_portable("Sereal::Decoder::scalar_looks_like_sereal", THX_xsfunc_looks_like_sereal, __FILE__, "$");
+        CvXSUBANY(cv).any_i32 = 0x010100 | OPOPT_LOOKS_LIKE;
+#if USE_CUSTOM_OPS
+        cv_set_call_checker(cv, THX_ck_entersub_args_sereal_decoder, (SV*)cv);
+#endif /* USE_CUSTOM_OPS */
+        cv = newXS("Sereal::Decoder::looks_like_sereal", THX_xsfunc_looks_like_sereal, __FILE__);
+        CvXSUBANY(cv).any_i32 = 0x020100 | OPOPT_LOOKS_LIKE;
+    }
+}
+
 srl_decoder_t *
 new(CLASS, opt = NULL)
     char *CLASS;
@@ -33,88 +416,6 @@ DESTROY(dec)
   CODE:
     srl_destroy_decoder(aTHX_ dec);
 
-
-void
-decode(dec, src, into = NULL)
-    srl_decoder_t *dec;
-    SV *src;
-    SV *into;
-  PPCODE:
-    ST(0)= srl_decode_into(aTHX_ dec, src, into, 0);
-    XSRETURN(1);
-
-AV *
-decode_with_header(dec, src, body_into = NULL, header_into = NULL)
-    srl_decoder_t *dec;
-    SV *src;
-    SV *body_into;
-    SV *header_into;
-  CODE:
-    if (header_into == NULL)
-        header_into = sv_newmortal();
-    if (body_into == NULL)
-        body_into = sv_newmortal();
-    srl_decode_all_into(aTHX_ dec, src, header_into, body_into, 0);
-    RETVAL = newAV();
-    sv_2mortal((SV *)RETVAL);
-    av_extend(RETVAL, 1);
-    av_store(RETVAL, 0, header_into);
-    av_store(RETVAL, 1, body_into);
-  OUTPUT: RETVAL
-
-AV *
-decode_with_header_and_offset(dec, src, offset, body_into = NULL, header_into = NULL)
-    srl_decoder_t *dec;
-    SV *src;
-    UV offset;
-    SV *body_into;
-    SV *header_into;
-  CODE:
-    if (header_into == NULL)
-        header_into = sv_newmortal();
-    if (body_into == NULL)
-        body_into = sv_newmortal();
-    srl_decode_all_into(aTHX_ dec, src, header_into, body_into, offset);
-    RETVAL = newAV();
-    sv_2mortal((SV *)RETVAL);
-    av_extend(RETVAL, 1);
-    av_store(RETVAL, 0, header_into);
-    av_store(RETVAL, 1, body_into);
-  OUTPUT: RETVAL
-
-
-
-void
-decode_only_header(dec, src, header_into = NULL)
-    srl_decoder_t *dec;
-    SV *src;
-    SV *header_into;
-  PPCODE:
-    ST(0)= srl_decode_header_into(aTHX_ dec, src, header_into, 0);
-    XSRETURN(1);
-
-
-void
-decode_with_offset(dec, src, offset, into = NULL)
-    srl_decoder_t *dec;
-    SV *src;
-    UV offset;
-    SV *into;
-  PPCODE:
-    ST(0)= srl_decode_into(aTHX_ dec, src, into, offset);
-    XSRETURN(1);
-
-void
-decode_only_header_with_offset(dec, src, offset, header_into = NULL)
-    srl_decoder_t *dec;
-    SV *src;
-    UV offset;
-    SV *header_into;
-  PPCODE:
-    ST(0)= srl_decode_header_into(aTHX_ dec, src, header_into, offset);
-    XSRETURN(1);
-
-
 void
 decode_sereal(src, opt = NULL, into = NULL)
     SV *src;
@@ -123,6 +424,8 @@ decode_sereal(src, opt = NULL, into = NULL)
   PREINIT:
     srl_decoder_t *dec= NULL;
   PPCODE:
+    if (SvROK(src))
+        croak("We can't decode a reference as Sereal!");
     /* Support no opt at all, undef, hashref */
     if (opt != NULL) {
         SvGETMAGIC(opt);
@@ -165,34 +468,8 @@ decode_sereal_with_header_data(src, opt = NULL, body_into = NULL, header_into =
     RETVAL = newAV();
     sv_2mortal((SV *)RETVAL);
     av_extend(RETVAL, 1);
-    av_store(RETVAL, 0, header_into);
-    av_store(RETVAL, 1, body_into);
-  OUTPUT: RETVAL
-
-IV
-looks_like_sereal(...)
-  PREINIT:
-    SV *data;
-    char *strdata;
-    STRLEN len;
-  CODE:
-    RETVAL = 1;
-    if (items > 2 || items == 0) {
-        croak("Invalid number of parameters to looks_like_sereal: "
-              "Need one data parameter, possibly preceded by an invocant.");
-    }
-    data = ST(items-1); /* 1 or two items, use the last parameter as data */
-    if (!SvOK(data))
-        RETVAL = 0;
-    else {
-        strdata = SvPV(data, len);
-        if (len < SRL_MAGIC_STRLEN+3 /* at least one version/flag byte, one byte for header len, one type byte (smallest payload) */
-            || strnNE(strdata, SRL_MAGIC_STRING, SRL_MAGIC_STRLEN)
-            || strdata[SRL_MAGIC_STRLEN] == (U8)0) /* FIXME this check could be much better using the proto versions and all*/
-        {
-            RETVAL = 0;
-        }
-    }
+    av_store(RETVAL, 0, SvREFCNT_inc(header_into));
+    av_store(RETVAL, 1, SvREFCNT_inc(body_into));
   OUTPUT: RETVAL
 
 UV
@@ -1,4 +1,5 @@
 author_tools/bench.pl
+author_tools/different_sereal_docs.sh
 author_tools/freeze_thaw_timing.pl
 author_tools/hobodecoder.pl
 author_tools/numeric_str_length.c
@@ -13,8 +14,11 @@ inc/Devel/CheckLib.pm
 inc/Sereal/BuildTools.pm
 lib/Sereal/Decoder.pm
 lib/Sereal/Decoder/Constants.pm
+lib/Sereal/Performance.pm
 Makefile.PL
 MANIFEST			This list of files
+miniz.c
+miniz.h
 ppport.h
 ptable.h
 snappy/csnappy.h
@@ -29,21 +33,54 @@ srl_decoder.h
 srl_inline.h
 srl_protocol.h
 t/001_load.t
+t/004_testset.t
 t/010_desperate.t
 t/020_incremental.t
 t/030_looks_like_sereal.t
-t/100_roundtrip.t
-t/101_roundtrip_v1.t
+t/040_special_vars.t
+t/060_each.t
+t/070_alias_options.t
+t/080_set_readonly.t
 t/110_nobless.t
 t/150_dec_exception.t
 t/160_recursion.t
 t/200_bulk.t
+t/210_bulk_readonly.t
 t/300_overload.t
 t/400_utf8validate.t
 t/500_utf8decoding.t
+t/550_decode_into.t
+t/700_roundtrip/v1/plain.t
+t/700_roundtrip/v1/plain_canon.t
+t/700_roundtrip/v1/snappy.t
+t/700_roundtrip/v1/snappy_canon.t
+t/700_roundtrip/v2/dedudep_strings.t
+t/700_roundtrip/v2/freeze_thaw.t
+t/700_roundtrip/v2/plain.t
+t/700_roundtrip/v2/plain_canon.t
+t/700_roundtrip/v2/readonly.t
+t/700_roundtrip/v2/snappy.t
+t/700_roundtrip/v2/snappy_canon.t
+t/700_roundtrip/v2/snappy_incr.t
+t/700_roundtrip/v2/snappy_incr_canon.t
+t/700_roundtrip/v2/sort_keys.t
+t/700_roundtrip/v3/dedudep_strings.t
+t/700_roundtrip/v3/freeze_thaw.t
+t/700_roundtrip/v3/plain.t
+t/700_roundtrip/v3/plain_canon.t
+t/700_roundtrip/v3/readonly.t
+t/700_roundtrip/v3/snappy.t
+t/700_roundtrip/v3/snappy_canon.t
+t/700_roundtrip/v3/snappy_incr.t
+t/700_roundtrip/v3/snappy_incr_canon.t
+t/700_roundtrip/v3/sort_keys.t
+t/700_roundtrip/v3/zlib.t
+t/700_roundtrip/v3/zlib_force.t
 t/800_threads.t
 t/900_regr_issue_15.t
 t/901_regr_segv.t
+t/902_bad_input.t
+t/903_reentrancy.t
 t/data/corpus
 t/lib/Sereal/BulkTest.pm
 t/lib/Sereal/TestSet.pm
@@ -4,7 +4,7 @@
       "Steffen Mueller <smueller@cpan.org>, Yves Orton <yves@cpan.org>"
    ],
    "dynamic_config" : 1,
-   "generated_by" : "ExtUtils::MakeMaker version 6.64, CPAN::Meta::Converter version 2.120921",
+   "generated_by" : "ExtUtils::MakeMaker version 6.9, CPAN::Meta::Converter version 2.141520",
    "license" : [
       "perl_5"
    ],
@@ -55,5 +55,5 @@
          "url" : "git://github.com/Sereal/Sereal.git"
       }
    },
-   "version" : "2.03"
+   "version" : "3.002"
 }
@@ -3,33 +3,33 @@ abstract: 'Fast, compact, powerful binary deserialization'
 author:
   - 'Steffen Mueller <smueller@cpan.org>, Yves Orton <yves@cpan.org>'
 build_requires:
-  Data::Dumper: 0
-  ExtUtils::ParseXS: 2.21
-  File::Find: 0
-  File::Path: 0
-  File::Spec: 0
-  Scalar::Util: 0
-  Test::LongString: 0
-  Test::More: 0.88
-  Test::Warn: 0
-  XSLoader: 0
+  Data::Dumper: '0'
+  ExtUtils::ParseXS: '2.21'
+  File::Find: '0'
+  File::Path: '0'
+  File::Spec: '0'
+  Scalar::Util: '0'
+  Test::LongString: '0'
+  Test::More: '0.88'
+  Test::Warn: '0'
+  XSLoader: '0'
 configure_requires:
-  ExtUtils::MakeMaker: 0
+  ExtUtils::MakeMaker: '0'
 dynamic_config: 1
-generated_by: 'ExtUtils::MakeMaker version 6.64, CPAN::Meta::Converter version 2.120921'
+generated_by: 'ExtUtils::MakeMaker version 6.9, CPAN::Meta::Converter version 2.141520'
 license: perl
 meta-spec:
   url: http://module-build.sourceforge.net/META-spec-v1.4.html
-  version: 1.4
+  version: '1.4'
 name: Sereal-Decoder
 no_index:
   directory:
     - t
     - inc
 requires:
-  XSLoader: 0
-  perl: 5.008
+  XSLoader: '0'
+  perl: '5.008'
 resources:
   bugtracker: https://github.com/Sereal/Sereal/issues
   repository: git://github.com/Sereal/Sereal.git
-version: 2.03
+version: '3.002'
@@ -11,13 +11,20 @@ my $in_source_repo = -d "../../.git" and -d $shared_dir;
 
 my $module = "Sereal::Decoder";
 
-unshift @INC, '.';
+unshift @INC, ".", "./inc";
+unshift @INC, $shared_dir, "$shared_dir/inc" if $in_source_repo;
 require inc::Sereal::BuildTools;
 inc::Sereal::BuildTools::link_files($shared_dir) if $in_source_repo;
 inc::Sereal::BuildTools::generate_constant_includes($module) if $in_source_repo;
 
 our $OPTIMIZE;
 
+# TODO Configure/optimize for miniz:
+#   * Important: For best perf. be sure to customize the below macros for your target platform:
+#     #define MINIZ_USE_UNALIGNED_LOADS_AND_STORES 1
+#     #define MINIZ_LITTLE_ENDIAN 1
+#     #define MINIZ_HAS_64BIT_REGISTERS 1
+
 my $defines = join " ", map "-D$_", grep exists $ENV{$_}, qw(NOINLINE DEBUG MEMDEBUG NDEBUG);
 if ($Config{gccversion}) {
     $OPTIMIZE = '-O3 -Wall -W';
@@ -34,6 +41,17 @@ else {
   $defines .= " -DNDEBUG";
 }
 
+# TODO: This is identical with Encoder.
+if ($Config{osname} eq 'hpux' && not $Config{gccversion}) {
+  # HP-UX cc does not support inline.
+  # Or rather, it does, but it depends on the compiler flags,
+  # assumedly -AC99 instead of -Ae would work.
+  # But we cannot change the compiler config too much from
+  # the one that was used to compile Perl,
+  # so we just fake the inline away.
+  $defines .= " -Dinline= ";
+}
+
 # See lib/ExtUtils/MakeMaker.pm for details of how to influence
 # the contents of the Makefile that is written.
 WriteMakefile1(
@@ -69,6 +87,9 @@ WriteMakefile1(
     INC               => '-I.', # e.g., '-I. -I/usr/include/other'
     OPTIMIZE          => $OPTIMIZE,
     OBJECT            => '$(O_FILES)',
+    test              => {
+        TESTS => "t/*.t t/*/*/*.t"
+    },
 );
 $ENV{OPTIMIZE} = $OPTIMIZE;
 
@@ -65,16 +65,20 @@ push @str, substr($chars, int(rand(int(length($chars)/2+1))), 10) for 1..1000;
 my @rand = map rand, 1..1000;
 our %data;
 
-$data{$_}= make_data() for qw(sereal sereal_func dd1 dd2 ddl mp json_xs storable sereal_snappy cbor);
+$data{$_}= make_data() for qw(sereal sereal_func dd1 dd2 ddl mp json_xs storable sereal_snappy sereal_zlib_fast sereal_zlib_small cbor);
 
 our $enc = Sereal::Encoder->new(\%opt);
-our $enc_snappy = Sereal::Encoder->new({%opt, snappy => 1});
+our $enc_snappy = Sereal::Encoder->new({%opt, compress => Sereal::Encoder::SRL_SNAPPY});
+our $enc_zlib_fast = Sereal::Encoder->new({%opt, compress => Sereal::Encoder::SRL_ZLIB, compress_level => 1, compress_threshold => 0});
+our $enc_zlib_small = Sereal::Encoder->new({%opt, compress => Sereal::Encoder::SRL_ZLIB, compress_level => 10, compress_threshold => 0});
 our $dec = Sereal::Decoder->new(\%opt);
 
-our ($json_xs, $dd1, $dd2, $ddl, $sereal, $storable, $mp, $sereal_snappy, $cbor);
+our ($json_xs, $dd1, $dd2, $ddl, $sereal, $storable, $mp, $sereal_snappy, $sereal_zlib_fast, $sereal_zlib_small, $cbor);
 # do this first before any of the other dumpers "contaminate" the iv/pv issue
-$sereal   = $enc->encode($data{sereal});
-$sereal_snappy   = $enc_snappy->encode($data{sereal_snappy});
+$sereal            = $enc->encode($data{sereal});
+$sereal_snappy     = $enc_snappy->encode($data{sereal_snappy});
+$sereal_zlib_fast  = $enc_zlib_fast->encode($data{sereal_zlib_fast});
+$sereal_zlib_small = $enc_zlib_small->encode($data{sereal_zlib_small});
 if (!SEREAL_ONLY) {
     $json_xs  = encode_json($data{json_xs}) if !$medium_data or $nobless;
     $dd1      = Data::Dumper->new([$data{dd1}])->Indent(0)->Dump();
@@ -102,6 +106,8 @@ if (!SEREAL_ONLY) {
         ["Storable", bytes::length($storable)],
         ["Sereal::Encoder",  bytes::length($sereal)],
         ["Sereal::Encoder, Snappy",  bytes::length($sereal_snappy)],
+        ["Sereal::Encoder, Zlib (fast)",  bytes::length($sereal_zlib_fast)],
+        ["Sereal::Encoder, Zlib (small)",  bytes::length($sereal_zlib_small)],
     );
     for my $tuple (@size_datasets) {
         my ($name, $size) = @$tuple;
@@ -130,6 +136,8 @@ if ($encoder) {
             sereal_func => '$::x = encode_sereal($::data{sereal_func}, \%::opt);',
             sereal => '$::x = $::enc->encode($::data{sereal});',
             sereal_snappy => '$::x = $::enc_snappy->encode($::data{sereal_snappy});',
+            sereal_zlib_fast => '$::x = $::enc_zlib_fast->encode($::data{sereal_zlib_fast});',
+            sereal_zlib_small => '$::x = $::enc_zlib_small->encode($::data{sereal_zlib_small});',
         }
     );
 }
@@ -152,6 +160,8 @@ if ($decoder) {
             sereal_func => '$::x = decode_sereal($::sereal, \%::opt);',
             sereal => '$::x = $::dec->decode($::sereal);',
             sereal_snappy => '$::x = $::dec->decode($::sereal_snappy);',
+            sereal_zlib_fast => '$::x = $::dec->decode($::sereal_zlib_fast);',
+            sereal_zlib_small => '$::x = $::dec->decode($::sereal_zlib_small);',
         }
     );
 }
@@ -0,0 +1,38 @@
+#!/bin/sh -e
+
+# This emits all possible types of Sereal docs, for testing the patch
+# to file(1) to detect the various types
+
+rm -v /tmp/sereal-*.srl
+
+perl -MSereal::Encoder=encode_sereal -we 'print encode_sereal("foo" x 8, {use_protocol_v1 => 1})' >/tmp/sereal-v1-plain.srl
+hexdump -C /tmp/sereal-v1-plain.srl
+sha1sum /tmp/sereal-v1-plain.srl
+echo
+
+perl -MSereal::Encoder=encode_sereal -we 'print encode_sereal("foo" x 8, {})' >/tmp/sereal-v2-plain.srl
+hexdump -C /tmp/sereal-v2-plain.srl
+sha1sum /tmp/sereal-v2-plain.srl
+echo
+
+perl -MSereal::Encoder=encode_sereal -we 'print encode_sereal("foo" x 8, {use_protocol_v1 => 1, snappy => 1, snappy_incr => 0, snappy_threshold => 0})' >/tmp/sereal-v1-snappy-nonincr.srl
+hexdump -C /tmp/sereal-v1-snappy-nonincr.srl
+sha1sum /tmp/sereal-v1-snappy-nonincr.srl
+echo
+
+perl -MSereal::Encoder=encode_sereal -we 'print encode_sereal("foo" x 8, {use_protocol_v1 => 1, snappy => 0, snappy_incr => 1, snappy_threshold => 0})' >/tmp/sereal-v1-snappy-incr.srl
+hexdump -C /tmp/sereal-v1-snappy-incr.srl
+sha1sum /tmp/sereal-v1-snappy-incr.srl
+echo
+
+# The non-incremental option doesn't do anything under v2
+perl -MSereal::Encoder=encode_sereal -we 'print encode_sereal("foo" x 8, {use_protocol_v1 => 0, snappy => 1, snappy_incr => 0, snappy_threshold => 0})' >/tmp/sereal-v2-snappy-nonincr.srl
+hexdump -C /tmp/sereal-v2-snappy-nonincr.srl
+sha1sum /tmp/sereal-v2-snappy-nonincr.srl
+echo
+
+perl -MSereal::Encoder=encode_sereal -we 'print encode_sereal("foo" x 8, {use_protocol_v1 => 0, snappy => 0, snappy_incr => 1, snappy_threshold => 0})' >/tmp/sereal-v2-snappy-incr.srl
+hexdump -C /tmp/sereal-v2-snappy-incr.srl
+sha1sum /tmp/sereal-v2-snappy-incr.srl
+echo
+
@@ -4,13 +4,18 @@ use warnings;
 use Data::Dumper;
 
 use Getopt::Long qw(GetOptions);
+our @constants;
 BEGIN {
     my $err;
     eval '
         use Sereal::Encoder::Constants qw(:all);
+        @constants= @Sereal::Encoder::Constants::EXPORT_OK;
+        print "Loaded constants from $INC{q(Sereal/Encoder/Constants.pm)}\n";
         1;
     ' or do { $err= $@; eval '
         use Sereal::Decoder::Constants qw(:all);
+        @constants= @Sereal::Decoder::Constants::EXPORT_OK;
+        print "Loaded constants from $INC{q(Sereal/Decoder/Constants.pm)}\n";
         1;
     ' } or die "No encoder/decoder constants: $err\n$@";
 }
@@ -19,10 +24,9 @@ my $done;
 my $data;
 my $hlen;
 my $indent = "";
-my %const_names = map {$_ => eval "$_"} @Sereal::Constants::EXPORT_OK;
 
 sub parse_header {
-  $data =~ s/^(=srl)(.)// or die "invalid header: $data";
+  $data =~ s/^(=[s\xF3]rl)(.)// or die "invalid header: $data";
   $done .= $1 . $2;
   my $flags = $2;
   my $len = varint();
@@ -44,9 +48,13 @@ sub parse_header {
   else {
     print "Empty Header.\n";
   }
+
   my $encoding= ord($flags) & SRL_PROTOCOL_ENCODING_MASK;
 
-  if ($encoding == SRL_PROTOCOL_ENCODING_SNAPPY) {
+  printf "%i %i %i\n", $encoding, ord(SRL_PROTOCOL_ENCODING_MASK), ord($flags);
+  if ($encoding == SRL_PROTOCOL_ENCODING_RAW) {
+    print "Header says: Document body is uncompressed.\n";
+  } elsif ($encoding == SRL_PROTOCOL_ENCODING_SNAPPY) {
     print "Header says: Document body is Snappy-compressed.\n";
     require Compress::Snappy;
     my $out = Compress::Snappy::decompress($data);
@@ -57,7 +65,14 @@ sub parse_header {
     require Compress::Snappy;
     my $out = Compress::Snappy::decompress($data);
     $data = $out;
-  } elsif ($encoding) {
+  } elsif ($encoding == SRL_PROTOCOL_ENCODING_ZLIB) {
+    print "Header says: Document body is ZLIB-compressed.\n";
+    my $uncompressed_len = varint();
+    my $compressed_len = varint();
+    require Compress::Zlib;
+    my $out = Compress::Zlib::uncompress($data);
+    $data = $out;
+  } else {
     die "Invalid encoding '" . ($encoding >> SRL_PROTOCOL_VERSION_BITS) . "'";
   }
   $hlen= length($done);
@@ -77,9 +92,9 @@ sub parse_double {
     return unpack("d",$v);
 }
 sub parse_long_double {
-    $len_D||= eval { length(pack("D",0)) };
+    $len_D ||= eval { length(pack("D",0.0)) };
     die "Long double not supported" unless $len_D;
-    my $v= substr($data,0,$len_D,"");
+    my $v= substr($data, 0, $len_D, "");
     $done .= $v;
     return unpack("D",$v);
 }
@@ -102,6 +117,9 @@ sub parse_sv {
   if ($o == SRL_HDR_VARINT) {
     printf "VARINT: %u\n", varint();
   }
+  elsif ($o == SRL_HDR_ZIGZAG) {
+    printf "ZIGZAG: %d\n", zigzag();
+  }
   elsif (SRL_HDR_POS_LOW <= $o && $o <= SRL_HDR_POS_HIGH) {
     printf "POS: %u\n", $o;
   }
@@ -159,6 +177,9 @@ sub parse_sv {
     printf "HASH";
     parse_hv($ind);
   }
+  elsif ($o == SRL_HDR_CANONICAL_UNDEF) {
+    printf "CANONICAL_UNDEF\n";
+  }
   elsif ($o == SRL_HDR_UNDEF) {
     printf "UNDEF\n";
   }
@@ -213,7 +234,8 @@ sub parse_sv {
   }
   else {
     printf "<UNKNOWN>\n";
-    die "unsupported type: $o ($t): $const_names{$o}";
+    die sprintf "unsupported type: 0x%02x (%d) %s: %s", $o, $o,
+        Data::Dumper::qquote($t), Data::Dumper->new([$TAG_INFO_ARRAY[$o]])->Terse(1)->Dump();
   }
   return 0;
 }
@@ -239,7 +261,7 @@ sub parse_hv {
   while ($len--) {
     my $t = substr($data, 0, 1);
     my $o = ord($t);
-    print( "               ", $ind, ($flipflop++ % 2 == 1 ? "VALUE" : "KEY"), ":\n" );
+    printf  "$fmt2%s:\n",("") x $lead_items, $ind, ($flipflop++ %2 == 1 ? "VALUE" : "KEY");
     parse_sv($ind."  ");
   }
 }
@@ -266,6 +288,14 @@ sub varint {
   return $x;
 }
 
+BEGIN{
+my $_shift= length(pack"j",0) * 8 - 1;
+sub zigzag {
+    my $n= varint();
+    return ($n >> 1) ^ (-($n & 1));
+}
+}
+
 GetOptions(
   my $opt = {},
   'e|stderr',
@@ -276,8 +306,6 @@ if ($opt->{e}) {
   select(STDERR);
 }
 
-#print Dumper \%const_names; exit;
-
 local $/ = undef;
 $data = <STDIN>;
 
@@ -3,6 +3,7 @@ use strict;
 use warnings;
 use Data::Dumper;
 my (
+    @meta,
     %name_to_value,             # just the names in the srl_protocol.h
     %name_to_value_expanded,    # names from srl_protocol, but with the LOW/HIGH data expanded
     %value_to_name_expanded,    # values from srl_protocol_expanded, mapping back, note value points at FIRST name
@@ -15,25 +16,48 @@ sub fill_ranges {
     $pfx=~s/_LOW//;
     defined(my $ofs= $name_to_value_expanded{$pfx})
         or die "unknown $pfx";
-    for my $i ( $name_to_value_expanded{$pfx . "_LOW"} .. $name_to_value_expanded{$pfx . "_HIGH"}) {
-        my $n= $pfx=~/NEG/ ? abs($i - 32) : $i - $ofs;
-        $name_to_value_expanded{ $pfx . "_" . $n } ||= $i;
-        $value_to_name_expanded{ $i } = $pfx . "_". $n;
-        $value_to_comment_expanded{ $i } ||= '';
+    for my $value ( $name_to_value_expanded{$pfx . "_LOW"} .. $name_to_value_expanded{$pfx . "_HIGH"}) {
+        my $n= $pfx=~/NEG/ ? abs($value - 32) : $value - $ofs;
+        my $name= $pfx . "_" . $n;
+        $name_to_value_expanded{ $name } ||= $value;
+        $value_to_name_expanded{ $value } = $name;
+        $value_to_comment_expanded{ $value } ||= '';
+
+        $meta[$value]{name}= $name;
+        $meta[$value]{value}= $value;
+        $meta[$value]{type_name}= $pfx;
+        $meta[$value]{type_value}= $ofs;
+        #$meta[$value]{comment}= $value_to_comment_expanded{ $ofs }
+        #    if exists $value_to_comment_expanded{ $ofs };
+
+        $meta[$value]{masked_val}= $n;
+        $meta[$value]{masked}= 1;
+
     }
     $value_to_comment_expanded{ $name_to_value_expanded{$pfx . "_HIGH"} } = $value_to_comment_expanded{ $ofs };
 }
 sub read_protocol {
     open my $fh,"<", "Perl/shared/srl_protocol.h"
         or die "Perl/shared/srl_protocol.h: $!";
+
     my @fill;
     while (<$fh>) {
         if(m!^#define\s+SRL_HDR_(\S+)\s+\(\(char\)(\d+)\)\s*(?:/\*\s*(.*?)\s*\*/)?\s*\z!i) {
-            $name_to_value{$1}= $2;
-            $name_to_value_expanded{$1}= $2;
-            $value_to_name_expanded{$2} ||= $1;
-            $value_to_comment_expanded{$2} ||= $3;
-            push @fill, $1 if substr($1,-4) eq '_LOW';
+            my ($name, $value, $comment)= ($1, $2, $3);
+            $value= 0+$value;
+            $name_to_value{$name}= $value;
+            $name_to_value_expanded{$name}= $value;
+            $value_to_name_expanded{$value} ||= $name;
+            $value_to_comment_expanded{$value} ||= $comment;
+            push @fill, $name if substr($name, -4) eq '_LOW';
+
+            if ( $value < 128 ) {
+                $meta[$value]{name}= $name;
+                $meta[$value]{value}= $value;
+                $meta[$value]{type_name}= $name;
+                $meta[$value]{type_value}= $value;
+                $meta[$value]{comment}= $comment if defined $comment;
+            }
         }
     }
     close $fh;
@@ -42,6 +66,7 @@ sub read_protocol {
         $max_name_length= length($pfx) if $max_name_length < length($pfx);
     }
 }
+
 sub open_swap {
     my $file= shift;
     open my $fh,"<", $file
@@ -58,12 +83,12 @@ sub replace_block {
     my ($in,$out)= open_swap($file);
     while (<$in>) {
         print $out $_;
-        last if /^=for autoupdater start/;
+        last if /^=for autoupdater start/ || /^# start autoupdated section/;
     }
     $blob=~s/\s+$//mg;
     print $out "\n$blob\n\n";
     while (<$in>) {
-        if (/^=for autoupdater stop/) {
+        if (/^=for autoupdater stop/ || /^# stop autoupdated section/) {
             print $out $_;
             last;
         }
@@ -74,6 +99,18 @@ sub replace_block {
     close $out;
     close $in;
 }
+sub update_buildtools {
+    my $dump= Data::Dumper->new([\@meta],['*TAG_INFO_ARRAY'])->Indent(1)->Dump();
+    $dump =~ s/^(\s*)\{/$1# autoupdated by $0 do not modify directly!\n$1\{/mg;
+    return replace_block(
+        "Perl/shared/inc/Sereal/BuildTools.pm",
+        join "\n",
+            "our (%TAG_INFO_HASH, \@TAG_INFO_ARRAY);",
+            $dump,
+            "\$TAG_INFO_HASH{chr \$_}= \$TAG_INFO_ARRAY[\$_] for 0 .. 127;",
+            "push \@EXPORT_OK, qw(%TAG_INFO_HASH \@TAG_INFO_ARRAY);",
+    )
+}
 sub update_srl_decoder_h {
     replace_block("Perl/Decoder/srl_decoder.h",
         join("\n",
@@ -135,6 +172,7 @@ chomp($git_dir);
 chdir "$git_dir/.."
     or die "Failed to chdir to root of repo '$git_dir/..': $!";
 read_protocol();
+update_buildtools();
 update_srl_decoder_h();
 update_table("sereal_spec.pod");
 update_table("Perl/shared/srl_protocol.h");
@@ -538,6 +538,95 @@ constant_20 (pTHX_ const char *name, IV *iv_return) {
 }
 
 static int
+constant_21 (pTHX_ const char *name, IV *iv_return) {
+  /* When generated this function returned values for the list of names given
+     here.  However, subsequent manual editing may have added or removed some.
+     SRL_HDR_ARRAYREF_HIGH SRL_HDR_OBJECT_FREEZE SRL_HDR_RESERVED_HIGH */
+  /* Offset 13 gives the best switch position.  */
+  switch (name[13]) {
+  case 'R':
+    if (memEQ(name, "SRL_HDR_ARRAYREF_HIGH", 21)) {
+    /*                            ^              */
+#ifdef SRL_HDR_ARRAYREF_HIGH
+      *iv_return = SRL_HDR_ARRAYREF_HIGH;
+      return PERL_constant_ISIV;
+#else
+      return PERL_constant_NOTDEF;
+#endif
+    }
+    break;
+  case 'T':
+    if (memEQ(name, "SRL_HDR_OBJECT_FREEZE", 21)) {
+    /*                            ^              */
+#ifdef SRL_HDR_OBJECT_FREEZE
+      *iv_return = SRL_HDR_OBJECT_FREEZE;
+      return PERL_constant_ISIV;
+#else
+      return PERL_constant_NOTDEF;
+#endif
+    }
+    break;
+  case 'V':
+    if (memEQ(name, "SRL_HDR_RESERVED_HIGH", 21)) {
+    /*                            ^              */
+#ifdef SRL_HDR_RESERVED_HIGH
+      *iv_return = SRL_HDR_RESERVED_HIGH;
+      return PERL_constant_ISIV;
+#else
+      return PERL_constant_NOTDEF;
+#endif
+    }
+    break;
+  }
+  return PERL_constant_NOTFOUND;
+}
+
+static int
+constant_24 (pTHX_ const char *name, IV *iv_return) {
+  /* When generated this function returned values for the list of names given
+     here.  However, subsequent manual editing may have added or removed some.
+     SRL_HDR_SHORT_BINARY_LOW SRL_MAGIC_STRING_UINT_BE SRL_MAGIC_STRING_UINT_LE
+     */
+  /* Offset 22 gives the best switch position.  */
+  switch (name[22]) {
+  case 'B':
+    if (memEQ(name, "SRL_MAGIC_STRING_UINT_BE", 24)) {
+    /*                                     ^        */
+#ifdef SRL_MAGIC_STRING_UINT_BE
+      *iv_return = SRL_MAGIC_STRING_UINT_BE;
+      return PERL_constant_ISIV;
+#else
+      return PERL_constant_NOTDEF;
+#endif
+    }
+    break;
+  case 'L':
+    if (memEQ(name, "SRL_MAGIC_STRING_UINT_LE", 24)) {
+    /*                                     ^        */
+#ifdef SRL_MAGIC_STRING_UINT_LE
+      *iv_return = SRL_MAGIC_STRING_UINT_LE;
+      return PERL_constant_ISIV;
+#else
+      return PERL_constant_NOTDEF;
+#endif
+    }
+    break;
+  case 'O':
+    if (memEQ(name, "SRL_HDR_SHORT_BINARY_LOW", 24)) {
+    /*                                     ^        */
+#ifdef SRL_HDR_SHORT_BINARY_LOW
+      *iv_return = SRL_HDR_SHORT_BINARY_LOW;
+      return PERL_constant_ISIV;
+#else
+      return PERL_constant_NOTDEF;
+#endif
+    }
+    break;
+  }
+  return PERL_constant_NOTFOUND;
+}
+
+static int
 constant_25 (pTHX_ const char *name, IV *iv_return) {
   /* When generated this function returned values for the list of names given
      here.  However, subsequent manual editing may have added or removed some.
@@ -617,6 +706,51 @@ constant_25 (pTHX_ const char *name, IV *iv_return) {
 }
 
 static int
+constant_26 (pTHX_ const char *name, IV *iv_return) {
+  /* When generated this function returned values for the list of names given
+     here.  However, subsequent manual editing may have added or removed some.
+     SRL_PROTOCOL_ENCODING_MASK SRL_PROTOCOL_ENCODING_ZLIB
+     SRL_PROTOCOL_HDR_USER_DATA */
+  /* Offset 25 gives the best switch position.  */
+  switch (name[25]) {
+  case 'A':
+    if (memEQ(name, "SRL_PROTOCOL_HDR_USER_DAT", 25)) {
+    /*                                        A      */
+#ifdef SRL_PROTOCOL_HDR_USER_DATA
+      *iv_return = SRL_PROTOCOL_HDR_USER_DATA;
+      return PERL_constant_ISIV;
+#else
+      return PERL_constant_NOTDEF;
+#endif
+    }
+    break;
+  case 'B':
+    if (memEQ(name, "SRL_PROTOCOL_ENCODING_ZLI", 25)) {
+    /*                                        B      */
+#ifdef SRL_PROTOCOL_ENCODING_ZLIB
+      *iv_return = SRL_PROTOCOL_ENCODING_ZLIB;
+      return PERL_constant_ISIV;
+#else
+      return PERL_constant_NOTDEF;
+#endif
+    }
+    break;
+  case 'K':
+    if (memEQ(name, "SRL_PROTOCOL_ENCODING_MAS", 25)) {
+    /*                                        K      */
+#ifdef SRL_PROTOCOL_ENCODING_MASK
+      *iv_return = SRL_PROTOCOL_ENCODING_MASK;
+      return PERL_constant_ISIV;
+#else
+      return PERL_constant_NOTDEF;
+#endif
+    }
+    break;
+  }
+  return PERL_constant_NOTFOUND;
+}
+
+static int
 constant (pTHX_ const char *name, STRLEN len, IV *iv_return) {
   /* Initially switch on the length of the name.  */
   /* When generated this function returned values for the list of names given
@@ -635,26 +769,31 @@ use ExtUtils::Constant qw (constant_types C_constant XS_constant);
 my $types = {map {($_, 1)} qw(IV)};
 my @names = (qw(SRL_HDR_ALIAS SRL_HDR_ARRAY SRL_HDR_ARRAYREF
 	       SRL_HDR_ARRAYREF_HIGH SRL_HDR_ARRAYREF_LOW SRL_HDR_BINARY
-	       SRL_HDR_COPY SRL_HDR_DOUBLE SRL_HDR_EXTEND SRL_HDR_FALSE
-	       SRL_HDR_FLOAT SRL_HDR_HASH SRL_HDR_HASHREF SRL_HDR_HASHREF_HIGH
-	       SRL_HDR_HASHREF_LOW SRL_HDR_LONG_DOUBLE SRL_HDR_MANY SRL_HDR_NEG
-	       SRL_HDR_NEG_HIGH SRL_HDR_NEG_LOW SRL_HDR_OBJECT SRL_HDR_OBJECTV
-	       SRL_HDR_OBJECTV_FREEZE SRL_HDR_PACKET_START SRL_HDR_PAD
-	       SRL_HDR_POS SRL_HDR_POS_HIGH SRL_HDR_POS_LOW SRL_HDR_REFN
-	       SRL_HDR_REFP SRL_HDR_REGEXP SRL_HDR_RESERVED
-	       SRL_HDR_RESERVED_HIGH SRL_HDR_RESERVED_LOW SRL_HDR_SHORT_BINARY
-	       SRL_HDR_SHORT_BINARY_HIGH SRL_HDR_SHORT_BINARY_LOW
-	       SRL_HDR_STR_UTF8 SRL_HDR_TRACK_FLAG SRL_HDR_TRUE SRL_HDR_UNDEF
-	       SRL_HDR_VARINT SRL_HDR_WEAKEN SRL_HDR_ZIGZAG
-	       SRL_MAGIC_STRING_LILIPUTIAN SRL_MAGIC_STRLEN
+	       SRL_HDR_CANONICAL_UNDEF SRL_HDR_COPY SRL_HDR_DOUBLE
+	       SRL_HDR_EXTEND SRL_HDR_FALSE SRL_HDR_FLOAT SRL_HDR_HASH
+	       SRL_HDR_HASHREF SRL_HDR_HASHREF_HIGH SRL_HDR_HASHREF_LOW
+	       SRL_HDR_LONG_DOUBLE SRL_HDR_MANY SRL_HDR_NEG SRL_HDR_NEG_HIGH
+	       SRL_HDR_NEG_LOW SRL_HDR_OBJECT SRL_HDR_OBJECTV
+	       SRL_HDR_OBJECTV_FREEZE SRL_HDR_OBJECT_FREEZE
+	       SRL_HDR_PACKET_START SRL_HDR_PAD SRL_HDR_POS SRL_HDR_POS_HIGH
+	       SRL_HDR_POS_LOW SRL_HDR_REFN SRL_HDR_REFP SRL_HDR_REGEXP
+	       SRL_HDR_RESERVED SRL_HDR_RESERVED_HIGH SRL_HDR_RESERVED_LOW
+	       SRL_HDR_SHORT_BINARY SRL_HDR_SHORT_BINARY_HIGH
+	       SRL_HDR_SHORT_BINARY_LOW SRL_HDR_STR_UTF8 SRL_HDR_TRACK_FLAG
+	       SRL_HDR_TRUE SRL_HDR_UNDEF SRL_HDR_VARINT SRL_HDR_WEAKEN
+	       SRL_HDR_ZIGZAG SRL_MAGIC_STRING_HIGHBIT_UINT_BE
+	       SRL_MAGIC_STRING_HIGHBIT_UINT_LE
+	       SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_BE
+	       SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_LE SRL_MAGIC_STRING_UINT_BE
+	       SRL_MAGIC_STRING_UINT_LE SRL_MAGIC_STRLEN
 	       SRL_MASK_ARRAYREF_COUNT SRL_MASK_HASHREF_COUNT
 	       SRL_MASK_SHORT_BINARY_LEN SRL_NEG_MIN_SIZE SRL_POS_MAX_SIZE
 	       SRL_PROTOCOL_ENCODING_MASK SRL_PROTOCOL_ENCODING_RAW
 	       SRL_PROTOCOL_ENCODING_SNAPPY
 	       SRL_PROTOCOL_ENCODING_SNAPPY_INCREMENTAL
-	       SRL_PROTOCOL_HDR_CONTINUE SRL_PROTOCOL_HDR_USER_DATA
-	       SRL_PROTOCOL_VERSION SRL_PROTOCOL_VERSION_BITS
-	       SRL_PROTOCOL_VERSION_MASK));
+	       SRL_PROTOCOL_ENCODING_ZLIB SRL_PROTOCOL_HDR_CONTINUE
+	       SRL_PROTOCOL_HDR_USER_DATA SRL_PROTOCOL_VERSION
+	       SRL_PROTOCOL_VERSION_BITS SRL_PROTOCOL_VERSION_MASK));
 
 print constant_types(), "\n"; # macro defs
 foreach (C_constant ("Sereal::Decoder::Constants", 'constant', 'IV', $types, undef, 3, @names) ) {
@@ -727,33 +866,7 @@ __END__
     return constant_20 (aTHX_ name, iv_return);
     break;
   case 21:
-    /* Names all of length 21.  */
-    /* SRL_HDR_ARRAYREF_HIGH SRL_HDR_RESERVED_HIGH */
-    /* Offset 10 gives the best switch position.  */
-    switch (name[10]) {
-    case 'R':
-      if (memEQ(name, "SRL_HDR_ARRAYREF_HIGH", 21)) {
-      /*                         ^                 */
-#ifdef SRL_HDR_ARRAYREF_HIGH
-        *iv_return = SRL_HDR_ARRAYREF_HIGH;
-        return PERL_constant_ISIV;
-#else
-        return PERL_constant_NOTDEF;
-#endif
-      }
-      break;
-    case 'S':
-      if (memEQ(name, "SRL_HDR_RESERVED_HIGH", 21)) {
-      /*                         ^                 */
-#ifdef SRL_HDR_RESERVED_HIGH
-        *iv_return = SRL_HDR_RESERVED_HIGH;
-        return PERL_constant_ISIV;
-#else
-        return PERL_constant_NOTDEF;
-#endif
-      }
-      break;
-    }
+    return constant_21 (aTHX_ name, iv_return);
     break;
   case 22:
     /* Names all of length 22.  */
@@ -785,49 +898,74 @@ __END__
     }
     break;
   case 23:
-    if (memEQ(name, "SRL_MASK_ARRAYREF_COUNT", 23)) {
+    /* Names all of length 23.  */
+    /* SRL_HDR_CANONICAL_UNDEF SRL_MASK_ARRAYREF_COUNT */
+    /* Offset 6 gives the best switch position.  */
+    switch (name[6]) {
+    case 'R':
+      if (memEQ(name, "SRL_HDR_CANONICAL_UNDEF", 23)) {
+      /*                     ^                       */
+#ifdef SRL_HDR_CANONICAL_UNDEF
+        *iv_return = SRL_HDR_CANONICAL_UNDEF;
+        return PERL_constant_ISIV;
+#else
+        return PERL_constant_NOTDEF;
+#endif
+      }
+      break;
+    case 'S':
+      if (memEQ(name, "SRL_MASK_ARRAYREF_COUNT", 23)) {
+      /*                     ^                       */
 #ifdef SRL_MASK_ARRAYREF_COUNT
-      *iv_return = SRL_MASK_ARRAYREF_COUNT;
-      return PERL_constant_ISIV;
+        *iv_return = SRL_MASK_ARRAYREF_COUNT;
+        return PERL_constant_ISIV;
 #else
-      return PERL_constant_NOTDEF;
+        return PERL_constant_NOTDEF;
 #endif
+      }
+      break;
     }
     break;
   case 24:
-    if (memEQ(name, "SRL_HDR_SHORT_BINARY_LOW", 24)) {
-#ifdef SRL_HDR_SHORT_BINARY_LOW
-      *iv_return = SRL_HDR_SHORT_BINARY_LOW;
+    return constant_24 (aTHX_ name, iv_return);
+    break;
+  case 25:
+    return constant_25 (aTHX_ name, iv_return);
+    break;
+  case 26:
+    return constant_26 (aTHX_ name, iv_return);
+    break;
+  case 28:
+    if (memEQ(name, "SRL_PROTOCOL_ENCODING_SNAPPY", 28)) {
+#ifdef SRL_PROTOCOL_ENCODING_SNAPPY
+      *iv_return = SRL_PROTOCOL_ENCODING_SNAPPY;
       return PERL_constant_ISIV;
 #else
       return PERL_constant_NOTDEF;
 #endif
     }
     break;
-  case 25:
-    return constant_25 (aTHX_ name, iv_return);
-    break;
-  case 26:
-    /* Names all of length 26.  */
-    /* SRL_PROTOCOL_ENCODING_MASK SRL_PROTOCOL_HDR_USER_DATA */
-    /* Offset 24 gives the best switch position.  */
-    switch (name[24]) {
-    case 'S':
-      if (memEQ(name, "SRL_PROTOCOL_ENCODING_MASK", 26)) {
-      /*                                       ^        */
-#ifdef SRL_PROTOCOL_ENCODING_MASK
-        *iv_return = SRL_PROTOCOL_ENCODING_MASK;
+  case 32:
+    /* Names all of length 32.  */
+    /* SRL_MAGIC_STRING_HIGHBIT_UINT_BE SRL_MAGIC_STRING_HIGHBIT_UINT_LE */
+    /* Offset 30 gives the best switch position.  */
+    switch (name[30]) {
+    case 'B':
+      if (memEQ(name, "SRL_MAGIC_STRING_HIGHBIT_UINT_BE", 32)) {
+      /*                                             ^        */
+#ifdef SRL_MAGIC_STRING_HIGHBIT_UINT_BE
+        *iv_return = SRL_MAGIC_STRING_HIGHBIT_UINT_BE;
         return PERL_constant_ISIV;
 #else
         return PERL_constant_NOTDEF;
 #endif
       }
       break;
-    case 'T':
-      if (memEQ(name, "SRL_PROTOCOL_HDR_USER_DATA", 26)) {
-      /*                                       ^        */
-#ifdef SRL_PROTOCOL_HDR_USER_DATA
-        *iv_return = SRL_PROTOCOL_HDR_USER_DATA;
+    case 'L':
+      if (memEQ(name, "SRL_MAGIC_STRING_HIGHBIT_UINT_LE", 32)) {
+      /*                                             ^        */
+#ifdef SRL_MAGIC_STRING_HIGHBIT_UINT_LE
+        *iv_return = SRL_MAGIC_STRING_HIGHBIT_UINT_LE;
         return PERL_constant_ISIV;
 #else
         return PERL_constant_NOTDEF;
@@ -836,24 +974,34 @@ __END__
       break;
     }
     break;
-  case 27:
-    if (memEQ(name, "SRL_MAGIC_STRING_LILIPUTIAN", 27)) {
-#ifdef SRL_MAGIC_STRING_LILIPUTIAN
-      *iv_return = SRL_MAGIC_STRING_LILIPUTIAN;
-      return PERL_constant_ISIV;
+  case 37:
+    /* Names all of length 37.  */
+    /* SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_BE
+       SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_LE */
+    /* Offset 35 gives the best switch position.  */
+    switch (name[35]) {
+    case 'B':
+      if (memEQ(name, "SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_BE", 37)) {
+      /*                                                  ^        */
+#ifdef SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_BE
+        *iv_return = SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_BE;
+        return PERL_constant_ISIV;
 #else
-      return PERL_constant_NOTDEF;
+        return PERL_constant_NOTDEF;
 #endif
-    }
-    break;
-  case 28:
-    if (memEQ(name, "SRL_PROTOCOL_ENCODING_SNAPPY", 28)) {
-#ifdef SRL_PROTOCOL_ENCODING_SNAPPY
-      *iv_return = SRL_PROTOCOL_ENCODING_SNAPPY;
-      return PERL_constant_ISIV;
+      }
+      break;
+    case 'L':
+      if (memEQ(name, "SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_LE", 37)) {
+      /*                                                  ^        */
+#ifdef SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_LE
+        *iv_return = SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_LE;
+        return PERL_constant_ISIV;
 #else
-      return PERL_constant_NOTDEF;
+        return PERL_constant_NOTDEF;
 #endif
+      }
+      break;
     }
     break;
   case 40:
@@ -61,7 +61,7 @@ sub generate_constant_includes {
     open my $fh, "<", "srl_protocol.h" or die $!;
     my (@string_const, @int_const);
     while (<$fh>) {
-      if (/^#\s*define\s*(SRL_\w+)\s*(.*)$/) {
+      if (/^#\s*define\s*(SRL_\w+)\s*(.*?)(?:\/\*|$)/) {
         my ($name, $value) = ($1, $2);
         next if $name =~ /_H_$/ or $name =~ /SET/ or $value =~ /"/;
         push @int_const, $name;
@@ -101,8 +101,1160 @@ HERE
   };
 }
 
-sub SRL_MAGIC_STRING () {"=srl"}
-push @EXPORT_OK, "SRL_MAGIC_STRING";
+sub SRL_MAGIC_STRING ()                 { "=srl" }
+sub SRL_MAGIC_STRING_HIGHBIT ()         { "=\xF3rl" }
+sub SRL_MAGIC_STRING_HIGHBIT_UTF8 ()    { "=\xC3\xB3rl" }
+
+push @EXPORT_OK, qw(
+    SRL_MAGIC_STRING
+    SRL_MAGIC_STRING_HIGHBIT
+    SRL_MAGIC_STRING_HIGHBIT_UTF8
+);
+
+# start autoupdated section - do not modify directly
+
+our (%TAG_INFO_HASH, @TAG_INFO_ARRAY);
+@TAG_INFO_ARRAY = (
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'comment' => 'small positive integer - value in low 4 bits (identity)',
+    'value' => 0,
+    'name' => 'POS_0',
+    'masked_val' => 0,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 1,
+    'name' => 'POS_1',
+    'masked_val' => 1,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 2,
+    'name' => 'POS_2',
+    'masked_val' => 2,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 3,
+    'name' => 'POS_3',
+    'masked_val' => 3,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 4,
+    'name' => 'POS_4',
+    'masked_val' => 4,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 5,
+    'name' => 'POS_5',
+    'masked_val' => 5,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 6,
+    'name' => 'POS_6',
+    'masked_val' => 6,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 7,
+    'name' => 'POS_7',
+    'masked_val' => 7,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 8,
+    'name' => 'POS_8',
+    'masked_val' => 8,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 9,
+    'name' => 'POS_9',
+    'masked_val' => 9,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 10,
+    'name' => 'POS_10',
+    'masked_val' => 10,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 11,
+    'name' => 'POS_11',
+    'masked_val' => 11,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 12,
+    'name' => 'POS_12',
+    'masked_val' => 12,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 13,
+    'name' => 'POS_13',
+    'masked_val' => 13,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 14,
+    'name' => 'POS_14',
+    'masked_val' => 14,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'comment' => 'small positive integer - value in low 4 bits (identity)',
+    'value' => 15,
+    'name' => 'POS_15',
+    'masked_val' => 15,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'comment' => 'small negative integer - value in low 4 bits (k+32)',
+    'value' => 16,
+    'name' => 'NEG_16',
+    'masked_val' => 16,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 17,
+    'name' => 'NEG_15',
+    'masked_val' => 15,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 18,
+    'name' => 'NEG_14',
+    'masked_val' => 14,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 19,
+    'name' => 'NEG_13',
+    'masked_val' => 13,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 20,
+    'name' => 'NEG_12',
+    'masked_val' => 12,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 21,
+    'name' => 'NEG_11',
+    'masked_val' => 11,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 22,
+    'name' => 'NEG_10',
+    'masked_val' => 10,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 23,
+    'name' => 'NEG_9',
+    'masked_val' => 9,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 24,
+    'name' => 'NEG_8',
+    'masked_val' => 8,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 25,
+    'name' => 'NEG_7',
+    'masked_val' => 7,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 26,
+    'name' => 'NEG_6',
+    'masked_val' => 6,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 27,
+    'name' => 'NEG_5',
+    'masked_val' => 5,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 28,
+    'name' => 'NEG_4',
+    'masked_val' => 4,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 29,
+    'name' => 'NEG_3',
+    'masked_val' => 3,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 30,
+    'name' => 'NEG_2',
+    'masked_val' => 2,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'comment' => 'small negative integer - value in low 4 bits (k+32)',
+    'value' => 31,
+    'name' => 'NEG_1',
+    'masked_val' => 1,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'VARINT',
+    'comment' => '<VARINT> - Varint variable length integer',
+    'value' => 32,
+    'name' => 'VARINT',
+    'type_value' => 32
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ZIGZAG',
+    'comment' => '<ZIGZAG-VARINT> - Zigzag variable length integer',
+    'value' => 33,
+    'name' => 'ZIGZAG',
+    'type_value' => 33
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'FLOAT',
+    'comment' => '<IEEE-FLOAT>',
+    'value' => 34,
+    'name' => 'FLOAT',
+    'type_value' => 34
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'DOUBLE',
+    'comment' => '<IEEE-DOUBLE>',
+    'value' => 35,
+    'name' => 'DOUBLE',
+    'type_value' => 35
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'LONG_DOUBLE',
+    'comment' => '<IEEE-LONG-DOUBLE>',
+    'value' => 36,
+    'name' => 'LONG_DOUBLE',
+    'type_value' => 36
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'UNDEF',
+    'comment' => 'None - Perl undef var; eg my $var= undef;',
+    'value' => 37,
+    'name' => 'UNDEF',
+    'type_value' => 37
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'BINARY',
+    'comment' => '<LEN-VARINT> <BYTES> - binary/(latin1) string',
+    'value' => 38,
+    'name' => 'BINARY',
+    'type_value' => 38
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'STR_UTF8',
+    'comment' => '<LEN-VARINT> <UTF8> - utf8 string',
+    'value' => 39,
+    'name' => 'STR_UTF8',
+    'type_value' => 39
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'REFN',
+    'comment' => '<ITEM-TAG>    - ref to next item',
+    'value' => 40,
+    'name' => 'REFN',
+    'type_value' => 40
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'REFP',
+    'comment' => '<OFFSET-VARINT> - ref to previous item stored at offset',
+    'value' => 41,
+    'name' => 'REFP',
+    'type_value' => 41
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASH',
+    'comment' => '<COUNT-VARINT> [<KEY-TAG> <ITEM-TAG> ...] - count followed by key/value pairs',
+    'value' => 42,
+    'name' => 'HASH',
+    'type_value' => 42
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAY',
+    'comment' => '<COUNT-VARINT> [<ITEM-TAG> ...] - count followed by items',
+    'value' => 43,
+    'name' => 'ARRAY',
+    'type_value' => 43
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'OBJECT',
+    'comment' => '<STR-TAG> <ITEM-TAG> - class, object-item',
+    'value' => 44,
+    'name' => 'OBJECT',
+    'type_value' => 44
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'OBJECTV',
+    'comment' => '<OFFSET-VARINT> <ITEM-TAG> - offset of previously used classname tag - object-item',
+    'value' => 45,
+    'name' => 'OBJECTV',
+    'type_value' => 45
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ALIAS',
+    'comment' => '<OFFSET-VARINT> - alias to item defined at offset',
+    'value' => 46,
+    'name' => 'ALIAS',
+    'type_value' => 46
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'COPY',
+    'comment' => '<OFFSET-VARINT> - copy of item defined at offset',
+    'value' => 47,
+    'name' => 'COPY',
+    'type_value' => 47
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'WEAKEN',
+    'comment' => '<REF-TAG> - Weaken the following reference',
+    'value' => 48,
+    'name' => 'WEAKEN',
+    'type_value' => 48
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'REGEXP',
+    'comment' => '<PATTERN-STR-TAG> <MODIFIERS-STR-TAG>',
+    'value' => 49,
+    'name' => 'REGEXP',
+    'type_value' => 49
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'OBJECT_FREEZE',
+    'comment' => '<STR-TAG> <ITEM-TAG> - class, object-item. Need to call "THAW" method on class after decoding',
+    'value' => 50,
+    'name' => 'OBJECT_FREEZE',
+    'type_value' => 50
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'OBJECTV_FREEZE',
+    'comment' => '<OFFSET-VARINT> <ITEM-TAG> - (OBJECTV_FREEZE is to OBJECT_FREEZE as OBJECTV is to OBJECT)',
+    'value' => 51,
+    'name' => 'OBJECTV_FREEZE',
+    'type_value' => 51
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'RESERVED',
+    'masked' => 1,
+    'comment' => 'reserved',
+    'value' => 52,
+    'name' => 'RESERVED_0',
+    'masked_val' => 0,
+    'type_value' => 52
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'RESERVED',
+    'masked' => 1,
+    'value' => 53,
+    'name' => 'RESERVED_1',
+    'masked_val' => 1,
+    'type_value' => 52
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'RESERVED',
+    'masked' => 1,
+    'value' => 54,
+    'name' => 'RESERVED_2',
+    'masked_val' => 2,
+    'type_value' => 52
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'RESERVED',
+    'masked' => 1,
+    'value' => 55,
+    'name' => 'RESERVED_3',
+    'masked_val' => 3,
+    'type_value' => 52
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'RESERVED',
+    'masked' => 1,
+    'value' => 56,
+    'name' => 'RESERVED_4',
+    'masked_val' => 4,
+    'type_value' => 52
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'CANONICAL_UNDEF',
+    'comment' => 'undef (PL_sv_undef) - "the" Perl undef (see notes)',
+    'value' => 57,
+    'name' => 'CANONICAL_UNDEF',
+    'type_value' => 57
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'FALSE',
+    'comment' => 'false (PL_sv_no)',
+    'value' => 58,
+    'name' => 'FALSE',
+    'type_value' => 58
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'TRUE',
+    'comment' => 'true  (PL_sv_yes)',
+    'value' => 59,
+    'name' => 'TRUE',
+    'type_value' => 59
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'MANY',
+    'comment' => '<LEN-VARINT> <TYPE-BYTE> <TAG-DATA> - repeated tag (not done yet, will be implemented in version 3)',
+    'value' => 60,
+    'name' => 'MANY',
+    'type_value' => 60
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'PACKET_START',
+    'comment' => '(first byte of magic string in header)',
+    'value' => 61,
+    'name' => 'PACKET_START',
+    'type_value' => 61
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'EXTEND',
+    'comment' => '<BYTE> - for additional tags',
+    'value' => 62,
+    'name' => 'EXTEND',
+    'type_value' => 62
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'PAD',
+    'comment' => '(ignored tag, skip to next byte)',
+    'value' => 63,
+    'name' => 'PAD',
+    'type_value' => 63
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'comment' => '[<ITEM-TAG> ...] - count of items in low 4 bits (ARRAY must be refcnt=1)',
+    'value' => 64,
+    'name' => 'ARRAYREF_0',
+    'masked_val' => 0,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 65,
+    'name' => 'ARRAYREF_1',
+    'masked_val' => 1,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 66,
+    'name' => 'ARRAYREF_2',
+    'masked_val' => 2,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 67,
+    'name' => 'ARRAYREF_3',
+    'masked_val' => 3,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 68,
+    'name' => 'ARRAYREF_4',
+    'masked_val' => 4,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 69,
+    'name' => 'ARRAYREF_5',
+    'masked_val' => 5,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 70,
+    'name' => 'ARRAYREF_6',
+    'masked_val' => 6,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 71,
+    'name' => 'ARRAYREF_7',
+    'masked_val' => 7,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 72,
+    'name' => 'ARRAYREF_8',
+    'masked_val' => 8,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 73,
+    'name' => 'ARRAYREF_9',
+    'masked_val' => 9,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 74,
+    'name' => 'ARRAYREF_10',
+    'masked_val' => 10,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 75,
+    'name' => 'ARRAYREF_11',
+    'masked_val' => 11,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 76,
+    'name' => 'ARRAYREF_12',
+    'masked_val' => 12,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 77,
+    'name' => 'ARRAYREF_13',
+    'masked_val' => 13,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 78,
+    'name' => 'ARRAYREF_14',
+    'masked_val' => 14,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 79,
+    'name' => 'ARRAYREF_15',
+    'masked_val' => 15,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'comment' => '[<KEY-TAG> <ITEM-TAG> ...] - count in low 4 bits, key/value pairs (HASH must be refcnt=1)',
+    'value' => 80,
+    'name' => 'HASHREF_0',
+    'masked_val' => 0,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 81,
+    'name' => 'HASHREF_1',
+    'masked_val' => 1,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 82,
+    'name' => 'HASHREF_2',
+    'masked_val' => 2,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 83,
+    'name' => 'HASHREF_3',
+    'masked_val' => 3,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 84,
+    'name' => 'HASHREF_4',
+    'masked_val' => 4,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 85,
+    'name' => 'HASHREF_5',
+    'masked_val' => 5,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 86,
+    'name' => 'HASHREF_6',
+    'masked_val' => 6,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 87,
+    'name' => 'HASHREF_7',
+    'masked_val' => 7,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 88,
+    'name' => 'HASHREF_8',
+    'masked_val' => 8,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 89,
+    'name' => 'HASHREF_9',
+    'masked_val' => 9,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 90,
+    'name' => 'HASHREF_10',
+    'masked_val' => 10,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 91,
+    'name' => 'HASHREF_11',
+    'masked_val' => 11,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 92,
+    'name' => 'HASHREF_12',
+    'masked_val' => 12,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 93,
+    'name' => 'HASHREF_13',
+    'masked_val' => 13,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 94,
+    'name' => 'HASHREF_14',
+    'masked_val' => 14,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 95,
+    'name' => 'HASHREF_15',
+    'masked_val' => 15,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'comment' => '<BYTES> - binary/latin1 string, length encoded in low 5 bits of tag',
+    'value' => 96,
+    'name' => 'SHORT_BINARY_0',
+    'masked_val' => 0,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 97,
+    'name' => 'SHORT_BINARY_1',
+    'masked_val' => 1,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 98,
+    'name' => 'SHORT_BINARY_2',
+    'masked_val' => 2,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 99,
+    'name' => 'SHORT_BINARY_3',
+    'masked_val' => 3,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 100,
+    'name' => 'SHORT_BINARY_4',
+    'masked_val' => 4,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 101,
+    'name' => 'SHORT_BINARY_5',
+    'masked_val' => 5,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 102,
+    'name' => 'SHORT_BINARY_6',
+    'masked_val' => 6,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 103,
+    'name' => 'SHORT_BINARY_7',
+    'masked_val' => 7,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 104,
+    'name' => 'SHORT_BINARY_8',
+    'masked_val' => 8,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 105,
+    'name' => 'SHORT_BINARY_9',
+    'masked_val' => 9,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 106,
+    'name' => 'SHORT_BINARY_10',
+    'masked_val' => 10,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 107,
+    'name' => 'SHORT_BINARY_11',
+    'masked_val' => 11,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 108,
+    'name' => 'SHORT_BINARY_12',
+    'masked_val' => 12,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 109,
+    'name' => 'SHORT_BINARY_13',
+    'masked_val' => 13,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 110,
+    'name' => 'SHORT_BINARY_14',
+    'masked_val' => 14,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 111,
+    'name' => 'SHORT_BINARY_15',
+    'masked_val' => 15,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 112,
+    'name' => 'SHORT_BINARY_16',
+    'masked_val' => 16,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 113,
+    'name' => 'SHORT_BINARY_17',
+    'masked_val' => 17,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 114,
+    'name' => 'SHORT_BINARY_18',
+    'masked_val' => 18,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 115,
+    'name' => 'SHORT_BINARY_19',
+    'masked_val' => 19,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 116,
+    'name' => 'SHORT_BINARY_20',
+    'masked_val' => 20,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 117,
+    'name' => 'SHORT_BINARY_21',
+    'masked_val' => 21,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 118,
+    'name' => 'SHORT_BINARY_22',
+    'masked_val' => 22,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 119,
+    'name' => 'SHORT_BINARY_23',
+    'masked_val' => 23,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 120,
+    'name' => 'SHORT_BINARY_24',
+    'masked_val' => 24,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 121,
+    'name' => 'SHORT_BINARY_25',
+    'masked_val' => 25,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 122,
+    'name' => 'SHORT_BINARY_26',
+    'masked_val' => 26,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 123,
+    'name' => 'SHORT_BINARY_27',
+    'masked_val' => 27,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 124,
+    'name' => 'SHORT_BINARY_28',
+    'masked_val' => 28,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 125,
+    'name' => 'SHORT_BINARY_29',
+    'masked_val' => 29,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 126,
+    'name' => 'SHORT_BINARY_30',
+    'masked_val' => 30,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 127,
+    'name' => 'SHORT_BINARY_31',
+    'masked_val' => 31,
+    'type_value' => 96
+  }
+);
+$TAG_INFO_HASH{chr $_}= $TAG_INFO_ARRAY[$_] for 0 .. 127;
+push @EXPORT_OK, qw(%TAG_INFO_HASH @TAG_INFO_ARRAY);
+
+# stop autoupdated section - do not modify directly!
+
+
 our %EXPORT_TAGS=(all => \@EXPORT_OK);
 HERE
     close $ofh;
@@ -10,7 +10,12 @@ require Exporter;
 our @EXPORT_OK;
 BEGIN { @EXPORT_OK = qw(
     SRL_MAGIC_STRLEN
-    SRL_MAGIC_STRING_LILIPUTIAN
+    SRL_MAGIC_STRING_UINT_LE
+    SRL_MAGIC_STRING_UINT_BE
+    SRL_MAGIC_STRING_HIGHBIT_UINT_LE
+    SRL_MAGIC_STRING_HIGHBIT_UINT_BE
+    SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_LE
+    SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_BE
     SRL_PROTOCOL_VERSION
     SRL_PROTOCOL_VERSION_BITS
     SRL_PROTOCOL_VERSION_MASK
@@ -18,6 +23,7 @@ BEGIN { @EXPORT_OK = qw(
     SRL_PROTOCOL_ENCODING_RAW
     SRL_PROTOCOL_ENCODING_SNAPPY
     SRL_PROTOCOL_ENCODING_SNAPPY_INCREMENTAL
+    SRL_PROTOCOL_ENCODING_ZLIB
     SRL_PROTOCOL_HDR_USER_DATA
     SRL_PROTOCOL_HDR_CONTINUE
     SRL_POS_MAX_SIZE
@@ -46,10 +52,12 @@ BEGIN { @EXPORT_OK = qw(
     SRL_HDR_COPY
     SRL_HDR_WEAKEN
     SRL_HDR_REGEXP
+    SRL_HDR_OBJECT_FREEZE
     SRL_HDR_OBJECTV_FREEZE
     SRL_HDR_RESERVED
     SRL_HDR_RESERVED_LOW
     SRL_HDR_RESERVED_HIGH
+    SRL_HDR_CANONICAL_UNDEF
     SRL_HDR_FALSE
     SRL_HDR_TRUE
     SRL_HDR_MANY
@@ -82,6 +90,1158 @@ BEGIN { @EXPORT_OK = qw(
   };
 }
 
-sub SRL_MAGIC_STRING () {"=srl"}
-push @EXPORT_OK, "SRL_MAGIC_STRING";
+sub SRL_MAGIC_STRING ()                 { "=srl" }
+sub SRL_MAGIC_STRING_HIGHBIT ()         { "=\xF3rl" }
+sub SRL_MAGIC_STRING_HIGHBIT_UTF8 ()    { "=\xC3\xB3rl" }
+
+push @EXPORT_OK, qw(
+    SRL_MAGIC_STRING
+    SRL_MAGIC_STRING_HIGHBIT
+    SRL_MAGIC_STRING_HIGHBIT_UTF8
+);
+
+# start autoupdated section - do not modify directly
+
+our (%TAG_INFO_HASH, @TAG_INFO_ARRAY);
+@TAG_INFO_ARRAY = (
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'comment' => 'small positive integer - value in low 4 bits (identity)',
+    'value' => 0,
+    'name' => 'POS_0',
+    'masked_val' => 0,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 1,
+    'name' => 'POS_1',
+    'masked_val' => 1,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 2,
+    'name' => 'POS_2',
+    'masked_val' => 2,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 3,
+    'name' => 'POS_3',
+    'masked_val' => 3,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 4,
+    'name' => 'POS_4',
+    'masked_val' => 4,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 5,
+    'name' => 'POS_5',
+    'masked_val' => 5,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 6,
+    'name' => 'POS_6',
+    'masked_val' => 6,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 7,
+    'name' => 'POS_7',
+    'masked_val' => 7,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 8,
+    'name' => 'POS_8',
+    'masked_val' => 8,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 9,
+    'name' => 'POS_9',
+    'masked_val' => 9,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 10,
+    'name' => 'POS_10',
+    'masked_val' => 10,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 11,
+    'name' => 'POS_11',
+    'masked_val' => 11,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 12,
+    'name' => 'POS_12',
+    'masked_val' => 12,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 13,
+    'name' => 'POS_13',
+    'masked_val' => 13,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'value' => 14,
+    'name' => 'POS_14',
+    'masked_val' => 14,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'POS',
+    'masked' => 1,
+    'comment' => 'small positive integer - value in low 4 bits (identity)',
+    'value' => 15,
+    'name' => 'POS_15',
+    'masked_val' => 15,
+    'type_value' => 0
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'comment' => 'small negative integer - value in low 4 bits (k+32)',
+    'value' => 16,
+    'name' => 'NEG_16',
+    'masked_val' => 16,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 17,
+    'name' => 'NEG_15',
+    'masked_val' => 15,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 18,
+    'name' => 'NEG_14',
+    'masked_val' => 14,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 19,
+    'name' => 'NEG_13',
+    'masked_val' => 13,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 20,
+    'name' => 'NEG_12',
+    'masked_val' => 12,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 21,
+    'name' => 'NEG_11',
+    'masked_val' => 11,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 22,
+    'name' => 'NEG_10',
+    'masked_val' => 10,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 23,
+    'name' => 'NEG_9',
+    'masked_val' => 9,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 24,
+    'name' => 'NEG_8',
+    'masked_val' => 8,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 25,
+    'name' => 'NEG_7',
+    'masked_val' => 7,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 26,
+    'name' => 'NEG_6',
+    'masked_val' => 6,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 27,
+    'name' => 'NEG_5',
+    'masked_val' => 5,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 28,
+    'name' => 'NEG_4',
+    'masked_val' => 4,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 29,
+    'name' => 'NEG_3',
+    'masked_val' => 3,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'value' => 30,
+    'name' => 'NEG_2',
+    'masked_val' => 2,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'NEG',
+    'masked' => 1,
+    'comment' => 'small negative integer - value in low 4 bits (k+32)',
+    'value' => 31,
+    'name' => 'NEG_1',
+    'masked_val' => 1,
+    'type_value' => 16
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'VARINT',
+    'comment' => '<VARINT> - Varint variable length integer',
+    'value' => 32,
+    'name' => 'VARINT',
+    'type_value' => 32
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ZIGZAG',
+    'comment' => '<ZIGZAG-VARINT> - Zigzag variable length integer',
+    'value' => 33,
+    'name' => 'ZIGZAG',
+    'type_value' => 33
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'FLOAT',
+    'comment' => '<IEEE-FLOAT>',
+    'value' => 34,
+    'name' => 'FLOAT',
+    'type_value' => 34
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'DOUBLE',
+    'comment' => '<IEEE-DOUBLE>',
+    'value' => 35,
+    'name' => 'DOUBLE',
+    'type_value' => 35
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'LONG_DOUBLE',
+    'comment' => '<IEEE-LONG-DOUBLE>',
+    'value' => 36,
+    'name' => 'LONG_DOUBLE',
+    'type_value' => 36
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'UNDEF',
+    'comment' => 'None - Perl undef var; eg my $var= undef;',
+    'value' => 37,
+    'name' => 'UNDEF',
+    'type_value' => 37
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'BINARY',
+    'comment' => '<LEN-VARINT> <BYTES> - binary/(latin1) string',
+    'value' => 38,
+    'name' => 'BINARY',
+    'type_value' => 38
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'STR_UTF8',
+    'comment' => '<LEN-VARINT> <UTF8> - utf8 string',
+    'value' => 39,
+    'name' => 'STR_UTF8',
+    'type_value' => 39
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'REFN',
+    'comment' => '<ITEM-TAG>    - ref to next item',
+    'value' => 40,
+    'name' => 'REFN',
+    'type_value' => 40
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'REFP',
+    'comment' => '<OFFSET-VARINT> - ref to previous item stored at offset',
+    'value' => 41,
+    'name' => 'REFP',
+    'type_value' => 41
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASH',
+    'comment' => '<COUNT-VARINT> [<KEY-TAG> <ITEM-TAG> ...] - count followed by key/value pairs',
+    'value' => 42,
+    'name' => 'HASH',
+    'type_value' => 42
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAY',
+    'comment' => '<COUNT-VARINT> [<ITEM-TAG> ...] - count followed by items',
+    'value' => 43,
+    'name' => 'ARRAY',
+    'type_value' => 43
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'OBJECT',
+    'comment' => '<STR-TAG> <ITEM-TAG> - class, object-item',
+    'value' => 44,
+    'name' => 'OBJECT',
+    'type_value' => 44
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'OBJECTV',
+    'comment' => '<OFFSET-VARINT> <ITEM-TAG> - offset of previously used classname tag - object-item',
+    'value' => 45,
+    'name' => 'OBJECTV',
+    'type_value' => 45
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ALIAS',
+    'comment' => '<OFFSET-VARINT> - alias to item defined at offset',
+    'value' => 46,
+    'name' => 'ALIAS',
+    'type_value' => 46
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'COPY',
+    'comment' => '<OFFSET-VARINT> - copy of item defined at offset',
+    'value' => 47,
+    'name' => 'COPY',
+    'type_value' => 47
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'WEAKEN',
+    'comment' => '<REF-TAG> - Weaken the following reference',
+    'value' => 48,
+    'name' => 'WEAKEN',
+    'type_value' => 48
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'REGEXP',
+    'comment' => '<PATTERN-STR-TAG> <MODIFIERS-STR-TAG>',
+    'value' => 49,
+    'name' => 'REGEXP',
+    'type_value' => 49
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'OBJECT_FREEZE',
+    'comment' => '<STR-TAG> <ITEM-TAG> - class, object-item. Need to call "THAW" method on class after decoding',
+    'value' => 50,
+    'name' => 'OBJECT_FREEZE',
+    'type_value' => 50
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'OBJECTV_FREEZE',
+    'comment' => '<OFFSET-VARINT> <ITEM-TAG> - (OBJECTV_FREEZE is to OBJECT_FREEZE as OBJECTV is to OBJECT)',
+    'value' => 51,
+    'name' => 'OBJECTV_FREEZE',
+    'type_value' => 51
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'RESERVED',
+    'masked' => 1,
+    'comment' => 'reserved',
+    'value' => 52,
+    'name' => 'RESERVED_0',
+    'masked_val' => 0,
+    'type_value' => 52
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'RESERVED',
+    'masked' => 1,
+    'value' => 53,
+    'name' => 'RESERVED_1',
+    'masked_val' => 1,
+    'type_value' => 52
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'RESERVED',
+    'masked' => 1,
+    'value' => 54,
+    'name' => 'RESERVED_2',
+    'masked_val' => 2,
+    'type_value' => 52
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'RESERVED',
+    'masked' => 1,
+    'value' => 55,
+    'name' => 'RESERVED_3',
+    'masked_val' => 3,
+    'type_value' => 52
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'RESERVED',
+    'masked' => 1,
+    'value' => 56,
+    'name' => 'RESERVED_4',
+    'masked_val' => 4,
+    'type_value' => 52
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'CANONICAL_UNDEF',
+    'comment' => 'undef (PL_sv_undef) - "the" Perl undef (see notes)',
+    'value' => 57,
+    'name' => 'CANONICAL_UNDEF',
+    'type_value' => 57
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'FALSE',
+    'comment' => 'false (PL_sv_no)',
+    'value' => 58,
+    'name' => 'FALSE',
+    'type_value' => 58
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'TRUE',
+    'comment' => 'true  (PL_sv_yes)',
+    'value' => 59,
+    'name' => 'TRUE',
+    'type_value' => 59
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'MANY',
+    'comment' => '<LEN-VARINT> <TYPE-BYTE> <TAG-DATA> - repeated tag (not done yet, will be implemented in version 3)',
+    'value' => 60,
+    'name' => 'MANY',
+    'type_value' => 60
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'PACKET_START',
+    'comment' => '(first byte of magic string in header)',
+    'value' => 61,
+    'name' => 'PACKET_START',
+    'type_value' => 61
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'EXTEND',
+    'comment' => '<BYTE> - for additional tags',
+    'value' => 62,
+    'name' => 'EXTEND',
+    'type_value' => 62
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'PAD',
+    'comment' => '(ignored tag, skip to next byte)',
+    'value' => 63,
+    'name' => 'PAD',
+    'type_value' => 63
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'comment' => '[<ITEM-TAG> ...] - count of items in low 4 bits (ARRAY must be refcnt=1)',
+    'value' => 64,
+    'name' => 'ARRAYREF_0',
+    'masked_val' => 0,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 65,
+    'name' => 'ARRAYREF_1',
+    'masked_val' => 1,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 66,
+    'name' => 'ARRAYREF_2',
+    'masked_val' => 2,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 67,
+    'name' => 'ARRAYREF_3',
+    'masked_val' => 3,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 68,
+    'name' => 'ARRAYREF_4',
+    'masked_val' => 4,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 69,
+    'name' => 'ARRAYREF_5',
+    'masked_val' => 5,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 70,
+    'name' => 'ARRAYREF_6',
+    'masked_val' => 6,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 71,
+    'name' => 'ARRAYREF_7',
+    'masked_val' => 7,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 72,
+    'name' => 'ARRAYREF_8',
+    'masked_val' => 8,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 73,
+    'name' => 'ARRAYREF_9',
+    'masked_val' => 9,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 74,
+    'name' => 'ARRAYREF_10',
+    'masked_val' => 10,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 75,
+    'name' => 'ARRAYREF_11',
+    'masked_val' => 11,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 76,
+    'name' => 'ARRAYREF_12',
+    'masked_val' => 12,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 77,
+    'name' => 'ARRAYREF_13',
+    'masked_val' => 13,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 78,
+    'name' => 'ARRAYREF_14',
+    'masked_val' => 14,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'ARRAYREF',
+    'masked' => 1,
+    'value' => 79,
+    'name' => 'ARRAYREF_15',
+    'masked_val' => 15,
+    'type_value' => 64
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'comment' => '[<KEY-TAG> <ITEM-TAG> ...] - count in low 4 bits, key/value pairs (HASH must be refcnt=1)',
+    'value' => 80,
+    'name' => 'HASHREF_0',
+    'masked_val' => 0,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 81,
+    'name' => 'HASHREF_1',
+    'masked_val' => 1,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 82,
+    'name' => 'HASHREF_2',
+    'masked_val' => 2,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 83,
+    'name' => 'HASHREF_3',
+    'masked_val' => 3,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 84,
+    'name' => 'HASHREF_4',
+    'masked_val' => 4,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 85,
+    'name' => 'HASHREF_5',
+    'masked_val' => 5,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 86,
+    'name' => 'HASHREF_6',
+    'masked_val' => 6,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 87,
+    'name' => 'HASHREF_7',
+    'masked_val' => 7,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 88,
+    'name' => 'HASHREF_8',
+    'masked_val' => 8,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 89,
+    'name' => 'HASHREF_9',
+    'masked_val' => 9,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 90,
+    'name' => 'HASHREF_10',
+    'masked_val' => 10,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 91,
+    'name' => 'HASHREF_11',
+    'masked_val' => 11,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 92,
+    'name' => 'HASHREF_12',
+    'masked_val' => 12,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 93,
+    'name' => 'HASHREF_13',
+    'masked_val' => 13,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 94,
+    'name' => 'HASHREF_14',
+    'masked_val' => 14,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'HASHREF',
+    'masked' => 1,
+    'value' => 95,
+    'name' => 'HASHREF_15',
+    'masked_val' => 15,
+    'type_value' => 80
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'comment' => '<BYTES> - binary/latin1 string, length encoded in low 5 bits of tag',
+    'value' => 96,
+    'name' => 'SHORT_BINARY_0',
+    'masked_val' => 0,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 97,
+    'name' => 'SHORT_BINARY_1',
+    'masked_val' => 1,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 98,
+    'name' => 'SHORT_BINARY_2',
+    'masked_val' => 2,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 99,
+    'name' => 'SHORT_BINARY_3',
+    'masked_val' => 3,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 100,
+    'name' => 'SHORT_BINARY_4',
+    'masked_val' => 4,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 101,
+    'name' => 'SHORT_BINARY_5',
+    'masked_val' => 5,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 102,
+    'name' => 'SHORT_BINARY_6',
+    'masked_val' => 6,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 103,
+    'name' => 'SHORT_BINARY_7',
+    'masked_val' => 7,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 104,
+    'name' => 'SHORT_BINARY_8',
+    'masked_val' => 8,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 105,
+    'name' => 'SHORT_BINARY_9',
+    'masked_val' => 9,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 106,
+    'name' => 'SHORT_BINARY_10',
+    'masked_val' => 10,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 107,
+    'name' => 'SHORT_BINARY_11',
+    'masked_val' => 11,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 108,
+    'name' => 'SHORT_BINARY_12',
+    'masked_val' => 12,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 109,
+    'name' => 'SHORT_BINARY_13',
+    'masked_val' => 13,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 110,
+    'name' => 'SHORT_BINARY_14',
+    'masked_val' => 14,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 111,
+    'name' => 'SHORT_BINARY_15',
+    'masked_val' => 15,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 112,
+    'name' => 'SHORT_BINARY_16',
+    'masked_val' => 16,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 113,
+    'name' => 'SHORT_BINARY_17',
+    'masked_val' => 17,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 114,
+    'name' => 'SHORT_BINARY_18',
+    'masked_val' => 18,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 115,
+    'name' => 'SHORT_BINARY_19',
+    'masked_val' => 19,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 116,
+    'name' => 'SHORT_BINARY_20',
+    'masked_val' => 20,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 117,
+    'name' => 'SHORT_BINARY_21',
+    'masked_val' => 21,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 118,
+    'name' => 'SHORT_BINARY_22',
+    'masked_val' => 22,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 119,
+    'name' => 'SHORT_BINARY_23',
+    'masked_val' => 23,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 120,
+    'name' => 'SHORT_BINARY_24',
+    'masked_val' => 24,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 121,
+    'name' => 'SHORT_BINARY_25',
+    'masked_val' => 25,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 122,
+    'name' => 'SHORT_BINARY_26',
+    'masked_val' => 26,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 123,
+    'name' => 'SHORT_BINARY_27',
+    'masked_val' => 27,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 124,
+    'name' => 'SHORT_BINARY_28',
+    'masked_val' => 28,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 125,
+    'name' => 'SHORT_BINARY_29',
+    'masked_val' => 29,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 126,
+    'name' => 'SHORT_BINARY_30',
+    'masked_val' => 30,
+    'type_value' => 96
+  },
+  # autoupdated by author_tools/update_from_header.pl do not modify directly!
+  {
+    'type_name' => 'SHORT_BINARY',
+    'masked' => 1,
+    'value' => 127,
+    'name' => 'SHORT_BINARY_31',
+    'masked_val' => 31,
+    'type_value' => 96
+  }
+);
+$TAG_INFO_HASH{chr $_}= $TAG_INFO_ARRAY[$_] for 0 .. 127;
+push @EXPORT_OK, qw(%TAG_INFO_HASH @TAG_INFO_ARRAY);
+
+# stop autoupdated section - do not modify directly!
+
+
 our %EXPORT_TAGS=(all => \@EXPORT_OK);
@@ -5,22 +5,27 @@ use warnings;
 use Carp qw/croak/;
 use XSLoader;
 
-our $VERSION = '2.03'; # Don't forget to update the TestCompat set for testing against installed encoders!
+our $VERSION = '3.002'; # Don't forget to update the TestCompat set for testing against installed encoders!
+our $XS_VERSION = $VERSION; $VERSION= eval $VERSION;
 
 # not for public consumption, just for testing.
 (my $num_version = $VERSION) =~ s/_//;
-my $TestCompat = [ map sprintf("%.2f", $_/100), reverse( 200 .. int($num_version * 100) ) ]; # compat with 2.00 to ...
+my $TestCompat = [ map sprintf("%.2f", $_/100), reverse( 300 .. int($num_version * 100) ) ]; # compat with 3.00 to ...
 sub _test_compat {return(@$TestCompat, $VERSION)}
 
 use Exporter 'import';
-our @EXPORT_OK = qw(decode_sereal looks_like_sereal decode_sereal_with_header_data);
+our @EXPORT_OK = qw(
+    decode_sereal looks_like_sereal decode_sereal_with_header_data
+    scalar_looks_like_sereal
+    sereal_decode_with_object sereal_decode_with_header_with_object
+);
 our %EXPORT_TAGS = (all => \@EXPORT_OK);
 # export by default if run from command line
 our @EXPORT = ((caller())[1] eq '-e' ? @EXPORT_OK : ());
 
 sub CLONE_SKIP { 1 }
 
-XSLoader::load('Sereal::Decoder', $VERSION);
+XSLoader::load('Sereal::Decoder', $XS_VERSION);
 
 1;
 
@@ -34,7 +39,8 @@ Sereal::Decoder - Fast, compact, powerful binary deserialization
 
 =head1 SYNOPSIS
 
-  use Sereal::Decoder qw(decode_sereal looks_like_sereal);
+  use Sereal::Decoder
+    qw(decode_sereal sereal_decode_with_object scalar_looks_like_sereal);
   
   my $decoder = Sereal::Decoder->new({...options...});
   
@@ -44,12 +50,16 @@ Sereal::Decoder - Fast, compact, powerful binary deserialization
   # or if you don't have references to the top level structure, this works, too:
   $structure = $decoder->decode($blob);
   
-  # alternatively functional interface:
+  # alternatively functional interface: (See Sereal::Performance)
+  sereal_decode_with_object($decoder, $blob, $structure);
+  $structure = sereal_decode_with_object($decoder, $blob);
+
+  # much slower functional interface with no persistent objects:
   decode_sereal($blob, {... options ...}, $structure);
   $structure = decode_sereal($blob, {... options ...});
   
   # Not a full validation, but just a quick check for a reasonable header:
-  my $is_likely_sereal = looks_like_sereal($some_string);
+  my $is_likely_sereal = scalar_looks_like_sereal($some_string);
   # or:
   $is_likely_sereal = $decoder->looks_like_sereal($some_string);
 
@@ -61,10 +71,10 @@ Its sister module L<Sereal::Encoder> implements an encoder for this format.
 The two are released separately to allow for independent and safer upgrading.
 
 The Sereal protocol versions that are compatible with this decoder implementation
-are currently protocol versions 1 and 2. As it stands, it will refuse to attempt to
+are currently protocol versions 1, 2, and 3. As it stands, it will refuse to attempt to
 decode future versions of the protocol, but if necessary there is likely
 going to be an option to decode the parts of the input that are compatible
-with version 2 of the protocol. The protocol was designed to allow for this.
+with version 3 of the protocol. The protocol was designed to allow for this.
 
 The protocol specification and many other bits of documentation
 can be found in the github repository. Right now, the specification is at
@@ -139,12 +149,60 @@ This means you can do this:
         my $data= decode_sereal($buffer,{incremental=>1});
     }
 
+=head3 alias_smallint
+
+If set to a true value then C<Sereal::Decoder> will share integers from
+-16 to 15 (encoded as either SRL_HDR_NEG and SRL_HDR_POS) as read-only
+aliases to a common SV.
+
+The result of this may be significant space savings in data structures with
+many integers in the specified range. The cost is more memory used by the
+decoder and a very modest speed penalty when deserializing.
+
+Note this option changes the structure of the dumped data. Use with caution.
+
+See also the "alias_varint_under" option.
+
+=head3 alias_varint_under
+
+If set to a true positive integer smaller than 16 then this option is
+similar to setting "alias_smallint" and causes all integers from -16 to 15
+to be shared as read-only aliases to the same SV, except that this treatment
+ALSO applies to SRL_HDR_VARINT. If set to a value larger than 16 then this
+applies to all varints varints under the value set. (In general SRL_HDR_VARINT
+is used only for integers larger than 15, and SRL_HDR_NEG and SRL_HDR_POS are
+used for -16 to -1  and 0 to 15 respectively.)
+
+In simple terms if you want to share values larger than 16 then you should use
+this option, if you want to share only values in the -16 to 15 range then you
+should use the "alias_smallint" option instead.
+
+The result of this may be significant space savings in data structures with
+many integers in the desire range. The cost is more memory used by the decoder
+and a very modest speed penalty when deserializing.
+
+Note this option changes the structure of the dumped data. Use with caution.
+
+=head3 use_undef
+
+If set to a true value then this any undef value to be deserialized as
+PL_sv_undef. This may change the structure of the data structure being
+dumped, do not enable this unless you know what you are doing.
+
+=head3 set_readonly
+
+If set to a true value then the output will be completely readonly (deeply).
+
+=head3 set_readonly_scalars
+
+If set to a true value then scalars in the output will be readonly (deeply).
+References won't be readonly.
 
 =head1 INSTANCE METHODS
 
 =head2 decode
 
-Given a byte string of Sereal data, the C<decode> call derializes that data
+Given a byte string of Sereal data, the C<decode> call deserializes that data
 structure. The result can be obtained in one of two ways: C<decode> accepts
 a second parameter, which is a scalar to write the result to, AND C<decode>
 will return the resulting data structure.
@@ -161,6 +219,44 @@ In other words,
 This is an unfortunate side-effect of perls standard copy semantics of
 assignment. Possibly one day we will have an alternative to this.
 
+=head2 decode_with_header
+
+Given a byte string of Sereal data, the C<decode_with_header> call deserializes
+that data structure as C<decode> would do, however it also decodes the optional
+user data structure that can be embedded into a Sereal document, inside the
+header  (see L<Sereal::Encoder::encode>).
+
+It accepts an optional second parameter, which is a scalar to write the body
+to, and an optional third parameter, which is a scalar to write the header to.
+
+Regardless of the number of parameters received, C<decode_with_header> returns
+an ArrayRef containing the deserialized body, and the deserialized header, in
+this order.
+
+See C<decode> for the subtle difference between the one, two and three
+parameters versions.
+
+If there is no header in a Sereal document, corresponding variable or return
+value will be set to undef.
+
+=head2 decode_only_header
+
+Given a byte string of Sereal data, the C<decode_only_header> deserializes
+only the optional user data structure that can be embedded into a Sereal
+document, inside the header (see L<Sereal::Encoder::encode>).
+
+It accepts an optional second parameter, which is a scalar
+to write the header to.
+
+Regardless of the number of parameters received, C<decode_only_header> returns
+the resulting data structure.
+
+See C<decode> for the subtle difference between the one and two parameters
+versions.
+
+If there is no header in a Sereal document, corresponding variable or return
+value will be set to undef.
+
 =head2 decode_with_offset
 
 Same as the C<decode> method, except as second parameter, you must
@@ -168,14 +264,28 @@ pass an integer offset into the input string, at which the decoding is
 to start. The optional "pass-in" style scalar (see C<decode> above)
 is relegated to being the third parameter.
 
+=head2 decode_only_header_with_offset
+
+Same as the C<decode_only_header> method, except as second parameter, you must
+pass an integer offset into the input string, at which the decoding is
+to start. The optional "pass-in" style scalar (see C<decode_only_header> above)
+is relegated to being the third parameter.
+
+=head2 decode_with_header_and_offset
+
+Same as the C<decode_with_header> method, except as second parameter, you must
+pass an integer offset into the input string, at which the decoding is
+to start. The optional "pass-in" style scalars (see C<decode_with_header> above)
+are relegated to being the third and fourth parameters.
+
 =head2 bytes_consumed
 
-After using the C<decode> method, C<bytes_consumed> can return the
-number of bytes of the input string that were actually consumed by
-the decoder. That is, if you append random garbage to a valid
-Sereal document, C<decode> will happily decode the data and ignore the
-garbage. If that is an error in your use case, you can use C<bytes_consumed>
-to catch it.
+After using the various C<decode> methods documented previously,
+C<bytes_consumed> can return the number of bytes B<from the body> of the input
+string that were actually consumed by the decoder. That is, if you append
+random garbage to a valid Sereal document, C<decode> will happily decode the
+data and ignore the garbage. If that is an error in your use case, you can use
+C<bytes_consumed> to catch it.
 
   my $out = $decoder->decode($sereal_string);
   if (length($sereal_string) != $decoder->bytes_consumed) {
@@ -199,17 +309,76 @@ is concatenated into the same string (code not very robust...):
     }
   };
 
+As mentioned, only the bytes consumed from the body are considered. So the
+following example is correct, as only the header is deserialized:
+
+  my $header = $decoder->decode_only_header($sereal_string);
+  my $count = $decoder->bytes_consumed;
+  # $count is 0
+
 =head2 looks_like_sereal
 
-Given a string (or undef), checks whether it looks like it starts
-with a valid Sereal packet. This is not a full-blown validation.
-Instead, this just checks the magic string and some header properties
-to provide a quick and efficient way to distinguish multiple well-formed
-serialization methods instead of really making sure it's valid Sereal.
-For reference, sereal's magic string is a four byte string C<=srl>.
+Performs some rudimentary check to determine if the argument
+appears to be a valid Sereal packet or not. These tests are not
+comprehensive and a true result does not mean that the document
+is valid, merely that it appears to be valid. On the other hand
+a false result is always reliable.
+
+The return of this function may be treated as a simple boolean but
+is in fact a more complex return. When the argument does not
+look anything like a Sereal document then the return is perl's FALSE,
+which has the property of being string equivalent to "" and
+numerically equivalent to 0. However when the argument appears to
+be a UTF-8 encoded protocol 3 Sereal document (by noticing that
+the \xF3 in the magic string has been replaced by \xC3\xB3) then
+it returns 0 (the number, which is string equivalent to "0"), and
+otherwise returns the protocol version of the document. This means
+you can write something like this:
+
+    $type= looks_like_sereal($thing);
+    if ($type eq '') {
+        say "Not a Sereal document";
+    } elsif ($type eq '0') {
+        say "Possibly utf8 encoded Sereal document";
+    } else {
+        say "Sereal document version $type";
+    }
+
+For reference, Sereal's magic value is a four byte string which is
+either C<=srl> for protocol version 1 and 2 or C<=\xF3rl> for protocol
+version 3 and later. This function checks that the magic string
+corresponds with the reported version number, as well as other
+checks, which may be enhanced in the future.
 
 =head1 EXPORTABLE FUNCTIONS
 
+=head2 sereal_decode_with_object
+
+The functional interface that is equivalent to using C<decode>. Takes a
+decoder object reference as first parameter, followed by a byte string
+to deserialize.  Optionally takes a third parameter, which is the output
+scalar to write to. See the documentation for C<decode> above for details.
+
+This functional interface is marginally faster than the OO interface
+since it avoids method resolution overhead and, on sufficiently modern
+Perl versions, can usually avoid subroutine call overhead. See
+L<Sereal::Performance> for a discussion on how to tune Sereal for maximum
+performance if you need to.
+
+=head2 sereal_decode_with_header_with_object
+
+The functional interface that is equivalent to using C<decode_with_header>.
+Takes a decoder object reference as first parameter, followed by a byte string
+to deserialize. Optionally takes third and fourth parameters, which are
+the output scalars to write to. See the documentation for C<decode_with_header>
+above for details.
+
+This functional interface is marginally faster than the OO interface
+since it avoids method resolution overhead and, on sufficiently modern
+Perl versions, can usually avoid subroutine call overhead. See
+L<Sereal::Performance> for a discussion on how to tune Sereal for maximum
+performance if you need to.
+
 =head2 decode_sereal
 
 The functional interface that is equivalent to using C<new> and C<decode>.
@@ -218,12 +387,23 @@ by a hash reference of options (see documentation for C<new()>). Finally,
 C<decode_sereal> supports a third parameter, which is the output scalar
 to write to. See the documentation for C<decode> above for details.
 
-The functional interface is marginally slower than the OO interface since
+This functional interface is significantly slower than the OO interface since
 it cannot reuse the decoder object.
 
-=head2 looks_like_sereal
+=head2 decode_sereal_with_header_data
+
+The functional interface that is equivalent to using C<new> and C<decode_with_header>.
+Expects a byte string to deserialize as first argument, optionally followed
+by a hash reference of options (see documentation for C<new()>). Finally,
+C<decode_sereal> supports third and fourth parameters, which are the output scalars
+to write to. See the documentation for C<decode_with_header> above for details.
 
-Same as the object method of the same name.
+This functional interface is significantly slower than the OO interface since
+it cannot reuse the decoder object.
+
+=head2 scalar_looks_like_sereal
+
+The functional interface that is equivalent to using C<looks_like_sereal>.
 
 =head1 ROBUSTNESS
 
@@ -232,7 +412,7 @@ input data as reasonably possible. This means that it should never
 (though read on) segfault. It may, however, cause a large malloc
 to fail. Generally speaking, invalid data should cause a Perl-trappable
 exception. The one exception is that for Snappy-compressed Sereal documents,
-the Snappy library may cause segmentation faults (invalid reads orwrites).
+the Snappy library may cause segmentation faults (invalid reads or writes).
 This should only be a problem if you do not checksum your data (internal
 checksum support is a To-Do) or if you accept data from potentially
 malicious sources.
@@ -263,11 +443,9 @@ the C<FREEZE/THAW> mechanism, please refer to L<Sereal::Encoder>.
 
 =head1 PERFORMANCE
 
-The exact performance in time and space depends heavily on the data structure
-to be serialized. For ready-made comparison scripts, see the
-F<author_tools/bench.pl> and F<author_tools/dbench.pl> programs that are part
-of this distribution. Suffice to say that this library is easily competitive
-in both time and space efficiency with the best alternatives.
+Please refer to the L<Sereal::Performance> document
+that has more detailed information about Sereal performance and
+tuning thereof.
 
 =head1 THREAD-SAFETY
 
@@ -290,7 +468,7 @@ L<https://groups.google.com/forum/?fromgroups#!forum/sereal-announce>
 Sereal development list:
 L<https://groups.google.com/forum/?fromgroups#!forum/sereal-dev>
 
-=head1 AUTHORS
+=head1 AUTHORS AND CONTRIBUTORS
 
 Yves Orton E<lt>demerphq@gmail.comE<gt>
 
@@ -302,8 +480,16 @@ Rafaël Garcia-Suarez
 
 Ævar Arnfjörð Bjarmason E<lt>avar@cpan.orgE<gt>
 
+Tim Bunce
+
 Daniel Dragan E<lt>bulkdd@cpan.orgE<gt> (Windows support and bugfixes)
 
+Zefram
+
+Borislav Nikolov
+
+Ivan Kruglov E<lt>ivan.kruglov@yahoo.comE<gt>
+
 Some inspiration and code was taken from Marc Lehmann's
 excellent JSON::XS module due to obvious overlap in
 problem domain.
@@ -0,0 +1,265 @@
+package Sereal::Performance;
+use 5.008;
+use warnings;
+1;
+
+__END__
+
+=encoding utf8
+
+=head1 NAME
+
+Sereal::Performance - Getting the most out of the Perl-Sereal implementation
+
+=head1 SYNOPSIS
+
+  # This is different from the standard module synopsis in
+  # that it chooses performance over ease-of-use.
+  # Think twice before micro-optimizing your Sereal usage.
+  # Usually, Sereal is a lot faster than most of one's code,
+  # so unless you are doing bulk encoding/decoding, you are
+  # better off optimizing for maintainability.
+  
+  use Sereal qw(sereal_encode_with_object
+                sereal_decode_with_object);
+  my $enc = Sereal::Encoder->new();
+  my $dec = Sereal::Decoder->new();
+  
+  my $big_data_structure = {...};
+  
+  my $srldoc = sereal_encode_with_object($enc, $big_data_structure);
+  
+  my $and_back = sereal_decode_with_object($dec, $srldoc);
+
+=head1 DESCRIPTION
+
+Using Sereal in the way that is optimally performant for your use
+case can make quite a significant difference in performance. Broadly
+speaking, there are two classes of tweaks you can do: choosing
+the right options during encoding (sometimes incurring
+trade-offs in output size) and calling the Sereal encode/decode
+functions in the most efficient way.
+
+If you are not yet using re-usable
+L<Sereal::Encoder> and L<Sereal::Decoder> objects, then
+read no further. By switching from the C<encode_sereal> and
+C<decode_sereal> functions to either the OO interface or the
+advanced functional interface, you will get a noticeable
+speed boost as encoder and decoder structures can be reused.
+This is particularly significant for the encoder, which can
+re-use its output buffer. In some cases, such a warmed-up
+encoder can avoid most memory allocations.
+
+B<I repeat, if you care about performance, then do not use
+the C<encode_sereal> and C<decode_sereal> interface.>
+
+The exact performance in time and space depends heavily on the data structure
+to be (de-)serialized. Often there is a trade-off between space and time. If in doubt,
+do your own testing and most importantly B<ALWAYS TEST WITH REAL DATA>. If you
+care purely about speed at the expense of output size, you can use the
+C<no_shared_hashkeys> option for a small speed-up, see below.
+If you need smaller output at the cost of higher CPU load and more memory
+used during encoding/decoding, try the C<dedupe_strings> option and
+enable Snappy compression.
+
+For ready-made comparison scripts, see the
+F<author_tools/bench.pl> and F<author_tools/dbench.pl> programs that are part
+of this distribution. Suffice to say that this library is easily competitive
+in both time and space efficiency with the best alternatives.
+
+If switching to the OO interface is not enough, you may consider
+switching to the advanced functional interface that avoids
+method lookup overhead, and by inlining as custom Perl OPs,
+may also avoid some of the Perl function call overhead (Perl
+5.14 and up). This additional speed-up is only a constant-offset,
+avoiding said method/function call, rather than speeding up encoding
+itself and so will be most significant if you are working with
+very small data sets.
+
+C<sereal_encode_with_object> and C<sereal_decode_with_object>
+are optionally exported from the L<Sereal> module (or
+C<Sereal::Encoder> and C<Sereal::Decoder> respectively).
+They work the same as the object-oriented interface except
+that they are invoked differently:
+
+  $srl_doc = $encoder->encode($data);
+
+becomes
+
+  $srl_doc = sereal_encode_with_object($encoder, $data);
+
+and
+
+  $data = $decoder->decode($srl_doc);
+
+becomes
+
+  $data = sereal_decode_with_object($decoder, $srl_doc);
+
+On Perl versions before 5.14, this will be marginally faster than
+the OO interface as it avoids method lookup. This should rarely matter.
+On Perl versions starting from 5.14, the function call to
+C<sereal_encode_with_object> or C<sereal_decode_with_object> will
+also be replaced with a custom Perl OP, thus avoiding most of the
+function call overhead as well.
+
+=head2 Tuning the C<Sereal::Encoder>
+
+Several of the C<Sereal::Encoder> options add or remove useful
+behaviour and some of them come at a runtime performance cost.
+
+=over 2
+
+=item C<no_shared_hashkeys>
+
+By default, Sereal will emit a "repetition" marker for hash keys
+that were already previously encountered. Depending on your data
+structure, this can save quite a bit of space in the generated
+document. Consider, for example, encoding an array of many objects
+of the same class. But it may not save anything if you don't have
+a lot of repeated hash keys or don't even encode any hashes to
+begin with.
+
+In those cases, you can turn this feature off with the C<no_shared_hashkeys>
+option for a small but measurable speed-up.
+
+=item C<dedupe_strings>
+
+If set, this option will apply the de-duplication logic to all
+strings that is only applied to hash keys by default. This
+can be quite expensive in both memory and performance.
+The same is true for C<aliased_dedupe_strings>.
+
+=item C<snappy> and C<snappy_incr>
+
+Enabling Snappy compression can (but doesn't have to) make your
+Sereal documents significantly smaller. How effective this
+compression is for you depends entirely on the nature of your data.
+Snappy compression is designed to be very fast. The additional
+space savings are very often worth the small overhead.
+
+=item C<freeze_callbacks>
+
+Using custom Perl C<FREEZE> callbacks is very expensive. If enabled,
+the encoder has to do a method lookup at least once per class of an
+object being serialized. If a C<FREEZE> hook actually exists, calling
+it will be even more expensive. If you care about ultimate performance,
+use with care.
+
+=item C<sort_keys>
+
+This option forces the encoder to always C<sort> the entries in a hash
+by its keys before writing them to the Sereal document. This can be
+somewhat expensive for large hashes.
+
+=back
+
+=head2 General Considerations
+
+Perl variables (scalars specifically) can, at the same time,
+hold multiple representations of the same data. If you create
+and integer and use it as a string, it will be cached in its
+string form. Sereal attempts to detect the most compact of
+these representations for encoding, but can not always
+succeed. For example, if a data structure was previously
+also traversed by certain other serialization modules
+(such as L<Storable>), then the scalars in the structure
+may have been irrevocably upgraded to a more complex
+(and bigger) type. This is only an issue in crude benchmarks.
+So if you plan to benchmark serialization, take care not
+to re-use the test data structure between serializers for
+results that do not depend on the order of operations.
+
+=head1 BUGS, CONTACT AND SUPPORT
+
+For reporting bugs, please use the github bug tracker at
+L<http://github.com/Sereal/Sereal/issues>.
+
+For support and discussion of Sereal, there are two Google Groups:
+
+Announcements around Sereal (extremely low volume):
+L<https://groups.google.com/forum/?fromgroups#!forum/sereal-announce>
+
+Sereal development list:
+L<https://groups.google.com/forum/?fromgroups#!forum/sereal-dev>
+
+=head1 AUTHORS AND CONTRIBUTORS
+
+Yves Orton E<lt>demerphq@gmail.comE<gt>
+
+Damian Gryski
+
+Steffen Mueller E<lt>smueller@cpan.orgE<gt>
+
+Rafaël Garcia-Suarez
+
+Ævar Arnfjörð Bjarmason E<lt>avar@cpan.orgE<gt>
+
+Tim Bunce
+
+Daniel Dragan E<lt>bulkdd@cpan.orgE<gt> (Windows support and bugfixes)
+
+Zefram
+
+Some inspiration and code was taken from Marc Lehmann's
+excellent JSON::XS module due to obvious overlap in
+problem domain.
+
+=head1 ACKNOWLEDGMENT
+
+This module was originally developed for Booking.com.
+With approval from Booking.com, this module was generalized
+and published on CPAN, for which the authors would like to express
+their gratitude.
+
+=head1 COPYRIGHT AND LICENSE
+
+Copyright (C) 2012, 2013, 2014 by Steffen Mueller
+Copyright (C) 2012, 2013, 2014 by Yves Orton
+
+The license for the code in this distribution is the following,
+with the exceptions listed below:
+
+This library is free software; you can redistribute it and/or modify
+it under the same terms as Perl itself.
+
+Except portions taken from Marc Lehmann's code for the JSON::XS
+module, which is licensed under the same terms as this module.
+(Many thanks to Marc for inspiration, and code.)
+
+Also except the code for Snappy compression library, whose license
+is reproduced below and which, to the best of our knowledge,
+is compatible with this module's license. The license for the
+enclosed Snappy code is:
+
+  Copyright 2011, Google Inc.
+  All rights reserved.
+
+  Redistribution and use in source and binary forms, with or without
+  modification, are permitted provided that the following conditions are
+  met:
+
+    * Redistributions of source code must retain the above copyright
+  notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+  copyright notice, this list of conditions and the following disclaimer
+  in the documentation and/or other materials provided with the
+  distribution.
+    * Neither the name of Google Inc. nor the names of its
+  contributors may be used to endorse or promote products derived from
+  this software without specific prior written permission.
+
+  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+  "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+  A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+  OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+  SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+  LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+  DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+  THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+  (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+  OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=cut
+
@@ -0,0 +1,4922 @@
+/* miniz.c v1.15 - public domain deflate/inflate, zlib-subset, ZIP reading/writing/appending, PNG writing
+   See "unlicense" statement at the end of this file.
+   Rich Geldreich <richgel99@gmail.com>, last updated Oct. 13, 2013
+   Implements RFC 1950: http://www.ietf.org/rfc/rfc1950.txt and RFC 1951: http://www.ietf.org/rfc/rfc1951.txt
+
+   Most API's defined in miniz.c are optional. For example, to disable the archive related functions just define
+   MINIZ_NO_ARCHIVE_APIS, or to get rid of all stdio usage define MINIZ_NO_STDIO (see the list below for more macros).
+
+   * Change History
+     10/13/13 v1.15 r4 - Interim bugfix release while I work on the next major release with Zip64 support (almost there!):
+       - Critical fix for the MZ_ZIP_FLAG_DO_NOT_SORT_CENTRAL_DIRECTORY bug (thanks kahmyong.moon@hp.com) which could cause locate files to not find files. This bug
+        would only have occured in earlier versions if you explicitly used this flag, OR if you used mz_zip_extract_archive_file_to_heap() or mz_zip_add_mem_to_archive_file_in_place()
+        (which used this flag). If you can't switch to v1.15 but want to fix this bug, just remove the uses of this flag from both helper funcs (and of course don't use the flag).
+       - Bugfix in mz_zip_reader_extract_to_mem_no_alloc() from kymoon when pUser_read_buf is not NULL and compressed size is > uncompressed size
+       - Fixing mz_zip_reader_extract_*() funcs so they don't try to extract compressed data from directory entries, to account for weird zipfiles which contain zero-size compressed data on dir entries.
+         Hopefully this fix won't cause any issues on weird zip archives, because it assumes the low 16-bits of zip external attributes are DOS attributes (which I believe they always are in practice).
+       - Fixing mz_zip_reader_is_file_a_directory() so it doesn't check the internal attributes, just the filename and external attributes
+       - mz_zip_reader_init_file() - missing MZ_FCLOSE() call if the seek failed
+       - Added cmake support for Linux builds which builds all the examples, tested with clang v3.3 and gcc v4.6.
+       - Clang fix for tdefl_write_image_to_png_file_in_memory() from toffaletti
+       - Merged MZ_FORCEINLINE fix from hdeanclark
+       - Fix <time.h> include before config #ifdef, thanks emil.brink
+       - Added tdefl_write_image_to_png_file_in_memory_ex(): supports Y flipping (super useful for OpenGL apps), and explicit control over the compression level (so you can
+        set it to 1 for real-time compression).
+       - Merged in some compiler fixes from paulharris's github repro.
+       - Retested this build under Windows (VS 2010, including static analysis), tcc  0.9.26, gcc v4.6 and clang v3.3.
+       - Added example6.c, which dumps an image of the mandelbrot set to a PNG file.
+       - Modified example2 to help test the MZ_ZIP_FLAG_DO_NOT_SORT_CENTRAL_DIRECTORY flag more.
+       - In r3: Bugfix to mz_zip_writer_add_file() found during merge: Fix possible src file fclose() leak if alignment bytes+local header file write faiiled
+		 - In r4: Minor bugfix to mz_zip_writer_add_from_zip_reader(): Was pushing the wrong central dir header offset, appears harmless in this release, but it became a problem in the zip64 branch
+     5/20/12 v1.14 - MinGW32/64 GCC 4.6.1 compiler fixes: added MZ_FORCEINLINE, #include <time.h> (thanks fermtect).
+     5/19/12 v1.13 - From jason@cornsyrup.org and kelwert@mtu.edu - Fix mz_crc32() so it doesn't compute the wrong CRC-32's when mz_ulong is 64-bit.
+       - Temporarily/locally slammed in "typedef unsigned long mz_ulong" and re-ran a randomized regression test on ~500k files.
+       - Eliminated a bunch of warnings when compiling with GCC 32-bit/64.
+       - Ran all examples, miniz.c, and tinfl.c through MSVC 2008's /analyze (static analysis) option and fixed all warnings (except for the silly
+        "Use of the comma-operator in a tested expression.." analysis warning, which I purposely use to work around a MSVC compiler warning).
+       - Created 32-bit and 64-bit Codeblocks projects/workspace. Built and tested Linux executables. The codeblocks workspace is compatible with Linux+Win32/x64.
+       - Added miniz_tester solution/project, which is a useful little app derived from LZHAM's tester app that I use as part of the regression test.
+       - Ran miniz.c and tinfl.c through another series of regression testing on ~500,000 files and archives.
+       - Modified example5.c so it purposely disables a bunch of high-level functionality (MINIZ_NO_STDIO, etc.). (Thanks to corysama for the MINIZ_NO_STDIO bug report.)
+       - Fix ftell() usage in examples so they exit with an error on files which are too large (a limitation of the examples, not miniz itself).
+     4/12/12 v1.12 - More comments, added low-level example5.c, fixed a couple minor level_and_flags issues in the archive API's.
+      level_and_flags can now be set to MZ_DEFAULT_COMPRESSION. Thanks to Bruce Dawson <bruced@valvesoftware.com> for the feedback/bug report.
+     5/28/11 v1.11 - Added statement from unlicense.org
+     5/27/11 v1.10 - Substantial compressor optimizations:
+      - Level 1 is now ~4x faster than before. The L1 compressor's throughput now varies between 70-110MB/sec. on a
+      - Core i7 (actual throughput varies depending on the type of data, and x64 vs. x86).
+      - Improved baseline L2-L9 compression perf. Also, greatly improved compression perf. issues on some file types.
+      - Refactored the compression code for better readability and maintainability.
+      - Added level 10 compression level (L10 has slightly better ratio than level 9, but could have a potentially large
+       drop in throughput on some files).
+     5/15/11 v1.09 - Initial stable release.
+
+   * Low-level Deflate/Inflate implementation notes:
+
+     Compression: Use the "tdefl" API's. The compressor supports raw, static, and dynamic blocks, lazy or
+     greedy parsing, match length filtering, RLE-only, and Huffman-only streams. It performs and compresses
+     approximately as well as zlib.
+
+     Decompression: Use the "tinfl" API's. The entire decompressor is implemented as a single function
+     coroutine: see tinfl_decompress(). It supports decompression into a 32KB (or larger power of 2) wrapping buffer, or into a memory
+     block large enough to hold the entire file.
+
+     The low-level tdefl/tinfl API's do not make any use of dynamic memory allocation.
+
+   * zlib-style API notes:
+
+     miniz.c implements a fairly large subset of zlib. There's enough functionality present for it to be a drop-in
+     zlib replacement in many apps:
+        The z_stream struct, optional memory allocation callbacks
+        deflateInit/deflateInit2/deflate/deflateReset/deflateEnd/deflateBound
+        inflateInit/inflateInit2/inflate/inflateEnd
+        compress, compress2, compressBound, uncompress
+        CRC-32, Adler-32 - Using modern, minimal code size, CPU cache friendly routines.
+        Supports raw deflate streams or standard zlib streams with adler-32 checking.
+
+     Limitations:
+      The callback API's are not implemented yet. No support for gzip headers or zlib static dictionaries.
+      I've tried to closely emulate zlib's various flavors of stream flushing and return status codes, but
+      there are no guarantees that miniz.c pulls this off perfectly.
+
+   * PNG writing: See the tdefl_write_image_to_png_file_in_memory() function, originally written by
+     Alex Evans. Supports 1-4 bytes/pixel images.
+
+   * ZIP archive API notes:
+
+     The ZIP archive API's where designed with simplicity and efficiency in mind, with just enough abstraction to
+     get the job done with minimal fuss. There are simple API's to retrieve file information, read files from
+     existing archives, create new archives, append new files to existing archives, or clone archive data from
+     one archive to another. It supports archives located in memory or the heap, on disk (using stdio.h),
+     or you can specify custom file read/write callbacks.
+
+     - Archive reading: Just call this function to read a single file from a disk archive:
+
+      void *mz_zip_extract_archive_file_to_heap(const char *pZip_filename, const char *pArchive_name,
+        size_t *pSize, mz_uint zip_flags);
+
+     For more complex cases, use the "mz_zip_reader" functions. Upon opening an archive, the entire central
+     directory is located and read as-is into memory, and subsequent file access only occurs when reading individual files.
+
+     - Archives file scanning: The simple way is to use this function to scan a loaded archive for a specific file:
+
+     int mz_zip_reader_locate_file(mz_zip_archive *pZip, const char *pName, const char *pComment, mz_uint flags);
+
+     The locate operation can optionally check file comments too, which (as one example) can be used to identify
+     multiple versions of the same file in an archive. This function uses a simple linear search through the central
+     directory, so it's not very fast.
+
+     Alternately, you can iterate through all the files in an archive (using mz_zip_reader_get_num_files()) and
+     retrieve detailed info on each file by calling mz_zip_reader_file_stat().
+
+     - Archive creation: Use the "mz_zip_writer" functions. The ZIP writer immediately writes compressed file data
+     to disk and builds an exact image of the central directory in memory. The central directory image is written
+     all at once at the end of the archive file when the archive is finalized.
+
+     The archive writer can optionally align each file's local header and file data to any power of 2 alignment,
+     which can be useful when the archive will be read from optical media. Also, the writer supports placing
+     arbitrary data blobs at the very beginning of ZIP archives. Archives written using either feature are still
+     readable by any ZIP tool.
+
+     - Archive appending: The simple way to add a single file to an archive is to call this function:
+
+      mz_bool mz_zip_add_mem_to_archive_file_in_place(const char *pZip_filename, const char *pArchive_name,
+        const void *pBuf, size_t buf_size, const void *pComment, mz_uint16 comment_size, mz_uint level_and_flags);
+
+     The archive will be created if it doesn't already exist, otherwise it'll be appended to.
+     Note the appending is done in-place and is not an atomic operation, so if something goes wrong
+     during the operation it's possible the archive could be left without a central directory (although the local
+     file headers and file data will be fine, so the archive will be recoverable).
+
+     For more complex archive modification scenarios:
+     1. The safest way is to use a mz_zip_reader to read the existing archive, cloning only those bits you want to
+     preserve into a new archive using using the mz_zip_writer_add_from_zip_reader() function (which compiles the
+     compressed file data as-is). When you're done, delete the old archive and rename the newly written archive, and
+     you're done. This is safe but requires a bunch of temporary disk space or heap memory.
+
+     2. Or, you can convert an mz_zip_reader in-place to an mz_zip_writer using mz_zip_writer_init_from_reader(),
+     append new files as needed, then finalize the archive which will write an updated central directory to the
+     original archive. (This is basically what mz_zip_add_mem_to_archive_file_in_place() does.) There's a
+     possibility that the archive's central directory could be lost with this method if anything goes wrong, though.
+
+     - ZIP archive support limitations:
+     No zip64 or spanning support. Extraction functions can only handle unencrypted, stored or deflated files.
+     Requires streams capable of seeking.
+
+   * This is a header file library, like stb_image.c. To get only a header file, either cut and paste the
+     below header, or create miniz.h, #define MINIZ_HEADER_FILE_ONLY, and then include miniz.c from it.
+
+   * Important: For best perf. be sure to customize the below macros for your target platform:
+     #define MINIZ_USE_UNALIGNED_LOADS_AND_STORES 1
+     #define MINIZ_LITTLE_ENDIAN 1
+     #define MINIZ_HAS_64BIT_REGISTERS 1
+
+   * On platforms using glibc, Be sure to "#define _LARGEFILE64_SOURCE 1" before including miniz.c to ensure miniz
+     uses the 64-bit variants: fopen64(), stat64(), etc. Otherwise you won't be able to process large files
+     (i.e. 32-bit stat() fails for me on files > 0x7FFFFFFF bytes).
+*/
+
+#ifndef MINIZ_HEADER_INCLUDED
+#define MINIZ_HEADER_INCLUDED
+
+#include <stdlib.h>
+
+// Defines to completely disable specific portions of miniz.c:
+// If all macros here are defined the only functionality remaining will be CRC-32, adler-32, tinfl, and tdefl.
+
+// Define MINIZ_NO_STDIO to disable all usage and any functions which rely on stdio for file I/O.
+//#define MINIZ_NO_STDIO
+
+// If MINIZ_NO_TIME is specified then the ZIP archive functions will not be able to get the current time, or
+// get/set file times, and the C run-time funcs that get/set times won't be called.
+// The current downside is the times written to your archives will be from 1979.
+//#define MINIZ_NO_TIME
+
+// Define MINIZ_NO_ARCHIVE_APIS to disable all ZIP archive API's.
+//#define MINIZ_NO_ARCHIVE_APIS
+
+// Define MINIZ_NO_ARCHIVE_APIS to disable all writing related ZIP archive API's.
+//#define MINIZ_NO_ARCHIVE_WRITING_APIS
+
+// Define MINIZ_NO_ZLIB_APIS to remove all ZLIB-style compression/decompression API's.
+//#define MINIZ_NO_ZLIB_APIS
+
+// Define MINIZ_NO_ZLIB_COMPATIBLE_NAME to disable zlib names, to prevent conflicts against stock zlib.
+//#define MINIZ_NO_ZLIB_COMPATIBLE_NAMES
+
+// Define MINIZ_NO_MALLOC to disable all calls to malloc, free, and realloc.
+// Note if MINIZ_NO_MALLOC is defined then the user must always provide custom user alloc/free/realloc
+// callbacks to the zlib and archive API's, and a few stand-alone helper API's which don't provide custom user
+// functions (such as tdefl_compress_mem_to_heap() and tinfl_decompress_mem_to_heap()) won't work.
+//#define MINIZ_NO_MALLOC
+
+#if defined(__TINYC__) && (defined(__linux) || defined(__linux__))
+  // TODO: Work around "error: include file 'sys\utime.h' when compiling with tcc on Linux
+  #define MINIZ_NO_TIME
+#endif
+
+#if !defined(MINIZ_NO_TIME) && !defined(MINIZ_NO_ARCHIVE_APIS)
+  #include <time.h>
+#endif
+
+#if defined(_M_IX86) || defined(_M_X64) || defined(__i386__) || defined(__i386) || defined(__i486__) || defined(__i486) || defined(i386) || defined(__ia64__) || defined(__x86_64__)
+// MINIZ_X86_OR_X64_CPU is only used to help set the below macros.
+#define MINIZ_X86_OR_X64_CPU 1
+#endif
+
+#if (__BYTE_ORDER__==__ORDER_LITTLE_ENDIAN__) || MINIZ_X86_OR_X64_CPU
+// Set MINIZ_LITTLE_ENDIAN to 1 if the processor is little endian.
+#define MINIZ_LITTLE_ENDIAN 1
+#endif
+
+#if MINIZ_X86_OR_X64_CPU
+// Set MINIZ_USE_UNALIGNED_LOADS_AND_STORES to 1 on CPU's that permit efficient integer loads and stores from unaligned addresses.
+#define MINIZ_USE_UNALIGNED_LOADS_AND_STORES 1
+#endif
+
+/* HP-UX runs on Itanium but has strict alignment. */
+#ifdef __hpux
+#undef MINIZ_USE_UNALIGNED_LOADS_AND_STORES
+#define MINIZ_USE_UNALIGNED_LOADS_AND_STORES 0
+#endif
+
+#if defined(_M_X64) || defined(_WIN64) || defined(__MINGW64__) || defined(_LP64) || defined(__LP64__) || defined(__ILP64__) || defined(__ia64__) || defined(__x86_64__)
+// Set MINIZ_HAS_64BIT_REGISTERS to 1 if operations on 64-bit integers are reasonably fast (and don't involve compiler generated calls to helper functions).
+#define MINIZ_HAS_64BIT_REGISTERS 1
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// ------------------- zlib-style API Definitions.
+
+// For more compatibility with zlib, miniz.c uses unsigned long for some parameters/struct members. Beware: mz_ulong can be either 32 or 64-bits!
+typedef unsigned long mz_ulong;
+
+// mz_free() internally uses the MZ_FREE() macro (which by default calls free() unless you've modified the MZ_MALLOC macro) to release a block allocated from the heap.
+void mz_free(void *p);
+
+#define MZ_ADLER32_INIT (1)
+// mz_adler32() returns the initial adler-32 value to use when called with ptr==NULL.
+mz_ulong mz_adler32(mz_ulong adler, const unsigned char *ptr, size_t buf_len);
+
+#define MZ_CRC32_INIT (0)
+// mz_crc32() returns the initial CRC-32 value to use when called with ptr==NULL.
+mz_ulong mz_crc32(mz_ulong crc, const unsigned char *ptr, size_t buf_len);
+
+// Compression strategies.
+enum { MZ_DEFAULT_STRATEGY = 0, MZ_FILTERED = 1, MZ_HUFFMAN_ONLY = 2, MZ_RLE = 3, MZ_FIXED = 4 };
+
+// Method
+#define MZ_DEFLATED 8
+
+#ifndef MINIZ_NO_ZLIB_APIS
+
+// Heap allocation callbacks.
+// Note that mz_alloc_func parameter types purpsosely differ from zlib's: items/size is size_t, not unsigned long.
+typedef void *(*mz_alloc_func)(void *opaque, size_t items, size_t size);
+typedef void (*mz_free_func)(void *opaque, void *address);
+typedef void *(*mz_realloc_func)(void *opaque, void *address, size_t items, size_t size);
+
+#define MZ_VERSION          "9.1.15"
+#define MZ_VERNUM           0x91F0
+#define MZ_VER_MAJOR        9
+#define MZ_VER_MINOR        1
+#define MZ_VER_REVISION     15
+#define MZ_VER_SUBREVISION  0
+
+// Flush values. For typical usage you only need MZ_NO_FLUSH and MZ_FINISH. The other values are for advanced use (refer to the zlib docs).
+enum { MZ_NO_FLUSH = 0, MZ_PARTIAL_FLUSH = 1, MZ_SYNC_FLUSH = 2, MZ_FULL_FLUSH = 3, MZ_FINISH = 4, MZ_BLOCK = 5 };
+
+// Return status codes. MZ_PARAM_ERROR is non-standard.
+enum { MZ_OK = 0, MZ_STREAM_END = 1, MZ_NEED_DICT = 2, MZ_ERRNO = -1, MZ_STREAM_ERROR = -2, MZ_DATA_ERROR = -3, MZ_MEM_ERROR = -4, MZ_BUF_ERROR = -5, MZ_VERSION_ERROR = -6, MZ_PARAM_ERROR = -10000 };
+
+// Compression levels: 0-9 are the standard zlib-style levels, 10 is best possible compression (not zlib compatible, and may be very slow), MZ_DEFAULT_COMPRESSION=MZ_DEFAULT_LEVEL.
+enum { MZ_NO_COMPRESSION = 0, MZ_BEST_SPEED = 1, MZ_BEST_COMPRESSION = 9, MZ_UBER_COMPRESSION = 10, MZ_DEFAULT_LEVEL = 6, MZ_DEFAULT_COMPRESSION = -1 };
+
+// Window bits
+#define MZ_DEFAULT_WINDOW_BITS 15
+
+struct mz_internal_state;
+
+// Compression/decompression stream struct.
+typedef struct mz_stream_s
+{
+  const unsigned char *next_in;     // pointer to next byte to read
+  unsigned int avail_in;            // number of bytes available at next_in
+  mz_ulong total_in;                // total number of bytes consumed so far
+
+  unsigned char *next_out;          // pointer to next byte to write
+  unsigned int avail_out;           // number of bytes that can be written to next_out
+  mz_ulong total_out;               // total number of bytes produced so far
+
+  char *msg;                        // error msg (unused)
+  struct mz_internal_state *state;  // internal state, allocated by zalloc/zfree
+
+  mz_alloc_func zalloc;             // optional heap allocation function (defaults to malloc)
+  mz_free_func zfree;               // optional heap free function (defaults to free)
+  void *opaque;                     // heap alloc function user pointer
+
+  int data_type;                    // data_type (unused)
+  mz_ulong adler;                   // adler32 of the source or uncompressed data
+  mz_ulong reserved;                // not used
+} mz_stream;
+
+typedef mz_stream *mz_streamp;
+
+// Returns the version string of miniz.c.
+const char *mz_version(void);
+
+// mz_deflateInit() initializes a compressor with default options:
+// Parameters:
+//  pStream must point to an initialized mz_stream struct.
+//  level must be between [MZ_NO_COMPRESSION, MZ_BEST_COMPRESSION].
+//  level 1 enables a specially optimized compression function that's been optimized purely for performance, not ratio.
+//  (This special func. is currently only enabled when MINIZ_USE_UNALIGNED_LOADS_AND_STORES and MINIZ_LITTLE_ENDIAN are defined.)
+// Return values:
+//  MZ_OK on success.
+//  MZ_STREAM_ERROR if the stream is bogus.
+//  MZ_PARAM_ERROR if the input parameters are bogus.
+//  MZ_MEM_ERROR on out of memory.
+int mz_deflateInit(mz_streamp pStream, int level);
+
+// mz_deflateInit2() is like mz_deflate(), except with more control:
+// Additional parameters:
+//   method must be MZ_DEFLATED
+//   window_bits must be MZ_DEFAULT_WINDOW_BITS (to wrap the deflate stream with zlib header/adler-32 footer) or -MZ_DEFAULT_WINDOW_BITS (raw deflate/no header or footer)
+//   mem_level must be between [1, 9] (it's checked but ignored by miniz.c)
+int mz_deflateInit2(mz_streamp pStream, int level, int method, int window_bits, int mem_level, int strategy);
+
+// Quickly resets a compressor without having to reallocate anything. Same as calling mz_deflateEnd() followed by mz_deflateInit()/mz_deflateInit2().
+int mz_deflateReset(mz_streamp pStream);
+
+// mz_deflate() compresses the input to output, consuming as much of the input and producing as much output as possible.
+// Parameters:
+//   pStream is the stream to read from and write to. You must initialize/update the next_in, avail_in, next_out, and avail_out members.
+//   flush may be MZ_NO_FLUSH, MZ_PARTIAL_FLUSH/MZ_SYNC_FLUSH, MZ_FULL_FLUSH, or MZ_FINISH.
+// Return values:
+//   MZ_OK on success (when flushing, or if more input is needed but not available, and/or there's more output to be written but the output buffer is full).
+//   MZ_STREAM_END if all input has been consumed and all output bytes have been written. Don't call mz_deflate() on the stream anymore.
+//   MZ_STREAM_ERROR if the stream is bogus.
+//   MZ_PARAM_ERROR if one of the parameters is invalid.
+//   MZ_BUF_ERROR if no forward progress is possible because the input and/or output buffers are empty. (Fill up the input buffer or free up some output space and try again.)
+int mz_deflate(mz_streamp pStream, int flush);
+
+// mz_deflateEnd() deinitializes a compressor:
+// Return values:
+//  MZ_OK on success.
+//  MZ_STREAM_ERROR if the stream is bogus.
+int mz_deflateEnd(mz_streamp pStream);
+
+// mz_deflateBound() returns a (very) conservative upper bound on the amount of data that could be generated by deflate(), assuming flush is set to only MZ_NO_FLUSH or MZ_FINISH.
+mz_ulong mz_deflateBound(mz_streamp pStream, mz_ulong source_len);
+
+// Single-call compression functions mz_compress() and mz_compress2():
+// Returns MZ_OK on success, or one of the error codes from mz_deflate() on failure.
+int mz_compress(unsigned char *pDest, mz_ulong *pDest_len, const unsigned char *pSource, mz_ulong source_len);
+int mz_compress2(unsigned char *pDest, mz_ulong *pDest_len, const unsigned char *pSource, mz_ulong source_len, int level);
+
+// mz_compressBound() returns a (very) conservative upper bound on the amount of data that could be generated by calling mz_compress().
+mz_ulong mz_compressBound(mz_ulong source_len);
+
+// Initializes a decompressor.
+int mz_inflateInit(mz_streamp pStream);
+
+// mz_inflateInit2() is like mz_inflateInit() with an additional option that controls the window size and whether or not the stream has been wrapped with a zlib header/footer:
+// window_bits must be MZ_DEFAULT_WINDOW_BITS (to parse zlib header/footer) or -MZ_DEFAULT_WINDOW_BITS (raw deflate).
+int mz_inflateInit2(mz_streamp pStream, int window_bits);
+
+// Decompresses the input stream to the output, consuming only as much of the input as needed, and writing as much to the output as possible.
+// Parameters:
+//   pStream is the stream to read from and write to. You must initialize/update the next_in, avail_in, next_out, and avail_out members.
+//   flush may be MZ_NO_FLUSH, MZ_SYNC_FLUSH, or MZ_FINISH.
+//   On the first call, if flush is MZ_FINISH it's assumed the input and output buffers are both sized large enough to decompress the entire stream in a single call (this is slightly faster).
+//   MZ_FINISH implies that there are no more source bytes available beside what's already in the input buffer, and that the output buffer is large enough to hold the rest of the decompressed data.
+// Return values:
+//   MZ_OK on success. Either more input is needed but not available, and/or there's more output to be written but the output buffer is full.
+//   MZ_STREAM_END if all needed input has been consumed and all output bytes have been written. For zlib streams, the adler-32 of the decompressed data has also been verified.
+//   MZ_STREAM_ERROR if the stream is bogus.
+//   MZ_DATA_ERROR if the deflate stream is invalid.
+//   MZ_PARAM_ERROR if one of the parameters is invalid.
+//   MZ_BUF_ERROR if no forward progress is possible because the input buffer is empty but the inflater needs more input to continue, or if the output buffer is not large enough. Call mz_inflate() again
+//   with more input data, or with more room in the output buffer (except when using single call decompression, described above).
+int mz_inflate(mz_streamp pStream, int flush);
+
+// Deinitializes a decompressor.
+int mz_inflateEnd(mz_streamp pStream);
+
+// Single-call decompression.
+// Returns MZ_OK on success, or one of the error codes from mz_inflate() on failure.
+int mz_uncompress(unsigned char *pDest, mz_ulong *pDest_len, const unsigned char *pSource, mz_ulong source_len);
+
+// Returns a string description of the specified error code, or NULL if the error code is invalid.
+const char *mz_error(int err);
+
+// Redefine zlib-compatible names to miniz equivalents, so miniz.c can be used as a drop-in replacement for the subset of zlib that miniz.c supports.
+// Define MINIZ_NO_ZLIB_COMPATIBLE_NAMES to disable zlib-compatibility if you use zlib in the same project.
+#ifndef MINIZ_NO_ZLIB_COMPATIBLE_NAMES
+  typedef unsigned char Byte;
+  typedef unsigned int uInt;
+  typedef mz_ulong uLong;
+  typedef Byte Bytef;
+  typedef uInt uIntf;
+  typedef char charf;
+  typedef int intf;
+  typedef void *voidpf;
+  typedef uLong uLongf;
+  typedef void *voidp;
+  typedef void *const voidpc;
+  #define Z_NULL                0
+  #define Z_NO_FLUSH            MZ_NO_FLUSH
+  #define Z_PARTIAL_FLUSH       MZ_PARTIAL_FLUSH
+  #define Z_SYNC_FLUSH          MZ_SYNC_FLUSH
+  #define Z_FULL_FLUSH          MZ_FULL_FLUSH
+  #define Z_FINISH              MZ_FINISH
+  #define Z_BLOCK               MZ_BLOCK
+  #define Z_OK                  MZ_OK
+  #define Z_STREAM_END          MZ_STREAM_END
+  #define Z_NEED_DICT           MZ_NEED_DICT
+  #define Z_ERRNO               MZ_ERRNO
+  #define Z_STREAM_ERROR        MZ_STREAM_ERROR
+  #define Z_DATA_ERROR          MZ_DATA_ERROR
+  #define Z_MEM_ERROR           MZ_MEM_ERROR
+  #define Z_BUF_ERROR           MZ_BUF_ERROR
+  #define Z_VERSION_ERROR       MZ_VERSION_ERROR
+  #define Z_PARAM_ERROR         MZ_PARAM_ERROR
+  #define Z_NO_COMPRESSION      MZ_NO_COMPRESSION
+  #define Z_BEST_SPEED          MZ_BEST_SPEED
+  #define Z_BEST_COMPRESSION    MZ_BEST_COMPRESSION
+  #define Z_DEFAULT_COMPRESSION MZ_DEFAULT_COMPRESSION
+  #define Z_DEFAULT_STRATEGY    MZ_DEFAULT_STRATEGY
+  #define Z_FILTERED            MZ_FILTERED
+  #define Z_HUFFMAN_ONLY        MZ_HUFFMAN_ONLY
+  #define Z_RLE                 MZ_RLE
+  #define Z_FIXED               MZ_FIXED
+  #define Z_DEFLATED            MZ_DEFLATED
+  #define Z_DEFAULT_WINDOW_BITS MZ_DEFAULT_WINDOW_BITS
+  #define alloc_func            mz_alloc_func
+  #define free_func             mz_free_func
+  #define internal_state        mz_internal_state
+  #define z_stream              mz_stream
+  #define deflateInit           mz_deflateInit
+  #define deflateInit2          mz_deflateInit2
+  #define deflateReset          mz_deflateReset
+  #define deflate               mz_deflate
+  #define deflateEnd            mz_deflateEnd
+  #define deflateBound          mz_deflateBound
+  #define compress              mz_compress
+  #define compress2             mz_compress2
+  #define compressBound         mz_compressBound
+  #define inflateInit           mz_inflateInit
+  #define inflateInit2          mz_inflateInit2
+  #define inflate               mz_inflate
+  #define inflateEnd            mz_inflateEnd
+  #define uncompress            mz_uncompress
+  #define crc32                 mz_crc32
+  #define adler32               mz_adler32
+  #define MAX_WBITS             15
+  #define MAX_MEM_LEVEL         9
+  #define zError                mz_error
+  #define ZLIB_VERSION          MZ_VERSION
+  #define ZLIB_VERNUM           MZ_VERNUM
+  #define ZLIB_VER_MAJOR        MZ_VER_MAJOR
+  #define ZLIB_VER_MINOR        MZ_VER_MINOR
+  #define ZLIB_VER_REVISION     MZ_VER_REVISION
+  #define ZLIB_VER_SUBREVISION  MZ_VER_SUBREVISION
+  #define zlibVersion           mz_version
+  #define zlib_version          mz_version()
+#endif // #ifndef MINIZ_NO_ZLIB_COMPATIBLE_NAMES
+
+#endif // MINIZ_NO_ZLIB_APIS
+
+// ------------------- Types and macros
+
+typedef unsigned char mz_uint8;
+typedef signed short mz_int16;
+typedef unsigned short mz_uint16;
+typedef unsigned int mz_uint32;
+typedef unsigned int mz_uint;
+typedef long long mz_int64;
+typedef unsigned long long mz_uint64;
+typedef int mz_bool;
+
+#define MZ_FALSE (0)
+#define MZ_TRUE (1)
+
+// An attempt to work around MSVC's spammy "warning C4127: conditional expression is constant" message.
+#ifdef _MSC_VER
+   #define MZ_MACRO_END while (0, 0)
+#else
+   #define MZ_MACRO_END while (0)
+#endif
+
+// ------------------- ZIP archive reading/writing
+
+#ifndef MINIZ_NO_ARCHIVE_APIS
+
+enum
+{
+  MZ_ZIP_MAX_IO_BUF_SIZE = 64*1024,
+  MZ_ZIP_MAX_ARCHIVE_FILENAME_SIZE = 260,
+  MZ_ZIP_MAX_ARCHIVE_FILE_COMMENT_SIZE = 256
+};
+
+typedef struct
+{
+  mz_uint32 m_file_index;
+  mz_uint32 m_central_dir_ofs;
+  mz_uint16 m_version_made_by;
+  mz_uint16 m_version_needed;
+  mz_uint16 m_bit_flag;
+  mz_uint16 m_method;
+#ifndef MINIZ_NO_TIME
+  time_t m_time;
+#endif
+  mz_uint32 m_crc32;
+  mz_uint64 m_comp_size;
+  mz_uint64 m_uncomp_size;
+  mz_uint16 m_internal_attr;
+  mz_uint32 m_external_attr;
+  mz_uint64 m_local_header_ofs;
+  mz_uint32 m_comment_size;
+  char m_filename[MZ_ZIP_MAX_ARCHIVE_FILENAME_SIZE];
+  char m_comment[MZ_ZIP_MAX_ARCHIVE_FILE_COMMENT_SIZE];
+} mz_zip_archive_file_stat;
+
+typedef size_t (*mz_file_read_func)(void *pOpaque, mz_uint64 file_ofs, void *pBuf, size_t n);
+typedef size_t (*mz_file_write_func)(void *pOpaque, mz_uint64 file_ofs, const void *pBuf, size_t n);
+
+struct mz_zip_internal_state_tag;
+typedef struct mz_zip_internal_state_tag mz_zip_internal_state;
+
+typedef enum
+{
+  MZ_ZIP_MODE_INVALID = 0,
+  MZ_ZIP_MODE_READING = 1,
+  MZ_ZIP_MODE_WRITING = 2,
+  MZ_ZIP_MODE_WRITING_HAS_BEEN_FINALIZED = 3
+} mz_zip_mode;
+
+typedef struct mz_zip_archive_tag
+{
+  mz_uint64 m_archive_size;
+  mz_uint64 m_central_directory_file_ofs;
+  mz_uint m_total_files;
+  mz_zip_mode m_zip_mode;
+
+  mz_uint m_file_offset_alignment;
+
+  mz_alloc_func m_pAlloc;
+  mz_free_func m_pFree;
+  mz_realloc_func m_pRealloc;
+  void *m_pAlloc_opaque;
+
+  mz_file_read_func m_pRead;
+  mz_file_write_func m_pWrite;
+  void *m_pIO_opaque;
+
+  mz_zip_internal_state *m_pState;
+
+} mz_zip_archive;
+
+typedef enum
+{
+  MZ_ZIP_FLAG_CASE_SENSITIVE                = 0x0100,
+  MZ_ZIP_FLAG_IGNORE_PATH                   = 0x0200,
+  MZ_ZIP_FLAG_COMPRESSED_DATA               = 0x0400,
+  MZ_ZIP_FLAG_DO_NOT_SORT_CENTRAL_DIRECTORY = 0x0800
+} mz_zip_flags;
+
+// ZIP archive reading
+
+// Inits a ZIP archive reader.
+// These functions read and validate the archive's central directory.
+mz_bool mz_zip_reader_init(mz_zip_archive *pZip, mz_uint64 size, mz_uint32 flags);
+mz_bool mz_zip_reader_init_mem(mz_zip_archive *pZip, const void *pMem, size_t size, mz_uint32 flags);
+
+#ifndef MINIZ_NO_STDIO
+mz_bool mz_zip_reader_init_file(mz_zip_archive *pZip, const char *pFilename, mz_uint32 flags);
+#endif
+
+// Returns the total number of files in the archive.
+mz_uint mz_zip_reader_get_num_files(mz_zip_archive *pZip);
+
+// Returns detailed information about an archive file entry.
+mz_bool mz_zip_reader_file_stat(mz_zip_archive *pZip, mz_uint file_index, mz_zip_archive_file_stat *pStat);
+
+// Determines if an archive file entry is a directory entry.
+mz_bool mz_zip_reader_is_file_a_directory(mz_zip_archive *pZip, mz_uint file_index);
+mz_bool mz_zip_reader_is_file_encrypted(mz_zip_archive *pZip, mz_uint file_index);
+
+// Retrieves the filename of an archive file entry.
+// Returns the number of bytes written to pFilename, or if filename_buf_size is 0 this function returns the number of bytes needed to fully store the filename.
+mz_uint mz_zip_reader_get_filename(mz_zip_archive *pZip, mz_uint file_index, char *pFilename, mz_uint filename_buf_size);
+
+// Attempts to locates a file in the archive's central directory.
+// Valid flags: MZ_ZIP_FLAG_CASE_SENSITIVE, MZ_ZIP_FLAG_IGNORE_PATH
+// Returns -1 if the file cannot be found.
+int mz_zip_reader_locate_file(mz_zip_archive *pZip, const char *pName, const char *pComment, mz_uint flags);
+
+// Extracts a archive file to a memory buffer using no memory allocation.
+mz_bool mz_zip_reader_extract_to_mem_no_alloc(mz_zip_archive *pZip, mz_uint file_index, void *pBuf, size_t buf_size, mz_uint flags, void *pUser_read_buf, size_t user_read_buf_size);
+mz_bool mz_zip_reader_extract_file_to_mem_no_alloc(mz_zip_archive *pZip, const char *pFilename, void *pBuf, size_t buf_size, mz_uint flags, void *pUser_read_buf, size_t user_read_buf_size);
+
+// Extracts a archive file to a memory buffer.
+mz_bool mz_zip_reader_extract_to_mem(mz_zip_archive *pZip, mz_uint file_index, void *pBuf, size_t buf_size, mz_uint flags);
+mz_bool mz_zip_reader_extract_file_to_mem(mz_zip_archive *pZip, const char *pFilename, void *pBuf, size_t buf_size, mz_uint flags);
+
+// Extracts a archive file to a dynamically allocated heap buffer.
+void *mz_zip_reader_extract_to_heap(mz_zip_archive *pZip, mz_uint file_index, size_t *pSize, mz_uint flags);
+void *mz_zip_reader_extract_file_to_heap(mz_zip_archive *pZip, const char *pFilename, size_t *pSize, mz_uint flags);
+
+// Extracts a archive file using a callback function to output the file's data.
+mz_bool mz_zip_reader_extract_to_callback(mz_zip_archive *pZip, mz_uint file_index, mz_file_write_func pCallback, void *pOpaque, mz_uint flags);
+mz_bool mz_zip_reader_extract_file_to_callback(mz_zip_archive *pZip, const char *pFilename, mz_file_write_func pCallback, void *pOpaque, mz_uint flags);
+
+#ifndef MINIZ_NO_STDIO
+// Extracts a archive file to a disk file and sets its last accessed and modified times.
+// This function only extracts files, not archive directory records.
+mz_bool mz_zip_reader_extract_to_file(mz_zip_archive *pZip, mz_uint file_index, const char *pDst_filename, mz_uint flags);
+mz_bool mz_zip_reader_extract_file_to_file(mz_zip_archive *pZip, const char *pArchive_filename, const char *pDst_filename, mz_uint flags);
+#endif
+
+// Ends archive reading, freeing all allocations, and closing the input archive file if mz_zip_reader_init_file() was used.
+mz_bool mz_zip_reader_end(mz_zip_archive *pZip);
+
+// ZIP archive writing
+
+#ifndef MINIZ_NO_ARCHIVE_WRITING_APIS
+
+// Inits a ZIP archive writer.
+mz_bool mz_zip_writer_init(mz_zip_archive *pZip, mz_uint64 existing_size);
+mz_bool mz_zip_writer_init_heap(mz_zip_archive *pZip, size_t size_to_reserve_at_beginning, size_t initial_allocation_size);
+
+#ifndef MINIZ_NO_STDIO
+mz_bool mz_zip_writer_init_file(mz_zip_archive *pZip, const char *pFilename, mz_uint64 size_to_reserve_at_beginning);
+#endif
+
+// Converts a ZIP archive reader object into a writer object, to allow efficient in-place file appends to occur on an existing archive.
+// For archives opened using mz_zip_reader_init_file, pFilename must be the archive's filename so it can be reopened for writing. If the file can't be reopened, mz_zip_reader_end() will be called.
+// For archives opened using mz_zip_reader_init_mem, the memory block must be growable using the realloc callback (which defaults to realloc unless you've overridden it).
+// Finally, for archives opened using mz_zip_reader_init, the mz_zip_archive's user provided m_pWrite function cannot be NULL.
+// Note: In-place archive modification is not recommended unless you know what you're doing, because if execution stops or something goes wrong before
+// the archive is finalized the file's central directory will be hosed.
+mz_bool mz_zip_writer_init_from_reader(mz_zip_archive *pZip, const char *pFilename);
+
+// Adds the contents of a memory buffer to an archive. These functions record the current local time into the archive.
+// To add a directory entry, call this method with an archive name ending in a forwardslash with empty buffer.
+// level_and_flags - compression level (0-10, see MZ_BEST_SPEED, MZ_BEST_COMPRESSION, etc.) logically OR'd with zero or more mz_zip_flags, or just set to MZ_DEFAULT_COMPRESSION.
+mz_bool mz_zip_writer_add_mem(mz_zip_archive *pZip, const char *pArchive_name, const void *pBuf, size_t buf_size, mz_uint level_and_flags);
+mz_bool mz_zip_writer_add_mem_ex(mz_zip_archive *pZip, const char *pArchive_name, const void *pBuf, size_t buf_size, const void *pComment, mz_uint16 comment_size, mz_uint level_and_flags, mz_uint64 uncomp_size, mz_uint32 uncomp_crc32);
+
+#ifndef MINIZ_NO_STDIO
+// Adds the contents of a disk file to an archive. This function also records the disk file's modified time into the archive.
+// level_and_flags - compression level (0-10, see MZ_BEST_SPEED, MZ_BEST_COMPRESSION, etc.) logically OR'd with zero or more mz_zip_flags, or just set to MZ_DEFAULT_COMPRESSION.
+mz_bool mz_zip_writer_add_file(mz_zip_archive *pZip, const char *pArchive_name, const char *pSrc_filename, const void *pComment, mz_uint16 comment_size, mz_uint level_and_flags);
+#endif
+
+// Adds a file to an archive by fully cloning the data from another archive.
+// This function fully clones the source file's compressed data (no recompression), along with its full filename, extra data, and comment fields.
+mz_bool mz_zip_writer_add_from_zip_reader(mz_zip_archive *pZip, mz_zip_archive *pSource_zip, mz_uint file_index);
+
+// Finalizes the archive by writing the central directory records followed by the end of central directory record.
+// After an archive is finalized, the only valid call on the mz_zip_archive struct is mz_zip_writer_end().
+// An archive must be manually finalized by calling this function for it to be valid.
+mz_bool mz_zip_writer_finalize_archive(mz_zip_archive *pZip);
+mz_bool mz_zip_writer_finalize_heap_archive(mz_zip_archive *pZip, void **pBuf, size_t *pSize);
+
+// Ends archive writing, freeing all allocations, and closing the output file if mz_zip_writer_init_file() was used.
+// Note for the archive to be valid, it must have been finalized before ending.
+mz_bool mz_zip_writer_end(mz_zip_archive *pZip);
+
+// Misc. high-level helper functions:
+
+// mz_zip_add_mem_to_archive_file_in_place() efficiently (but not atomically) appends a memory blob to a ZIP archive.
+// level_and_flags - compression level (0-10, see MZ_BEST_SPEED, MZ_BEST_COMPRESSION, etc.) logically OR'd with zero or more mz_zip_flags, or just set to MZ_DEFAULT_COMPRESSION.
+mz_bool mz_zip_add_mem_to_archive_file_in_place(const char *pZip_filename, const char *pArchive_name, const void *pBuf, size_t buf_size, const void *pComment, mz_uint16 comment_size, mz_uint level_and_flags);
+
+// Reads a single file from an archive into a heap block.
+// Returns NULL on failure.
+void *mz_zip_extract_archive_file_to_heap(const char *pZip_filename, const char *pArchive_name, size_t *pSize, mz_uint zip_flags);
+
+#endif // #ifndef MINIZ_NO_ARCHIVE_WRITING_APIS
+
+#endif // #ifndef MINIZ_NO_ARCHIVE_APIS
+
+// ------------------- Low-level Decompression API Definitions
+
+// Decompression flags used by tinfl_decompress().
+// TINFL_FLAG_PARSE_ZLIB_HEADER: If set, the input has a valid zlib header and ends with an adler32 checksum (it's a valid zlib stream). Otherwise, the input is a raw deflate stream.
+// TINFL_FLAG_HAS_MORE_INPUT: If set, there are more input bytes available beyond the end of the supplied input buffer. If clear, the input buffer contains all remaining input.
+// TINFL_FLAG_USING_NON_WRAPPING_OUTPUT_BUF: If set, the output buffer is large enough to hold the entire decompressed stream. If clear, the output buffer is at least the size of the dictionary (typically 32KB).
+// TINFL_FLAG_COMPUTE_ADLER32: Force adler-32 checksum computation of the decompressed bytes.
+enum
+{
+  TINFL_FLAG_PARSE_ZLIB_HEADER = 1,
+  TINFL_FLAG_HAS_MORE_INPUT = 2,
+  TINFL_FLAG_USING_NON_WRAPPING_OUTPUT_BUF = 4,
+  TINFL_FLAG_COMPUTE_ADLER32 = 8
+};
+
+// High level decompression functions:
+// tinfl_decompress_mem_to_heap() decompresses a block in memory to a heap block allocated via malloc().
+// On entry:
+//  pSrc_buf, src_buf_len: Pointer and size of the Deflate or zlib source data to decompress.
+// On return:
+//  Function returns a pointer to the decompressed data, or NULL on failure.
+//  *pOut_len will be set to the decompressed data's size, which could be larger than src_buf_len on uncompressible data.
+//  The caller must call mz_free() on the returned block when it's no longer needed.
+void *tinfl_decompress_mem_to_heap(const void *pSrc_buf, size_t src_buf_len, size_t *pOut_len, int flags);
+
+// tinfl_decompress_mem_to_mem() decompresses a block in memory to another block in memory.
+// Returns TINFL_DECOMPRESS_MEM_TO_MEM_FAILED on failure, or the number of bytes written on success.
+#define TINFL_DECOMPRESS_MEM_TO_MEM_FAILED ((size_t)(-1))
+size_t tinfl_decompress_mem_to_mem(void *pOut_buf, size_t out_buf_len, const void *pSrc_buf, size_t src_buf_len, int flags);
+
+// tinfl_decompress_mem_to_callback() decompresses a block in memory to an internal 32KB buffer, and a user provided callback function will be called to flush the buffer.
+// Returns 1 on success or 0 on failure.
+typedef int (*tinfl_put_buf_func_ptr)(const void* pBuf, int len, void *pUser);
+int tinfl_decompress_mem_to_callback(const void *pIn_buf, size_t *pIn_buf_size, tinfl_put_buf_func_ptr pPut_buf_func, void *pPut_buf_user, int flags);
+
+struct tinfl_decompressor_tag; typedef struct tinfl_decompressor_tag tinfl_decompressor;
+
+// Max size of LZ dictionary.
+#define TINFL_LZ_DICT_SIZE 32768
+
+// Return status.
+typedef enum
+{
+  TINFL_STATUS_BAD_PARAM = -3,
+  TINFL_STATUS_ADLER32_MISMATCH = -2,
+  TINFL_STATUS_FAILED = -1,
+  TINFL_STATUS_DONE = 0,
+  TINFL_STATUS_NEEDS_MORE_INPUT = 1,
+  TINFL_STATUS_HAS_MORE_OUTPUT = 2
+} tinfl_status;
+
+// Initializes the decompressor to its initial state.
+#define tinfl_init(r) do { (r)->m_state = 0; } MZ_MACRO_END
+#define tinfl_get_adler32(r) (r)->m_check_adler32
+
+// Main low-level decompressor coroutine function. This is the only function actually needed for decompression. All the other functions are just high-level helpers for improved usability.
+// This is a universal API, i.e. it can be used as a building block to build any desired higher level decompression API. In the limit case, it can be called once per every byte input or output.
+tinfl_status tinfl_decompress(tinfl_decompressor *r, const mz_uint8 *pIn_buf_next, size_t *pIn_buf_size, mz_uint8 *pOut_buf_start, mz_uint8 *pOut_buf_next, size_t *pOut_buf_size, const mz_uint32 decomp_flags);
+
+// Internal/private bits follow.
+enum
+{
+  TINFL_MAX_HUFF_TABLES = 3, TINFL_MAX_HUFF_SYMBOLS_0 = 288, TINFL_MAX_HUFF_SYMBOLS_1 = 32, TINFL_MAX_HUFF_SYMBOLS_2 = 19,
+  TINFL_FAST_LOOKUP_BITS = 10, TINFL_FAST_LOOKUP_SIZE = 1 << TINFL_FAST_LOOKUP_BITS
+};
+
+typedef struct
+{
+  mz_uint8 m_code_size[TINFL_MAX_HUFF_SYMBOLS_0];
+  mz_int16 m_look_up[TINFL_FAST_LOOKUP_SIZE], m_tree[TINFL_MAX_HUFF_SYMBOLS_0 * 2];
+} tinfl_huff_table;
+
+#if MINIZ_HAS_64BIT_REGISTERS
+  #define TINFL_USE_64BIT_BITBUF 1
+#endif
+
+#if TINFL_USE_64BIT_BITBUF
+  typedef mz_uint64 tinfl_bit_buf_t;
+  #define TINFL_BITBUF_SIZE (64)
+#else
+  typedef mz_uint32 tinfl_bit_buf_t;
+  #define TINFL_BITBUF_SIZE (32)
+#endif
+
+struct tinfl_decompressor_tag
+{
+  mz_uint32 m_state, m_num_bits, m_zhdr0, m_zhdr1, m_z_adler32, m_final, m_type, m_check_adler32, m_dist, m_counter, m_num_extra, m_table_sizes[TINFL_MAX_HUFF_TABLES];
+  tinfl_bit_buf_t m_bit_buf;
+  size_t m_dist_from_out_buf_start;
+  tinfl_huff_table m_tables[TINFL_MAX_HUFF_TABLES];
+  mz_uint8 m_raw_header[4], m_len_codes[TINFL_MAX_HUFF_SYMBOLS_0 + TINFL_MAX_HUFF_SYMBOLS_1 + 137];
+};
+
+// ------------------- Low-level Compression API Definitions
+
+// Set TDEFL_LESS_MEMORY to 1 to use less memory (compression will be slightly slower, and raw/dynamic blocks will be output more frequently).
+#define TDEFL_LESS_MEMORY 0
+
+// tdefl_init() compression flags logically OR'd together (low 12 bits contain the max. number of probes per dictionary search):
+// TDEFL_DEFAULT_MAX_PROBES: The compressor defaults to 128 dictionary probes per dictionary search. 0=Huffman only, 1=Huffman+LZ (fastest/crap compression), 4095=Huffman+LZ (slowest/best compression).
+enum
+{
+  TDEFL_HUFFMAN_ONLY = 0, TDEFL_DEFAULT_MAX_PROBES = 128, TDEFL_MAX_PROBES_MASK = 0xFFF
+};
+
+// TDEFL_WRITE_ZLIB_HEADER: If set, the compressor outputs a zlib header before the deflate data, and the Adler-32 of the source data at the end. Otherwise, you'll get raw deflate data.
+// TDEFL_COMPUTE_ADLER32: Always compute the adler-32 of the input data (even when not writing zlib headers).
+// TDEFL_GREEDY_PARSING_FLAG: Set to use faster greedy parsing, instead of more efficient lazy parsing.
+// TDEFL_NONDETERMINISTIC_PARSING_FLAG: Enable to decrease the compressor's initialization time to the minimum, but the output may vary from run to run given the same input (depending on the contents of memory).
+// TDEFL_RLE_MATCHES: Only look for RLE matches (matches with a distance of 1)
+// TDEFL_FILTER_MATCHES: Discards matches <= 5 chars if enabled.
+// TDEFL_FORCE_ALL_STATIC_BLOCKS: Disable usage of optimized Huffman tables.
+// TDEFL_FORCE_ALL_RAW_BLOCKS: Only use raw (uncompressed) deflate blocks.
+// The low 12 bits are reserved to control the max # of hash probes per dictionary lookup (see TDEFL_MAX_PROBES_MASK).
+enum
+{
+  TDEFL_WRITE_ZLIB_HEADER             = 0x01000,
+  TDEFL_COMPUTE_ADLER32               = 0x02000,
+  TDEFL_GREEDY_PARSING_FLAG           = 0x04000,
+  TDEFL_NONDETERMINISTIC_PARSING_FLAG = 0x08000,
+  TDEFL_RLE_MATCHES                   = 0x10000,
+  TDEFL_FILTER_MATCHES                = 0x20000,
+  TDEFL_FORCE_ALL_STATIC_BLOCKS       = 0x40000,
+  TDEFL_FORCE_ALL_RAW_BLOCKS          = 0x80000
+};
+
+// High level compression functions:
+// tdefl_compress_mem_to_heap() compresses a block in memory to a heap block allocated via malloc().
+// On entry:
+//  pSrc_buf, src_buf_len: Pointer and size of source block to compress.
+//  flags: The max match finder probes (default is 128) logically OR'd against the above flags. Higher probes are slower but improve compression.
+// On return:
+//  Function returns a pointer to the compressed data, or NULL on failure.
+//  *pOut_len will be set to the compressed data's size, which could be larger than src_buf_len on uncompressible data.
+//  The caller must free() the returned block when it's no longer needed.
+void *tdefl_compress_mem_to_heap(const void *pSrc_buf, size_t src_buf_len, size_t *pOut_len, int flags);
+
+// tdefl_compress_mem_to_mem() compresses a block in memory to another block in memory.
+// Returns 0 on failure.
+size_t tdefl_compress_mem_to_mem(void *pOut_buf, size_t out_buf_len, const void *pSrc_buf, size_t src_buf_len, int flags);
+
+// Compresses an image to a compressed PNG file in memory.
+// On entry:
+//  pImage, w, h, and num_chans describe the image to compress. num_chans may be 1, 2, 3, or 4. 
+//  The image pitch in bytes per scanline will be w*num_chans. The leftmost pixel on the top scanline is stored first in memory.
+//  level may range from [0,10], use MZ_NO_COMPRESSION, MZ_BEST_SPEED, MZ_BEST_COMPRESSION, etc. or a decent default is MZ_DEFAULT_LEVEL
+//  If flip is true, the image will be flipped on the Y axis (useful for OpenGL apps).
+// On return:
+//  Function returns a pointer to the compressed data, or NULL on failure.
+//  *pLen_out will be set to the size of the PNG image file.
+//  The caller must mz_free() the returned heap block (which will typically be larger than *pLen_out) when it's no longer needed.
+void *tdefl_write_image_to_png_file_in_memory_ex(const void *pImage, int w, int h, int num_chans, size_t *pLen_out, mz_uint level, mz_bool flip);
+void *tdefl_write_image_to_png_file_in_memory(const void *pImage, int w, int h, int num_chans, size_t *pLen_out);
+
+// Output stream interface. The compressor uses this interface to write compressed data. It'll typically be called TDEFL_OUT_BUF_SIZE at a time.
+typedef mz_bool (*tdefl_put_buf_func_ptr)(const void* pBuf, int len, void *pUser);
+
+// tdefl_compress_mem_to_output() compresses a block to an output stream. The above helpers use this function internally.
+mz_bool tdefl_compress_mem_to_output(const void *pBuf, size_t buf_len, tdefl_put_buf_func_ptr pPut_buf_func, void *pPut_buf_user, int flags);
+
+enum { TDEFL_MAX_HUFF_TABLES = 3, TDEFL_MAX_HUFF_SYMBOLS_0 = 288, TDEFL_MAX_HUFF_SYMBOLS_1 = 32, TDEFL_MAX_HUFF_SYMBOLS_2 = 19, TDEFL_LZ_DICT_SIZE = 32768, TDEFL_LZ_DICT_SIZE_MASK = TDEFL_LZ_DICT_SIZE - 1, TDEFL_MIN_MATCH_LEN = 3, TDEFL_MAX_MATCH_LEN = 258 };
+
+// TDEFL_OUT_BUF_SIZE MUST be large enough to hold a single entire compressed output block (using static/fixed Huffman codes).
+#if TDEFL_LESS_MEMORY
+enum { TDEFL_LZ_CODE_BUF_SIZE = 24 * 1024, TDEFL_OUT_BUF_SIZE = (TDEFL_LZ_CODE_BUF_SIZE * 13 ) / 10, TDEFL_MAX_HUFF_SYMBOLS = 288, TDEFL_LZ_HASH_BITS = 12, TDEFL_LEVEL1_HASH_SIZE_MASK = 4095, TDEFL_LZ_HASH_SHIFT = (TDEFL_LZ_HASH_BITS + 2) / 3, TDEFL_LZ_HASH_SIZE = 1 << TDEFL_LZ_HASH_BITS };
+#else
+enum { TDEFL_LZ_CODE_BUF_SIZE = 64 * 1024, TDEFL_OUT_BUF_SIZE = (TDEFL_LZ_CODE_BUF_SIZE * 13 ) / 10, TDEFL_MAX_HUFF_SYMBOLS = 288, TDEFL_LZ_HASH_BITS = 15, TDEFL_LEVEL1_HASH_SIZE_MASK = 4095, TDEFL_LZ_HASH_SHIFT = (TDEFL_LZ_HASH_BITS + 2) / 3, TDEFL_LZ_HASH_SIZE = 1 << TDEFL_LZ_HASH_BITS };
+#endif
+
+// The low-level tdefl functions below may be used directly if the above helper functions aren't flexible enough. The low-level functions don't make any heap allocations, unlike the above helper functions.
+typedef enum
+{
+  TDEFL_STATUS_BAD_PARAM = -2,
+  TDEFL_STATUS_PUT_BUF_FAILED = -1,
+  TDEFL_STATUS_OKAY = 0,
+  TDEFL_STATUS_DONE = 1,
+} tdefl_status;
+
+// Must map to MZ_NO_FLUSH, MZ_SYNC_FLUSH, etc. enums
+typedef enum
+{
+  TDEFL_NO_FLUSH = 0,
+  TDEFL_SYNC_FLUSH = 2,
+  TDEFL_FULL_FLUSH = 3,
+  TDEFL_FINISH = 4
+} tdefl_flush;
+
+// tdefl's compression state structure.
+typedef struct
+{
+  tdefl_put_buf_func_ptr m_pPut_buf_func;
+  void *m_pPut_buf_user;
+  mz_uint m_flags, m_max_probes[2];
+  int m_greedy_parsing;
+  mz_uint m_adler32, m_lookahead_pos, m_lookahead_size, m_dict_size;
+  mz_uint8 *m_pLZ_code_buf, *m_pLZ_flags, *m_pOutput_buf, *m_pOutput_buf_end;
+  mz_uint m_num_flags_left, m_total_lz_bytes, m_lz_code_buf_dict_pos, m_bits_in, m_bit_buffer;
+  mz_uint m_saved_match_dist, m_saved_match_len, m_saved_lit, m_output_flush_ofs, m_output_flush_remaining, m_finished, m_block_index, m_wants_to_finish;
+  tdefl_status m_prev_return_status;
+  const void *m_pIn_buf;
+  void *m_pOut_buf;
+  size_t *m_pIn_buf_size, *m_pOut_buf_size;
+  tdefl_flush m_flush;
+  const mz_uint8 *m_pSrc;
+  size_t m_src_buf_left, m_out_buf_ofs;
+  mz_uint8 m_dict[TDEFL_LZ_DICT_SIZE + TDEFL_MAX_MATCH_LEN - 1];
+  mz_uint16 m_huff_count[TDEFL_MAX_HUFF_TABLES][TDEFL_MAX_HUFF_SYMBOLS];
+  mz_uint16 m_huff_codes[TDEFL_MAX_HUFF_TABLES][TDEFL_MAX_HUFF_SYMBOLS];
+  mz_uint8 m_huff_code_sizes[TDEFL_MAX_HUFF_TABLES][TDEFL_MAX_HUFF_SYMBOLS];
+  mz_uint8 m_lz_code_buf[TDEFL_LZ_CODE_BUF_SIZE];
+  mz_uint16 m_next[TDEFL_LZ_DICT_SIZE];
+  mz_uint16 m_hash[TDEFL_LZ_HASH_SIZE];
+  mz_uint8 m_output_buf[TDEFL_OUT_BUF_SIZE];
+} tdefl_compressor;
+
+// Initializes the compressor.
+// There is no corresponding deinit() function because the tdefl API's do not dynamically allocate memory.
+// pBut_buf_func: If NULL, output data will be supplied to the specified callback. In this case, the user should call the tdefl_compress_buffer() API for compression.
+// If pBut_buf_func is NULL the user should always call the tdefl_compress() API.
+// flags: See the above enums (TDEFL_HUFFMAN_ONLY, TDEFL_WRITE_ZLIB_HEADER, etc.)
+tdefl_status tdefl_init(tdefl_compressor *d, tdefl_put_buf_func_ptr pPut_buf_func, void *pPut_buf_user, int flags);
+
+// Compresses a block of data, consuming as much of the specified input buffer as possible, and writing as much compressed data to the specified output buffer as possible.
+tdefl_status tdefl_compress(tdefl_compressor *d, const void *pIn_buf, size_t *pIn_buf_size, void *pOut_buf, size_t *pOut_buf_size, tdefl_flush flush);
+
+// tdefl_compress_buffer() is only usable when the tdefl_init() is called with a non-NULL tdefl_put_buf_func_ptr.
+// tdefl_compress_buffer() always consumes the entire input buffer.
+tdefl_status tdefl_compress_buffer(tdefl_compressor *d, const void *pIn_buf, size_t in_buf_size, tdefl_flush flush);
+
+tdefl_status tdefl_get_prev_return_status(tdefl_compressor *d);
+mz_uint32 tdefl_get_adler32(tdefl_compressor *d);
+
+// Can't use tdefl_create_comp_flags_from_zip_params if MINIZ_NO_ZLIB_APIS isn't defined, because it uses some of its macros.
+#ifndef MINIZ_NO_ZLIB_APIS
+// Create tdefl_compress() flags given zlib-style compression parameters.
+// level may range from [0,10] (where 10 is absolute max compression, but may be much slower on some files)
+// window_bits may be -15 (raw deflate) or 15 (zlib)
+// strategy may be either MZ_DEFAULT_STRATEGY, MZ_FILTERED, MZ_HUFFMAN_ONLY, MZ_RLE, or MZ_FIXED
+mz_uint tdefl_create_comp_flags_from_zip_params(int level, int window_bits, int strategy);
+#endif // #ifndef MINIZ_NO_ZLIB_APIS
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // MINIZ_HEADER_INCLUDED
+
+// ------------------- End of Header: Implementation follows. (If you only want the header, define MINIZ_HEADER_FILE_ONLY.)
+
+#ifndef MINIZ_HEADER_FILE_ONLY
+
+typedef unsigned char mz_validate_uint16[sizeof(mz_uint16)==2 ? 1 : -1];
+typedef unsigned char mz_validate_uint32[sizeof(mz_uint32)==4 ? 1 : -1];
+typedef unsigned char mz_validate_uint64[sizeof(mz_uint64)==8 ? 1 : -1];
+
+#include <string.h>
+#include <assert.h>
+
+#define MZ_ASSERT(x) assert(x)
+
+#ifdef MINIZ_NO_MALLOC
+  #define MZ_MALLOC(x) NULL
+  #define MZ_FREE(x) (void)x, ((void)0)
+  #define MZ_REALLOC(p, x) NULL
+#else
+  #define MZ_MALLOC(x) malloc(x)
+  #define MZ_FREE(x) free(x)
+  #define MZ_REALLOC(p, x) realloc(p, x)
+#endif
+
+#define MZ_MAX(a,b) (((a)>(b))?(a):(b))
+#define MZ_MIN(a,b) (((a)<(b))?(a):(b))
+#define MZ_CLEAR_OBJ(obj) memset(&(obj), 0, sizeof(obj))
+
+#if MINIZ_USE_UNALIGNED_LOADS_AND_STORES && MINIZ_LITTLE_ENDIAN
+  #define MZ_READ_LE16(p) *((const mz_uint16 *)(p))
+  #define MZ_READ_LE32(p) *((const mz_uint32 *)(p))
+#else
+  #define MZ_READ_LE16(p) ((mz_uint32)(((const mz_uint8 *)(p))[0]) | ((mz_uint32)(((const mz_uint8 *)(p))[1]) << 8U))
+  #define MZ_READ_LE32(p) ((mz_uint32)(((const mz_uint8 *)(p))[0]) | ((mz_uint32)(((const mz_uint8 *)(p))[1]) << 8U) | ((mz_uint32)(((const mz_uint8 *)(p))[2]) << 16U) | ((mz_uint32)(((const mz_uint8 *)(p))[3]) << 24U))
+#endif
+
+#ifdef _MSC_VER
+  #define MZ_FORCEINLINE __forceinline
+#elif defined(__GNUC__)
+  #define MZ_FORCEINLINE inline __attribute__((__always_inline__))
+#else
+  #define MZ_FORCEINLINE inline
+#endif
+
+#ifdef __cplusplus
+  extern "C" {
+#endif
+
+// ------------------- zlib-style API's
+
+mz_ulong mz_adler32(mz_ulong adler, const unsigned char *ptr, size_t buf_len)
+{
+  mz_uint32 i, s1 = (mz_uint32)(adler & 0xffff), s2 = (mz_uint32)(adler >> 16); size_t block_len = buf_len % 5552;
+  if (!ptr) return MZ_ADLER32_INIT;
+  while (buf_len) {
+    for (i = 0; i + 7 < block_len; i += 8, ptr += 8) {
+      s1 += ptr[0], s2 += s1; s1 += ptr[1], s2 += s1; s1 += ptr[2], s2 += s1; s1 += ptr[3], s2 += s1;
+      s1 += ptr[4], s2 += s1; s1 += ptr[5], s2 += s1; s1 += ptr[6], s2 += s1; s1 += ptr[7], s2 += s1;
+    }
+    for ( ; i < block_len; ++i) s1 += *ptr++, s2 += s1;
+    s1 %= 65521U, s2 %= 65521U; buf_len -= block_len; block_len = 5552;
+  }
+  return (s2 << 16) + s1;
+}
+
+// Karl Malbrain's compact CRC-32. See "A compact CCITT crc16 and crc32 C implementation that balances processor cache usage against speed": http://www.geocities.com/malbrain/
+mz_ulong mz_crc32(mz_ulong crc, const mz_uint8 *ptr, size_t buf_len)
+{
+  static const mz_uint32 s_crc32[16] = { 0, 0x1db71064, 0x3b6e20c8, 0x26d930ac, 0x76dc4190, 0x6b6b51f4, 0x4db26158, 0x5005713c,
+    0xedb88320, 0xf00f9344, 0xd6d6a3e8, 0xcb61b38c, 0x9b64c2b0, 0x86d3d2d4, 0xa00ae278, 0xbdbdf21c };
+  mz_uint32 crcu32 = (mz_uint32)crc;
+  if (!ptr) return MZ_CRC32_INIT;
+  crcu32 = ~crcu32; while (buf_len--) { mz_uint8 b = *ptr++; crcu32 = (crcu32 >> 4) ^ s_crc32[(crcu32 & 0xF) ^ (b & 0xF)]; crcu32 = (crcu32 >> 4) ^ s_crc32[(crcu32 & 0xF) ^ (b >> 4)]; }
+  return ~crcu32;
+}
+
+void mz_free(void *p)
+{
+  MZ_FREE(p);
+}
+
+#ifndef MINIZ_NO_ZLIB_APIS
+
+static void *def_alloc_func(void *opaque, size_t items, size_t size) { (void)opaque, (void)items, (void)size; return MZ_MALLOC(items * size); }
+static void def_free_func(void *opaque, void *address) { (void)opaque, (void)address; MZ_FREE(address); }
+static void *def_realloc_func(void *opaque, void *address, size_t items, size_t size) { (void)opaque, (void)address, (void)items, (void)size; return MZ_REALLOC(address, items * size); }
+
+const char *mz_version(void)
+{
+  return MZ_VERSION;
+}
+
+int mz_deflateInit(mz_streamp pStream, int level)
+{
+  return mz_deflateInit2(pStream, level, MZ_DEFLATED, MZ_DEFAULT_WINDOW_BITS, 9, MZ_DEFAULT_STRATEGY);
+}
+
+int mz_deflateInit2(mz_streamp pStream, int level, int method, int window_bits, int mem_level, int strategy)
+{
+  tdefl_compressor *pComp;
+  mz_uint comp_flags = TDEFL_COMPUTE_ADLER32 | tdefl_create_comp_flags_from_zip_params(level, window_bits, strategy);
+
+  if (!pStream) return MZ_STREAM_ERROR;
+  if ((method != MZ_DEFLATED) || ((mem_level < 1) || (mem_level > 9)) || ((window_bits != MZ_DEFAULT_WINDOW_BITS) && (-window_bits != MZ_DEFAULT_WINDOW_BITS))) return MZ_PARAM_ERROR;
+
+  pStream->data_type = 0;
+  pStream->adler = MZ_ADLER32_INIT;
+  pStream->msg = NULL;
+  pStream->reserved = 0;
+  pStream->total_in = 0;
+  pStream->total_out = 0;
+  if (!pStream->zalloc) pStream->zalloc = def_alloc_func;
+  if (!pStream->zfree) pStream->zfree = def_free_func;
+
+  pComp = (tdefl_compressor *)pStream->zalloc(pStream->opaque, 1, sizeof(tdefl_compressor));
+  if (!pComp)
+    return MZ_MEM_ERROR;
+
+  pStream->state = (struct mz_internal_state *)pComp;
+
+  if (tdefl_init(pComp, NULL, NULL, comp_flags) != TDEFL_STATUS_OKAY)
+  {
+    mz_deflateEnd(pStream);
+    return MZ_PARAM_ERROR;
+  }
+
+  return MZ_OK;
+}
+
+int mz_deflateReset(mz_streamp pStream)
+{
+  if ((!pStream) || (!pStream->state) || (!pStream->zalloc) || (!pStream->zfree)) return MZ_STREAM_ERROR;
+  pStream->total_in = pStream->total_out = 0;
+  tdefl_init((tdefl_compressor*)pStream->state, NULL, NULL, ((tdefl_compressor*)pStream->state)->m_flags);
+  return MZ_OK;
+}
+
+int mz_deflate(mz_streamp pStream, int flush)
+{
+  size_t in_bytes, out_bytes;
+  mz_ulong orig_total_in, orig_total_out;
+  int mz_status = MZ_OK;
+
+  if ((!pStream) || (!pStream->state) || (flush < 0) || (flush > MZ_FINISH) || (!pStream->next_out)) return MZ_STREAM_ERROR;
+  if (!pStream->avail_out) return MZ_BUF_ERROR;
+
+  if (flush == MZ_PARTIAL_FLUSH) flush = MZ_SYNC_FLUSH;
+
+  if (((tdefl_compressor*)pStream->state)->m_prev_return_status == TDEFL_STATUS_DONE)
+    return (flush == MZ_FINISH) ? MZ_STREAM_END : MZ_BUF_ERROR;
+
+  orig_total_in = pStream->total_in; orig_total_out = pStream->total_out;
+  for ( ; ; )
+  {
+    tdefl_status defl_status;
+    in_bytes = pStream->avail_in; out_bytes = pStream->avail_out;
+
+    defl_status = tdefl_compress((tdefl_compressor*)pStream->state, pStream->next_in, &in_bytes, pStream->next_out, &out_bytes, (tdefl_flush)flush);
+    pStream->next_in += (mz_uint)in_bytes; pStream->avail_in -= (mz_uint)in_bytes;
+    pStream->total_in += (mz_uint)in_bytes; pStream->adler = tdefl_get_adler32((tdefl_compressor*)pStream->state);
+
+    pStream->next_out += (mz_uint)out_bytes; pStream->avail_out -= (mz_uint)out_bytes;
+    pStream->total_out += (mz_uint)out_bytes;
+
+    if (defl_status < 0)
+    {
+      mz_status = MZ_STREAM_ERROR;
+      break;
+    }
+    else if (defl_status == TDEFL_STATUS_DONE)
+    {
+      mz_status = MZ_STREAM_END;
+      break;
+    }
+    else if (!pStream->avail_out)
+      break;
+    else if ((!pStream->avail_in) && (flush != MZ_FINISH))
+    {
+      if ((flush) || (pStream->total_in != orig_total_in) || (pStream->total_out != orig_total_out))
+        break;
+      return MZ_BUF_ERROR; // Can't make forward progress without some input.
+    }
+  }
+  return mz_status;
+}
+
+int mz_deflateEnd(mz_streamp pStream)
+{
+  if (!pStream) return MZ_STREAM_ERROR;
+  if (pStream->state)
+  {
+    pStream->zfree(pStream->opaque, pStream->state);
+    pStream->state = NULL;
+  }
+  return MZ_OK;
+}
+
+mz_ulong mz_deflateBound(mz_streamp pStream, mz_ulong source_len)
+{
+  (void)pStream;
+  // This is really over conservative. (And lame, but it's actually pretty tricky to compute a true upper bound given the way tdefl's blocking works.)
+  return MZ_MAX(128 + (source_len * 110) / 100, 128 + source_len + ((source_len / (31 * 1024)) + 1) * 5);
+}
+
+int mz_compress2(unsigned char *pDest, mz_ulong *pDest_len, const unsigned char *pSource, mz_ulong source_len, int level)
+{
+  int status;
+  mz_stream stream;
+  memset(&stream, 0, sizeof(stream));
+
+  // In case mz_ulong is 64-bits (argh I hate longs).
+  if ((source_len | *pDest_len) > 0xFFFFFFFFU) return MZ_PARAM_ERROR;
+
+  stream.next_in = pSource;
+  stream.avail_in = (mz_uint32)source_len;
+  stream.next_out = pDest;
+  stream.avail_out = (mz_uint32)*pDest_len;
+
+  status = mz_deflateInit(&stream, level);
+  if (status != MZ_OK) return status;
+
+  status = mz_deflate(&stream, MZ_FINISH);
+  if (status != MZ_STREAM_END)
+  {
+    mz_deflateEnd(&stream);
+    return (status == MZ_OK) ? MZ_BUF_ERROR : status;
+  }
+
+  *pDest_len = stream.total_out;
+  return mz_deflateEnd(&stream);
+}
+
+int mz_compress(unsigned char *pDest, mz_ulong *pDest_len, const unsigned char *pSource, mz_ulong source_len)
+{
+  return mz_compress2(pDest, pDest_len, pSource, source_len, MZ_DEFAULT_COMPRESSION);
+}
+
+mz_ulong mz_compressBound(mz_ulong source_len)
+{
+  return mz_deflateBound(NULL, source_len);
+}
+
+typedef struct
+{
+  tinfl_decompressor m_decomp;
+  mz_uint m_dict_ofs, m_dict_avail, m_first_call, m_has_flushed; int m_window_bits;
+  mz_uint8 m_dict[TINFL_LZ_DICT_SIZE];
+  tinfl_status m_last_status;
+} inflate_state;
+
+int mz_inflateInit2(mz_streamp pStream, int window_bits)
+{
+  inflate_state *pDecomp;
+  if (!pStream) return MZ_STREAM_ERROR;
+  if ((window_bits != MZ_DEFAULT_WINDOW_BITS) && (-window_bits != MZ_DEFAULT_WINDOW_BITS)) return MZ_PARAM_ERROR;
+
+  pStream->data_type = 0;
+  pStream->adler = 0;
+  pStream->msg = NULL;
+  pStream->total_in = 0;
+  pStream->total_out = 0;
+  pStream->reserved = 0;
+  if (!pStream->zalloc) pStream->zalloc = def_alloc_func;
+  if (!pStream->zfree) pStream->zfree = def_free_func;
+
+  pDecomp = (inflate_state*)pStream->zalloc(pStream->opaque, 1, sizeof(inflate_state));
+  if (!pDecomp) return MZ_MEM_ERROR;
+
+  pStream->state = (struct mz_internal_state *)pDecomp;
+
+  tinfl_init(&pDecomp->m_decomp);
+  pDecomp->m_dict_ofs = 0;
+  pDecomp->m_dict_avail = 0;
+  pDecomp->m_last_status = TINFL_STATUS_NEEDS_MORE_INPUT;
+  pDecomp->m_first_call = 1;
+  pDecomp->m_has_flushed = 0;
+  pDecomp->m_window_bits = window_bits;
+
+  return MZ_OK;
+}
+
+int mz_inflateInit(mz_streamp pStream)
+{
+   return mz_inflateInit2(pStream, MZ_DEFAULT_WINDOW_BITS);
+}
+
+int mz_inflate(mz_streamp pStream, int flush)
+{
+  inflate_state* pState;
+  mz_uint n, first_call, decomp_flags = TINFL_FLAG_COMPUTE_ADLER32;
+  size_t in_bytes, out_bytes, orig_avail_in;
+  tinfl_status status;
+
+  if ((!pStream) || (!pStream->state)) return MZ_STREAM_ERROR;
+  if (flush == MZ_PARTIAL_FLUSH) flush = MZ_SYNC_FLUSH;
+  if ((flush) && (flush != MZ_SYNC_FLUSH) && (flush != MZ_FINISH)) return MZ_STREAM_ERROR;
+
+  pState = (inflate_state*)pStream->state;
+  if (pState->m_window_bits > 0) decomp_flags |= TINFL_FLAG_PARSE_ZLIB_HEADER;
+  orig_avail_in = pStream->avail_in;
+
+  first_call = pState->m_first_call; pState->m_first_call = 0;
+  if (pState->m_last_status < 0) return MZ_DATA_ERROR;
+
+  if (pState->m_has_flushed && (flush != MZ_FINISH)) return MZ_STREAM_ERROR;
+  pState->m_has_flushed |= (flush == MZ_FINISH);
+
+  if ((flush == MZ_FINISH) && (first_call))
+  {
+    // MZ_FINISH on the first call implies that the input and output buffers are large enough to hold the entire compressed/decompressed file.
+    decomp_flags |= TINFL_FLAG_USING_NON_WRAPPING_OUTPUT_BUF;
+    in_bytes = pStream->avail_in; out_bytes = pStream->avail_out;
+    status = tinfl_decompress(&pState->m_decomp, pStream->next_in, &in_bytes, pStream->next_out, pStream->next_out, &out_bytes, decomp_flags);
+    pState->m_last_status = status;
+    pStream->next_in += (mz_uint)in_bytes; pStream->avail_in -= (mz_uint)in_bytes; pStream->total_in += (mz_uint)in_bytes;
+    pStream->adler = tinfl_get_adler32(&pState->m_decomp);
+    pStream->next_out += (mz_uint)out_bytes; pStream->avail_out -= (mz_uint)out_bytes; pStream->total_out += (mz_uint)out_bytes;
+
+    if (status < 0)
+      return MZ_DATA_ERROR;
+    else if (status != TINFL_STATUS_DONE)
+    {
+      pState->m_last_status = TINFL_STATUS_FAILED;
+      return MZ_BUF_ERROR;
+    }
+    return MZ_STREAM_END;
+  }
+  // flush != MZ_FINISH then we must assume there's more input.
+  if (flush != MZ_FINISH) decomp_flags |= TINFL_FLAG_HAS_MORE_INPUT;
+
+  if (pState->m_dict_avail)
+  {
+    n = MZ_MIN(pState->m_dict_avail, pStream->avail_out);
+    memcpy(pStream->next_out, pState->m_dict + pState->m_dict_ofs, n);
+    pStream->next_out += n; pStream->avail_out -= n; pStream->total_out += n;
+    pState->m_dict_avail -= n; pState->m_dict_ofs = (pState->m_dict_ofs + n) & (TINFL_LZ_DICT_SIZE - 1);
+    return ((pState->m_last_status == TINFL_STATUS_DONE) && (!pState->m_dict_avail)) ? MZ_STREAM_END : MZ_OK;
+  }
+
+  for ( ; ; )
+  {
+    in_bytes = pStream->avail_in;
+    out_bytes = TINFL_LZ_DICT_SIZE - pState->m_dict_ofs;
+
+    status = tinfl_decompress(&pState->m_decomp, pStream->next_in, &in_bytes, pState->m_dict, pState->m_dict + pState->m_dict_ofs, &out_bytes, decomp_flags);
+    pState->m_last_status = status;
+
+    pStream->next_in += (mz_uint)in_bytes; pStream->avail_in -= (mz_uint)in_bytes;
+    pStream->total_in += (mz_uint)in_bytes; pStream->adler = tinfl_get_adler32(&pState->m_decomp);
+
+    pState->m_dict_avail = (mz_uint)out_bytes;
+
+    n = MZ_MIN(pState->m_dict_avail, pStream->avail_out);
+    memcpy(pStream->next_out, pState->m_dict + pState->m_dict_ofs, n);
+    pStream->next_out += n; pStream->avail_out -= n; pStream->total_out += n;
+    pState->m_dict_avail -= n; pState->m_dict_ofs = (pState->m_dict_ofs + n) & (TINFL_LZ_DICT_SIZE - 1);
+
+    if (status < 0)
+       return MZ_DATA_ERROR; // Stream is corrupted (there could be some uncompressed data left in the output dictionary - oh well).
+    else if ((status == TINFL_STATUS_NEEDS_MORE_INPUT) && (!orig_avail_in))
+      return MZ_BUF_ERROR; // Signal caller that we can't make forward progress without supplying more input or by setting flush to MZ_FINISH.
+    else if (flush == MZ_FINISH)
+    {
+       // The output buffer MUST be large to hold the remaining uncompressed data when flush==MZ_FINISH.
+       if (status == TINFL_STATUS_DONE)
+          return pState->m_dict_avail ? MZ_BUF_ERROR : MZ_STREAM_END;
+       // status here must be TINFL_STATUS_HAS_MORE_OUTPUT, which means there's at least 1 more byte on the way. If there's no more room left in the output buffer then something is wrong.
+       else if (!pStream->avail_out)
+          return MZ_BUF_ERROR;
+    }
+    else if ((status == TINFL_STATUS_DONE) || (!pStream->avail_in) || (!pStream->avail_out) || (pState->m_dict_avail))
+      break;
+  }
+
+  return ((status == TINFL_STATUS_DONE) && (!pState->m_dict_avail)) ? MZ_STREAM_END : MZ_OK;
+}
+
+int mz_inflateEnd(mz_streamp pStream)
+{
+  if (!pStream)
+    return MZ_STREAM_ERROR;
+  if (pStream->state)
+  {
+    pStream->zfree(pStream->opaque, pStream->state);
+    pStream->state = NULL;
+  }
+  return MZ_OK;
+}
+
+int mz_uncompress(unsigned char *pDest, mz_ulong *pDest_len, const unsigned char *pSource, mz_ulong source_len)
+{
+  mz_stream stream;
+  int status;
+  memset(&stream, 0, sizeof(stream));
+
+  // In case mz_ulong is 64-bits (argh I hate longs).
+  if ((source_len | *pDest_len) > 0xFFFFFFFFU) return MZ_PARAM_ERROR;
+
+  stream.next_in = pSource;
+  stream.avail_in = (mz_uint32)source_len;
+  stream.next_out = pDest;
+  stream.avail_out = (mz_uint32)*pDest_len;
+
+  status = mz_inflateInit(&stream);
+  if (status != MZ_OK)
+    return status;
+
+  status = mz_inflate(&stream, MZ_FINISH);
+  if (status != MZ_STREAM_END)
+  {
+    mz_inflateEnd(&stream);
+    return ((status == MZ_BUF_ERROR) && (!stream.avail_in)) ? MZ_DATA_ERROR : status;
+  }
+  *pDest_len = stream.total_out;
+
+  return mz_inflateEnd(&stream);
+}
+
+const char *mz_error(int err)
+{
+  static struct { int m_err; const char *m_pDesc; } s_error_descs[] =
+  {
+    { MZ_OK, "" }, { MZ_STREAM_END, "stream end" }, { MZ_NEED_DICT, "need dictionary" }, { MZ_ERRNO, "file error" }, { MZ_STREAM_ERROR, "stream error" },
+    { MZ_DATA_ERROR, "data error" }, { MZ_MEM_ERROR, "out of memory" }, { MZ_BUF_ERROR, "buf error" }, { MZ_VERSION_ERROR, "version error" }, { MZ_PARAM_ERROR, "parameter error" }
+  };
+  mz_uint i; for (i = 0; i < sizeof(s_error_descs) / sizeof(s_error_descs[0]); ++i) if (s_error_descs[i].m_err == err) return s_error_descs[i].m_pDesc;
+  return NULL;
+}
+
+#endif //MINIZ_NO_ZLIB_APIS
+
+// ------------------- Low-level Decompression (completely independent from all compression API's)
+
+#define TINFL_MEMCPY(d, s, l) memcpy(d, s, l)
+#define TINFL_MEMSET(p, c, l) memset(p, c, l)
+
+#define TINFL_CR_BEGIN switch(r->m_state) { case 0:
+#define TINFL_CR_RETURN(state_index, result) do { status = result; r->m_state = state_index; goto common_exit; case state_index:; } MZ_MACRO_END
+#define TINFL_CR_RETURN_FOREVER(state_index, result) do { for ( ; ; ) { TINFL_CR_RETURN(state_index, result); } } MZ_MACRO_END
+#define TINFL_CR_FINISH }
+
+// TODO: If the caller has indicated that there's no more input, and we attempt to read beyond the input buf, then something is wrong with the input because the inflator never
+// reads ahead more than it needs to. Currently TINFL_GET_BYTE() pads the end of the stream with 0's in this scenario.
+#define TINFL_GET_BYTE(state_index, c) do { \
+  if (pIn_buf_cur >= pIn_buf_end) { \
+    for ( ; ; ) { \
+      if (decomp_flags & TINFL_FLAG_HAS_MORE_INPUT) { \
+        TINFL_CR_RETURN(state_index, TINFL_STATUS_NEEDS_MORE_INPUT); \
+        if (pIn_buf_cur < pIn_buf_end) { \
+          c = *pIn_buf_cur++; \
+          break; \
+        } \
+      } else { \
+        c = 0; \
+        break; \
+      } \
+    } \
+  } else c = *pIn_buf_cur++; } MZ_MACRO_END
+
+#define TINFL_NEED_BITS(state_index, n) do { mz_uint c; TINFL_GET_BYTE(state_index, c); bit_buf |= (((tinfl_bit_buf_t)c) << num_bits); num_bits += 8; } while (num_bits < (mz_uint)(n))
+#define TINFL_SKIP_BITS(state_index, n) do { if (num_bits < (mz_uint)(n)) { TINFL_NEED_BITS(state_index, n); } bit_buf >>= (n); num_bits -= (n); } MZ_MACRO_END
+#define TINFL_GET_BITS(state_index, b, n) do { if (num_bits < (mz_uint)(n)) { TINFL_NEED_BITS(state_index, n); } b = bit_buf & ((1 << (n)) - 1); bit_buf >>= (n); num_bits -= (n); } MZ_MACRO_END
+
+// TINFL_HUFF_BITBUF_FILL() is only used rarely, when the number of bytes remaining in the input buffer falls below 2.
+// It reads just enough bytes from the input stream that are needed to decode the next Huffman code (and absolutely no more). It works by trying to fully decode a
+// Huffman code by using whatever bits are currently present in the bit buffer. If this fails, it reads another byte, and tries again until it succeeds or until the
+// bit buffer contains >=15 bits (deflate's max. Huffman code size).
+#define TINFL_HUFF_BITBUF_FILL(state_index, pHuff) \
+  do { \
+    temp = (pHuff)->m_look_up[bit_buf & (TINFL_FAST_LOOKUP_SIZE - 1)]; \
+    if (temp >= 0) { \
+      code_len = temp >> 9; \
+      if ((code_len) && (num_bits >= code_len)) \
+      break; \
+    } else if (num_bits > TINFL_FAST_LOOKUP_BITS) { \
+       code_len = TINFL_FAST_LOOKUP_BITS; \
+       do { \
+          temp = (pHuff)->m_tree[~temp + ((bit_buf >> code_len++) & 1)]; \
+       } while ((temp < 0) && (num_bits >= (code_len + 1))); if (temp >= 0) break; \
+    } TINFL_GET_BYTE(state_index, c); bit_buf |= (((tinfl_bit_buf_t)c) << num_bits); num_bits += 8; \
+  } while (num_bits < 15);
+
+// TINFL_HUFF_DECODE() decodes the next Huffman coded symbol. It's more complex than you would initially expect because the zlib API expects the decompressor to never read
+// beyond the final byte of the deflate stream. (In other words, when this macro wants to read another byte from the input, it REALLY needs another byte in order to fully
+// decode the next Huffman code.) Handling this properly is particularly important on raw deflate (non-zlib) streams, which aren't followed by a byte aligned adler-32.
+// The slow path is only executed at the very end of the input buffer.
+#define TINFL_HUFF_DECODE(state_index, sym, pHuff) do { \
+  int temp; mz_uint code_len, c; \
+  if (num_bits < 15) { \
+    if ((pIn_buf_end - pIn_buf_cur) < 2) { \
+       TINFL_HUFF_BITBUF_FILL(state_index, pHuff); \
+    } else { \
+       bit_buf |= (((tinfl_bit_buf_t)pIn_buf_cur[0]) << num_bits) | (((tinfl_bit_buf_t)pIn_buf_cur[1]) << (num_bits + 8)); pIn_buf_cur += 2; num_bits += 16; \
+    } \
+  } \
+  if ((temp = (pHuff)->m_look_up[bit_buf & (TINFL_FAST_LOOKUP_SIZE - 1)]) >= 0) \
+    code_len = temp >> 9, temp &= 511; \
+  else { \
+    code_len = TINFL_FAST_LOOKUP_BITS; do { temp = (pHuff)->m_tree[~temp + ((bit_buf >> code_len++) & 1)]; } while (temp < 0); \
+  } sym = temp; bit_buf >>= code_len; num_bits -= code_len; } MZ_MACRO_END
+
+tinfl_status tinfl_decompress(tinfl_decompressor *r, const mz_uint8 *pIn_buf_next, size_t *pIn_buf_size, mz_uint8 *pOut_buf_start, mz_uint8 *pOut_buf_next, size_t *pOut_buf_size, const mz_uint32 decomp_flags)
+{
+  static const int s_length_base[31] = { 3,4,5,6,7,8,9,10,11,13, 15,17,19,23,27,31,35,43,51,59, 67,83,99,115,131,163,195,227,258,0,0 };
+  static const int s_length_extra[31]= { 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 };
+  static const int s_dist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193, 257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0};
+  static const int s_dist_extra[32] = { 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13};
+  static const mz_uint8 s_length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 };
+  static const int s_min_table_sizes[3] = { 257, 1, 4 };
+
+  tinfl_status status = TINFL_STATUS_FAILED; mz_uint32 num_bits, dist, counter, num_extra; tinfl_bit_buf_t bit_buf;
+  const mz_uint8 *pIn_buf_cur = pIn_buf_next, *const pIn_buf_end = pIn_buf_next + *pIn_buf_size;
+  mz_uint8 *pOut_buf_cur = pOut_buf_next, *const pOut_buf_end = pOut_buf_next + *pOut_buf_size;
+  size_t out_buf_size_mask = (decomp_flags & TINFL_FLAG_USING_NON_WRAPPING_OUTPUT_BUF) ? (size_t)-1 : ((pOut_buf_next - pOut_buf_start) + *pOut_buf_size) - 1, dist_from_out_buf_start;
+
+  // Ensure the output buffer's size is a power of 2, unless the output buffer is large enough to hold the entire output file (in which case it doesn't matter).
+  if (((out_buf_size_mask + 1) & out_buf_size_mask) || (pOut_buf_next < pOut_buf_start)) { *pIn_buf_size = *pOut_buf_size = 0; return TINFL_STATUS_BAD_PARAM; }
+
+  num_bits = r->m_num_bits; bit_buf = r->m_bit_buf; dist = r->m_dist; counter = r->m_counter; num_extra = r->m_num_extra; dist_from_out_buf_start = r->m_dist_from_out_buf_start;
+  TINFL_CR_BEGIN
+
+  bit_buf = num_bits = dist = counter = num_extra = r->m_zhdr0 = r->m_zhdr1 = 0; r->m_z_adler32 = r->m_check_adler32 = 1;
+  if (decomp_flags & TINFL_FLAG_PARSE_ZLIB_HEADER)
+  {
+    TINFL_GET_BYTE(1, r->m_zhdr0); TINFL_GET_BYTE(2, r->m_zhdr1);
+    counter = (((r->m_zhdr0 * 256 + r->m_zhdr1) % 31 != 0) || (r->m_zhdr1 & 32) || ((r->m_zhdr0 & 15) != 8));
+    if (!(decomp_flags & TINFL_FLAG_USING_NON_WRAPPING_OUTPUT_BUF)) counter |= (((1U << (8U + (r->m_zhdr0 >> 4))) > 32768U) || ((out_buf_size_mask + 1) < (size_t)(1U << (8U + (r->m_zhdr0 >> 4)))));
+    if (counter) { TINFL_CR_RETURN_FOREVER(36, TINFL_STATUS_FAILED); }
+  }
+
+  do
+  {
+    TINFL_GET_BITS(3, r->m_final, 3); r->m_type = r->m_final >> 1;
+    if (r->m_type == 0)
+    {
+      TINFL_SKIP_BITS(5, num_bits & 7);
+      for (counter = 0; counter < 4; ++counter) { if (num_bits) TINFL_GET_BITS(6, r->m_raw_header[counter], 8); else TINFL_GET_BYTE(7, r->m_raw_header[counter]); }
+      if ((counter = (r->m_raw_header[0] | (r->m_raw_header[1] << 8))) != (mz_uint)(0xFFFF ^ (r->m_raw_header[2] | (r->m_raw_header[3] << 8)))) { TINFL_CR_RETURN_FOREVER(39, TINFL_STATUS_FAILED); }
+      while ((counter) && (num_bits))
+      {
+        TINFL_GET_BITS(51, dist, 8);
+        while (pOut_buf_cur >= pOut_buf_end) { TINFL_CR_RETURN(52, TINFL_STATUS_HAS_MORE_OUTPUT); }
+        *pOut_buf_cur++ = (mz_uint8)dist;
+        counter--;
+      }
+      while (counter)
+      {
+        size_t n; while (pOut_buf_cur >= pOut_buf_end) { TINFL_CR_RETURN(9, TINFL_STATUS_HAS_MORE_OUTPUT); }
+        while (pIn_buf_cur >= pIn_buf_end)
+        {
+          if (decomp_flags & TINFL_FLAG_HAS_MORE_INPUT)
+          {
+            TINFL_CR_RETURN(38, TINFL_STATUS_NEEDS_MORE_INPUT);
+          }
+          else
+          {
+            TINFL_CR_RETURN_FOREVER(40, TINFL_STATUS_FAILED);
+          }
+        }
+        n = MZ_MIN(MZ_MIN((size_t)(pOut_buf_end - pOut_buf_cur), (size_t)(pIn_buf_end - pIn_buf_cur)), counter);
+        TINFL_MEMCPY(pOut_buf_cur, pIn_buf_cur, n); pIn_buf_cur += n; pOut_buf_cur += n; counter -= (mz_uint)n;
+      }
+    }
+    else if (r->m_type == 3)
+    {
+      TINFL_CR_RETURN_FOREVER(10, TINFL_STATUS_FAILED);
+    }
+    else
+    {
+      if (r->m_type == 1)
+      {
+        mz_uint8 *p = r->m_tables[0].m_code_size; mz_uint i;
+        r->m_table_sizes[0] = 288; r->m_table_sizes[1] = 32; TINFL_MEMSET(r->m_tables[1].m_code_size, 5, 32);
+        for ( i = 0; i <= 143; ++i) *p++ = 8; for ( ; i <= 255; ++i) *p++ = 9; for ( ; i <= 279; ++i) *p++ = 7; for ( ; i <= 287; ++i) *p++ = 8;
+      }
+      else
+      {
+        for (counter = 0; counter < 3; counter++) { TINFL_GET_BITS(11, r->m_table_sizes[counter], "\05\05\04"[counter]); r->m_table_sizes[counter] += s_min_table_sizes[counter]; }
+        MZ_CLEAR_OBJ(r->m_tables[2].m_code_size); for (counter = 0; counter < r->m_table_sizes[2]; counter++) { mz_uint s; TINFL_GET_BITS(14, s, 3); r->m_tables[2].m_code_size[s_length_dezigzag[counter]] = (mz_uint8)s; }
+        r->m_table_sizes[2] = 19;
+      }
+      for ( ; (int)r->m_type >= 0; r->m_type--)
+      {
+        int tree_next, tree_cur; tinfl_huff_table *pTable;
+        mz_uint i, j, used_syms, total, sym_index, next_code[17], total_syms[16]; pTable = &r->m_tables[r->m_type]; MZ_CLEAR_OBJ(total_syms); MZ_CLEAR_OBJ(pTable->m_look_up); MZ_CLEAR_OBJ(pTable->m_tree);
+        for (i = 0; i < r->m_table_sizes[r->m_type]; ++i) total_syms[pTable->m_code_size[i]]++;
+        used_syms = 0, total = 0; next_code[0] = next_code[1] = 0;
+        for (i = 1; i <= 15; ++i) { used_syms += total_syms[i]; next_code[i + 1] = (total = ((total + total_syms[i]) << 1)); }
+        if ((65536 != total) && (used_syms > 1))
+        {
+          TINFL_CR_RETURN_FOREVER(35, TINFL_STATUS_FAILED);
+        }
+        for (tree_next = -1, sym_index = 0; sym_index < r->m_table_sizes[r->m_type]; ++sym_index)
+        {
+          mz_uint rev_code = 0, l, cur_code, code_size = pTable->m_code_size[sym_index]; if (!code_size) continue;
+          cur_code = next_code[code_size]++; for (l = code_size; l > 0; l--, cur_code >>= 1) rev_code = (rev_code << 1) | (cur_code & 1);
+          if (code_size <= TINFL_FAST_LOOKUP_BITS) { mz_int16 k = (mz_int16)((code_size << 9) | sym_index); while (rev_code < TINFL_FAST_LOOKUP_SIZE) { pTable->m_look_up[rev_code] = k; rev_code += (1 << code_size); } continue; }
+          if (0 == (tree_cur = pTable->m_look_up[rev_code & (TINFL_FAST_LOOKUP_SIZE - 1)])) { pTable->m_look_up[rev_code & (TINFL_FAST_LOOKUP_SIZE - 1)] = (mz_int16)tree_next; tree_cur = tree_next; tree_next -= 2; }
+          rev_code >>= (TINFL_FAST_LOOKUP_BITS - 1);
+          for (j = code_size; j > (TINFL_FAST_LOOKUP_BITS + 1); j--)
+          {
+            tree_cur -= ((rev_code >>= 1) & 1);
+            if (!pTable->m_tree[-tree_cur - 1]) { pTable->m_tree[-tree_cur - 1] = (mz_int16)tree_next; tree_cur = tree_next; tree_next -= 2; } else tree_cur = pTable->m_tree[-tree_cur - 1];
+          }
+          tree_cur -= ((rev_code >>= 1) & 1); pTable->m_tree[-tree_cur - 1] = (mz_int16)sym_index;
+        }
+        if (r->m_type == 2)
+        {
+          for (counter = 0; counter < (r->m_table_sizes[0] + r->m_table_sizes[1]); )
+          {
+            mz_uint s; TINFL_HUFF_DECODE(16, dist, &r->m_tables[2]); if (dist < 16) { r->m_len_codes[counter++] = (mz_uint8)dist; continue; }
+            if ((dist == 16) && (!counter))
+            {
+              TINFL_CR_RETURN_FOREVER(17, TINFL_STATUS_FAILED);
+            }
+            num_extra = "\02\03\07"[dist - 16]; TINFL_GET_BITS(18, s, num_extra); s += "\03\03\013"[dist - 16];
+            TINFL_MEMSET(r->m_len_codes + counter, (dist == 16) ? r->m_len_codes[counter - 1] : 0, s); counter += s;
+          }
+          if ((r->m_table_sizes[0] + r->m_table_sizes[1]) != counter)
+          {
+            TINFL_CR_RETURN_FOREVER(21, TINFL_STATUS_FAILED);
+          }
+          TINFL_MEMCPY(r->m_tables[0].m_code_size, r->m_len_codes, r->m_table_sizes[0]); TINFL_MEMCPY(r->m_tables[1].m_code_size, r->m_len_codes + r->m_table_sizes[0], r->m_table_sizes[1]);
+        }
+      }
+      for ( ; ; )
+      {
+        mz_uint8 *pSrc;
+        for ( ; ; )
+        {
+          if (((pIn_buf_end - pIn_buf_cur) < 4) || ((pOut_buf_end - pOut_buf_cur) < 2))
+          {
+            TINFL_HUFF_DECODE(23, counter, &r->m_tables[0]);
+            if (counter >= 256)
+              break;
+            while (pOut_buf_cur >= pOut_buf_end) { TINFL_CR_RETURN(24, TINFL_STATUS_HAS_MORE_OUTPUT); }
+            *pOut_buf_cur++ = (mz_uint8)counter;
+          }
+          else
+          {
+            int sym2; mz_uint code_len;
+#if TINFL_USE_64BIT_BITBUF
+            if (num_bits < 30) { bit_buf |= (((tinfl_bit_buf_t)MZ_READ_LE32(pIn_buf_cur)) << num_bits); pIn_buf_cur += 4; num_bits += 32; }
+#else
+            if (num_bits < 15) { bit_buf |= (((tinfl_bit_buf_t)MZ_READ_LE16(pIn_buf_cur)) << num_bits); pIn_buf_cur += 2; num_bits += 16; }
+#endif
+            if ((sym2 = r->m_tables[0].m_look_up[bit_buf & (TINFL_FAST_LOOKUP_SIZE - 1)]) >= 0)
+              code_len = sym2 >> 9;
+            else
+            {
+              code_len = TINFL_FAST_LOOKUP_BITS; do { sym2 = r->m_tables[0].m_tree[~sym2 + ((bit_buf >> code_len++) & 1)]; } while (sym2 < 0);
+            }
+            counter = sym2; bit_buf >>= code_len; num_bits -= code_len;
+            if (counter & 256)
+              break;
+
+#if !TINFL_USE_64BIT_BITBUF
+            if (num_bits < 15) { bit_buf |= (((tinfl_bit_buf_t)MZ_READ_LE16(pIn_buf_cur)) << num_bits); pIn_buf_cur += 2; num_bits += 16; }
+#endif
+            if ((sym2 = r->m_tables[0].m_look_up[bit_buf & (TINFL_FAST_LOOKUP_SIZE - 1)]) >= 0)
+              code_len = sym2 >> 9;
+            else
+            {
+              code_len = TINFL_FAST_LOOKUP_BITS; do { sym2 = r->m_tables[0].m_tree[~sym2 + ((bit_buf >> code_len++) & 1)]; } while (sym2 < 0);
+            }
+            bit_buf >>= code_len; num_bits -= code_len;
+
+            pOut_buf_cur[0] = (mz_uint8)counter;
+            if (sym2 & 256)
+            {
+              pOut_buf_cur++;
+              counter = sym2;
+              break;
+            }
+            pOut_buf_cur[1] = (mz_uint8)sym2;
+            pOut_buf_cur += 2;
+          }
+        }
+        if ((counter &= 511) == 256) break;
+
+        num_extra = s_length_extra[counter - 257]; counter = s_length_base[counter - 257];
+        if (num_extra) { mz_uint extra_bits; TINFL_GET_BITS(25, extra_bits, num_extra); counter += extra_bits; }
+
+        TINFL_HUFF_DECODE(26, dist, &r->m_tables[1]);
+        num_extra = s_dist_extra[dist]; dist = s_dist_base[dist];
+        if (num_extra) { mz_uint extra_bits; TINFL_GET_BITS(27, extra_bits, num_extra); dist += extra_bits; }
+
+        dist_from_out_buf_start = pOut_buf_cur - pOut_buf_start;
+        if ((dist > dist_from_out_buf_start) && (decomp_flags & TINFL_FLAG_USING_NON_WRAPPING_OUTPUT_BUF))
+        {
+          TINFL_CR_RETURN_FOREVER(37, TINFL_STATUS_FAILED);
+        }
+
+        pSrc = pOut_buf_start + ((dist_from_out_buf_start - dist) & out_buf_size_mask);
+
+        if ((MZ_MAX(pOut_buf_cur, pSrc) + counter) > pOut_buf_end)
+        {
+          while (counter--)
+          {
+            while (pOut_buf_cur >= pOut_buf_end) { TINFL_CR_RETURN(53, TINFL_STATUS_HAS_MORE_OUTPUT); }
+            *pOut_buf_cur++ = pOut_buf_start[(dist_from_out_buf_start++ - dist) & out_buf_size_mask];
+          }
+          continue;
+        }
+#if MINIZ_USE_UNALIGNED_LOADS_AND_STORES
+        else if ((counter >= 9) && (counter <= dist))
+        {
+          const mz_uint8 *pSrc_end = pSrc + (counter & ~7);
+          do
+          {
+            ((mz_uint32 *)pOut_buf_cur)[0] = ((const mz_uint32 *)pSrc)[0];
+            ((mz_uint32 *)pOut_buf_cur)[1] = ((const mz_uint32 *)pSrc)[1];
+            pOut_buf_cur += 8;
+          } while ((pSrc += 8) < pSrc_end);
+          if ((counter &= 7) < 3)
+          {
+            if (counter)
+            {
+              pOut_buf_cur[0] = pSrc[0];
+              if (counter > 1)
+                pOut_buf_cur[1] = pSrc[1];
+              pOut_buf_cur += counter;
+            }
+            continue;
+          }
+        }
+#endif
+        do
+        {
+          pOut_buf_cur[0] = pSrc[0];
+          pOut_buf_cur[1] = pSrc[1];
+          pOut_buf_cur[2] = pSrc[2];
+          pOut_buf_cur += 3; pSrc += 3;
+        } while ((int)(counter -= 3) > 2);
+        if ((int)counter > 0)
+        {
+          pOut_buf_cur[0] = pSrc[0];
+          if ((int)counter > 1)
+            pOut_buf_cur[1] = pSrc[1];
+          pOut_buf_cur += counter;
+        }
+      }
+    }
+  } while (!(r->m_final & 1));
+  if (decomp_flags & TINFL_FLAG_PARSE_ZLIB_HEADER)
+  {
+    TINFL_SKIP_BITS(32, num_bits & 7); for (counter = 0; counter < 4; ++counter) { mz_uint s; if (num_bits) TINFL_GET_BITS(41, s, 8); else TINFL_GET_BYTE(42, s); r->m_z_adler32 = (r->m_z_adler32 << 8) | s; }
+  }
+  TINFL_CR_RETURN_FOREVER(34, TINFL_STATUS_DONE);
+  TINFL_CR_FINISH
+
+common_exit:
+  r->m_num_bits = num_bits; r->m_bit_buf = bit_buf; r->m_dist = dist; r->m_counter = counter; r->m_num_extra = num_extra; r->m_dist_from_out_buf_start = dist_from_out_buf_start;
+  *pIn_buf_size = pIn_buf_cur - pIn_buf_next; *pOut_buf_size = pOut_buf_cur - pOut_buf_next;
+  if ((decomp_flags & (TINFL_FLAG_PARSE_ZLIB_HEADER | TINFL_FLAG_COMPUTE_ADLER32)) && (status >= 0))
+  {
+    const mz_uint8 *ptr = pOut_buf_next; size_t buf_len = *pOut_buf_size;
+    mz_uint32 i, s1 = r->m_check_adler32 & 0xffff, s2 = r->m_check_adler32 >> 16; size_t block_len = buf_len % 5552;
+    while (buf_len)
+    {
+      for (i = 0; i + 7 < block_len; i += 8, ptr += 8)
+      {
+        s1 += ptr[0], s2 += s1; s1 += ptr[1], s2 += s1; s1 += ptr[2], s2 += s1; s1 += ptr[3], s2 += s1;
+        s1 += ptr[4], s2 += s1; s1 += ptr[5], s2 += s1; s1 += ptr[6], s2 += s1; s1 += ptr[7], s2 += s1;
+      }
+      for ( ; i < block_len; ++i) s1 += *ptr++, s2 += s1;
+      s1 %= 65521U, s2 %= 65521U; buf_len -= block_len; block_len = 5552;
+    }
+    r->m_check_adler32 = (s2 << 16) + s1; if ((status == TINFL_STATUS_DONE) && (decomp_flags & TINFL_FLAG_PARSE_ZLIB_HEADER) && (r->m_check_adler32 != r->m_z_adler32)) status = TINFL_STATUS_ADLER32_MISMATCH;
+  }
+  return status;
+}
+
+// Higher level helper functions.
+void *tinfl_decompress_mem_to_heap(const void *pSrc_buf, size_t src_buf_len, size_t *pOut_len, int flags)
+{
+  tinfl_decompressor decomp; void *pBuf = NULL, *pNew_buf; size_t src_buf_ofs = 0, out_buf_capacity = 0;
+  *pOut_len = 0;
+  tinfl_init(&decomp);
+  for ( ; ; )
+  {
+    size_t src_buf_size = src_buf_len - src_buf_ofs, dst_buf_size = out_buf_capacity - *pOut_len, new_out_buf_capacity;
+    tinfl_status status = tinfl_decompress(&decomp, (const mz_uint8*)pSrc_buf + src_buf_ofs, &src_buf_size, (mz_uint8*)pBuf, pBuf ? (mz_uint8*)pBuf + *pOut_len : NULL, &dst_buf_size,
+      (flags & ~TINFL_FLAG_HAS_MORE_INPUT) | TINFL_FLAG_USING_NON_WRAPPING_OUTPUT_BUF);
+    if ((status < 0) || (status == TINFL_STATUS_NEEDS_MORE_INPUT))
+    {
+      MZ_FREE(pBuf); *pOut_len = 0; return NULL;
+    }
+    src_buf_ofs += src_buf_size;
+    *pOut_len += dst_buf_size;
+    if (status == TINFL_STATUS_DONE) break;
+    new_out_buf_capacity = out_buf_capacity * 2; if (new_out_buf_capacity < 128) new_out_buf_capacity = 128;
+    pNew_buf = MZ_REALLOC(pBuf, new_out_buf_capacity);
+    if (!pNew_buf)
+    {
+      MZ_FREE(pBuf); *pOut_len = 0; return NULL;
+    }
+    pBuf = pNew_buf; out_buf_capacity = new_out_buf_capacity;
+  }
+  return pBuf;
+}
+
+size_t tinfl_decompress_mem_to_mem(void *pOut_buf, size_t out_buf_len, const void *pSrc_buf, size_t src_buf_len, int flags)
+{
+  tinfl_decompressor decomp; tinfl_status status; tinfl_init(&decomp);
+  status = tinfl_decompress(&decomp, (const mz_uint8*)pSrc_buf, &src_buf_len, (mz_uint8*)pOut_buf, (mz_uint8*)pOut_buf, &out_buf_len, (flags & ~TINFL_FLAG_HAS_MORE_INPUT) | TINFL_FLAG_USING_NON_WRAPPING_OUTPUT_BUF);
+  return (status != TINFL_STATUS_DONE) ? TINFL_DECOMPRESS_MEM_TO_MEM_FAILED : out_buf_len;
+}
+
+int tinfl_decompress_mem_to_callback(const void *pIn_buf, size_t *pIn_buf_size, tinfl_put_buf_func_ptr pPut_buf_func, void *pPut_buf_user, int flags)
+{
+  int result = 0;
+  tinfl_decompressor decomp;
+  mz_uint8 *pDict = (mz_uint8*)MZ_MALLOC(TINFL_LZ_DICT_SIZE); size_t in_buf_ofs = 0, dict_ofs = 0;
+  if (!pDict)
+    return TINFL_STATUS_FAILED;
+  tinfl_init(&decomp);
+  for ( ; ; )
+  {
+    size_t in_buf_size = *pIn_buf_size - in_buf_ofs, dst_buf_size = TINFL_LZ_DICT_SIZE - dict_ofs;
+    tinfl_status status = tinfl_decompress(&decomp, (const mz_uint8*)pIn_buf + in_buf_ofs, &in_buf_size, pDict, pDict + dict_ofs, &dst_buf_size,
+      (flags & ~(TINFL_FLAG_HAS_MORE_INPUT | TINFL_FLAG_USING_NON_WRAPPING_OUTPUT_BUF)));
+    in_buf_ofs += in_buf_size;
+    if ((dst_buf_size) && (!(*pPut_buf_func)(pDict + dict_ofs, (int)dst_buf_size, pPut_buf_user)))
+      break;
+    if (status != TINFL_STATUS_HAS_MORE_OUTPUT)
+    {
+      result = (status == TINFL_STATUS_DONE);
+      break;
+    }
+    dict_ofs = (dict_ofs + dst_buf_size) & (TINFL_LZ_DICT_SIZE - 1);
+  }
+  MZ_FREE(pDict);
+  *pIn_buf_size = in_buf_ofs;
+  return result;
+}
+
+// ------------------- Low-level Compression (independent from all decompression API's)
+
+// Purposely making these tables static for faster init and thread safety.
+static const mz_uint16 s_tdefl_len_sym[256] = {
+  257,258,259,260,261,262,263,264,265,265,266,266,267,267,268,268,269,269,269,269,270,270,270,270,271,271,271,271,272,272,272,272,
+  273,273,273,273,273,273,273,273,274,274,274,274,274,274,274,274,275,275,275,275,275,275,275,275,276,276,276,276,276,276,276,276,
+  277,277,277,277,277,277,277,277,277,277,277,277,277,277,277,277,278,278,278,278,278,278,278,278,278,278,278,278,278,278,278,278,
+  279,279,279,279,279,279,279,279,279,279,279,279,279,279,279,279,280,280,280,280,280,280,280,280,280,280,280,280,280,280,280,280,
+  281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,
+  282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,
+  283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,
+  284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,285 };
+
+static const mz_uint8 s_tdefl_len_extra[256] = {
+  0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,
+  4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,
+  5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,
+  5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,0 };
+
+static const mz_uint8 s_tdefl_small_dist_sym[512] = {
+  0,1,2,3,4,4,5,5,6,6,6,6,7,7,7,7,8,8,8,8,8,8,8,8,9,9,9,9,9,9,9,9,10,10,10,10,10,10,10,10,10,10,10,10,10,10,10,10,11,11,11,11,11,11,
+  11,11,11,11,11,11,11,11,11,11,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,13,
+  13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,14,14,14,14,14,14,14,14,14,14,14,14,
+  14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,
+  14,14,14,14,14,14,14,14,14,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,
+  15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,16,16,16,16,16,16,16,16,16,16,16,16,16,
+  16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,
+  16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,
+  16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,17,17,17,17,17,17,17,17,17,17,17,17,17,17,
+  17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,
+  17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,
+  17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17 };
+
+static const mz_uint8 s_tdefl_small_dist_extra[512] = {
+  0,0,0,0,1,1,1,1,2,2,2,2,2,2,2,2,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,5,5,5,5,5,5,5,5,
+  5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
+  6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
+  6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7 };
+
+static const mz_uint8 s_tdefl_large_dist_sym[128] = {
+  0,0,18,19,20,20,21,21,22,22,22,22,23,23,23,23,24,24,24,24,24,24,24,24,25,25,25,25,25,25,25,25,26,26,26,26,26,26,26,26,26,26,26,26,
+  26,26,26,26,27,27,27,27,27,27,27,27,27,27,27,27,27,27,27,27,28,28,28,28,28,28,28,28,28,28,28,28,28,28,28,28,28,28,28,28,28,28,28,28,
+  28,28,28,28,28,28,28,28,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29,29 };
+
+static const mz_uint8 s_tdefl_large_dist_extra[128] = {
+  0,0,8,8,9,9,9,9,10,10,10,10,10,10,10,10,11,11,11,11,11,11,11,11,11,11,11,11,11,11,11,11,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,
+  12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,
+  13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13 };
+
+// Radix sorts tdefl_sym_freq[] array by 16-bit key m_key. Returns ptr to sorted values.
+typedef struct { mz_uint16 m_key, m_sym_index; } tdefl_sym_freq;
+static tdefl_sym_freq* tdefl_radix_sort_syms(mz_uint num_syms, tdefl_sym_freq* pSyms0, tdefl_sym_freq* pSyms1)
+{
+  mz_uint32 total_passes = 2, pass_shift, pass, i, hist[256 * 2]; tdefl_sym_freq* pCur_syms = pSyms0, *pNew_syms = pSyms1; MZ_CLEAR_OBJ(hist);
+  for (i = 0; i < num_syms; i++) { mz_uint freq = pSyms0[i].m_key; hist[freq & 0xFF]++; hist[256 + ((freq >> 8) & 0xFF)]++; }
+  while ((total_passes > 1) && (num_syms == hist[(total_passes - 1) * 256])) total_passes--;
+  for (pass_shift = 0, pass = 0; pass < total_passes; pass++, pass_shift += 8)
+  {
+    const mz_uint32* pHist = &hist[pass << 8];
+    mz_uint offsets[256], cur_ofs = 0;
+    for (i = 0; i < 256; i++) { offsets[i] = cur_ofs; cur_ofs += pHist[i]; }
+    for (i = 0; i < num_syms; i++) pNew_syms[offsets[(pCur_syms[i].m_key >> pass_shift) & 0xFF]++] = pCur_syms[i];
+    { tdefl_sym_freq* t = pCur_syms; pCur_syms = pNew_syms; pNew_syms = t; }
+  }
+  return pCur_syms;
+}
+
+// tdefl_calculate_minimum_redundancy() originally written by: Alistair Moffat, alistair@cs.mu.oz.au, Jyrki Katajainen, jyrki@diku.dk, November 1996.
+static void tdefl_calculate_minimum_redundancy(tdefl_sym_freq *A, int n)
+{
+  int root, leaf, next, avbl, used, dpth;
+  if (n==0) return; else if (n==1) { A[0].m_key = 1; return; }
+  A[0].m_key += A[1].m_key; root = 0; leaf = 2;
+  for (next=1; next < n-1; next++)
+  {
+    if (leaf>=n || A[root].m_key<A[leaf].m_key) { A[next].m_key = A[root].m_key; A[root++].m_key = (mz_uint16)next; } else A[next].m_key = A[leaf++].m_key;
+    if (leaf>=n || (root<next && A[root].m_key<A[leaf].m_key)) { A[next].m_key = (mz_uint16)(A[next].m_key + A[root].m_key); A[root++].m_key = (mz_uint16)next; } else A[next].m_key = (mz_uint16)(A[next].m_key + A[leaf++].m_key);
+  }
+  A[n-2].m_key = 0; for (next=n-3; next>=0; next--) A[next].m_key = A[A[next].m_key].m_key+1;
+  avbl = 1; used = dpth = 0; root = n-2; next = n-1;
+  while (avbl>0)
+  {
+    while (root>=0 && (int)A[root].m_key==dpth) { used++; root--; }
+    while (avbl>used) { A[next--].m_key = (mz_uint16)(dpth); avbl--; }
+    avbl = 2*used; dpth++; used = 0;
+  }
+}
+
+// Limits canonical Huffman code table's max code size.
+enum { TDEFL_MAX_SUPPORTED_HUFF_CODESIZE = 32 };
+static void tdefl_huffman_enforce_max_code_size(int *pNum_codes, int code_list_len, int max_code_size)
+{
+  int i; mz_uint32 total = 0; if (code_list_len <= 1) return;
+  for (i = max_code_size + 1; i <= TDEFL_MAX_SUPPORTED_HUFF_CODESIZE; i++) pNum_codes[max_code_size] += pNum_codes[i];
+  for (i = max_code_size; i > 0; i--) total += (((mz_uint32)pNum_codes[i]) << (max_code_size - i));
+  while (total != (1UL << max_code_size))
+  {
+    pNum_codes[max_code_size]--;
+    for (i = max_code_size - 1; i > 0; i--) if (pNum_codes[i]) { pNum_codes[i]--; pNum_codes[i + 1] += 2; break; }
+    total--;
+  }
+}
+
+static void tdefl_optimize_huffman_table(tdefl_compressor *d, int table_num, int table_len, int code_size_limit, int static_table)
+{
+  int i, j, l, num_codes[1 + TDEFL_MAX_SUPPORTED_HUFF_CODESIZE]; mz_uint next_code[TDEFL_MAX_SUPPORTED_HUFF_CODESIZE + 1]; MZ_CLEAR_OBJ(num_codes);
+  if (static_table)
+  {
+    for (i = 0; i < table_len; i++) num_codes[d->m_huff_code_sizes[table_num][i]]++;
+  }
+  else
+  {
+    tdefl_sym_freq syms0[TDEFL_MAX_HUFF_SYMBOLS], syms1[TDEFL_MAX_HUFF_SYMBOLS], *pSyms;
+    int num_used_syms = 0;
+    const mz_uint16 *pSym_count = &d->m_huff_count[table_num][0];
+    for (i = 0; i < table_len; i++) if (pSym_count[i]) { syms0[num_used_syms].m_key = (mz_uint16)pSym_count[i]; syms0[num_used_syms++].m_sym_index = (mz_uint16)i; }
+
+    pSyms = tdefl_radix_sort_syms(num_used_syms, syms0, syms1); tdefl_calculate_minimum_redundancy(pSyms, num_used_syms);
+
+    for (i = 0; i < num_used_syms; i++) num_codes[pSyms[i].m_key]++;
+
+    tdefl_huffman_enforce_max_code_size(num_codes, num_used_syms, code_size_limit);
+
+    MZ_CLEAR_OBJ(d->m_huff_code_sizes[table_num]); MZ_CLEAR_OBJ(d->m_huff_codes[table_num]);
+    for (i = 1, j = num_used_syms; i <= code_size_limit; i++)
+      for (l = num_codes[i]; l > 0; l--) d->m_huff_code_sizes[table_num][pSyms[--j].m_sym_index] = (mz_uint8)(i);
+  }
+
+  next_code[1] = 0; for (j = 0, i = 2; i <= code_size_limit; i++) next_code[i] = j = ((j + num_codes[i - 1]) << 1);
+
+  for (i = 0; i < table_len; i++)
+  {
+    mz_uint rev_code = 0, code, code_size; if ((code_size = d->m_huff_code_sizes[table_num][i]) == 0) continue;
+    code = next_code[code_size]++; for (l = code_size; l > 0; l--, code >>= 1) rev_code = (rev_code << 1) | (code & 1);
+    d->m_huff_codes[table_num][i] = (mz_uint16)rev_code;
+  }
+}
+
+#define TDEFL_PUT_BITS(b, l) do { \
+  mz_uint bits = b; mz_uint len = l; MZ_ASSERT(bits <= ((1U << len) - 1U)); \
+  d->m_bit_buffer |= (bits << d->m_bits_in); d->m_bits_in += len; \
+  while (d->m_bits_in >= 8) { \
+    if (d->m_pOutput_buf < d->m_pOutput_buf_end) \
+      *d->m_pOutput_buf++ = (mz_uint8)(d->m_bit_buffer); \
+      d->m_bit_buffer >>= 8; \
+      d->m_bits_in -= 8; \
+  } \
+} MZ_MACRO_END
+
+#define TDEFL_RLE_PREV_CODE_SIZE() { if (rle_repeat_count) { \
+  if (rle_repeat_count < 3) { \
+    d->m_huff_count[2][prev_code_size] = (mz_uint16)(d->m_huff_count[2][prev_code_size] + rle_repeat_count); \
+    while (rle_repeat_count--) packed_code_sizes[num_packed_code_sizes++] = prev_code_size; \
+  } else { \
+    d->m_huff_count[2][16] = (mz_uint16)(d->m_huff_count[2][16] + 1); packed_code_sizes[num_packed_code_sizes++] = 16; packed_code_sizes[num_packed_code_sizes++] = (mz_uint8)(rle_repeat_count - 3); \
+} rle_repeat_count = 0; } }
+
+#define TDEFL_RLE_ZERO_CODE_SIZE() { if (rle_z_count) { \
+  if (rle_z_count < 3) { \
+    d->m_huff_count[2][0] = (mz_uint16)(d->m_huff_count[2][0] + rle_z_count); while (rle_z_count--) packed_code_sizes[num_packed_code_sizes++] = 0; \
+  } else if (rle_z_count <= 10) { \
+    d->m_huff_count[2][17] = (mz_uint16)(d->m_huff_count[2][17] + 1); packed_code_sizes[num_packed_code_sizes++] = 17; packed_code_sizes[num_packed_code_sizes++] = (mz_uint8)(rle_z_count - 3); \
+  } else { \
+    d->m_huff_count[2][18] = (mz_uint16)(d->m_huff_count[2][18] + 1); packed_code_sizes[num_packed_code_sizes++] = 18; packed_code_sizes[num_packed_code_sizes++] = (mz_uint8)(rle_z_count - 11); \
+} rle_z_count = 0; } }
+
+static mz_uint8 s_tdefl_packed_code_size_syms_swizzle[] = { 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 };
+
+static void tdefl_start_dynamic_block(tdefl_compressor *d)
+{
+  int num_lit_codes, num_dist_codes, num_bit_lengths; mz_uint i, total_code_sizes_to_pack, num_packed_code_sizes, rle_z_count, rle_repeat_count, packed_code_sizes_index;
+  mz_uint8 code_sizes_to_pack[TDEFL_MAX_HUFF_SYMBOLS_0 + TDEFL_MAX_HUFF_SYMBOLS_1], packed_code_sizes[TDEFL_MAX_HUFF_SYMBOLS_0 + TDEFL_MAX_HUFF_SYMBOLS_1], prev_code_size = 0xFF;
+
+  d->m_huff_count[0][256] = 1;
+
+  tdefl_optimize_huffman_table(d, 0, TDEFL_MAX_HUFF_SYMBOLS_0, 15, MZ_FALSE);
+  tdefl_optimize_huffman_table(d, 1, TDEFL_MAX_HUFF_SYMBOLS_1, 15, MZ_FALSE);
+
+  for (num_lit_codes = 286; num_lit_codes > 257; num_lit_codes--) if (d->m_huff_code_sizes[0][num_lit_codes - 1]) break;
+  for (num_dist_codes = 30; num_dist_codes > 1; num_dist_codes--) if (d->m_huff_code_sizes[1][num_dist_codes - 1]) break;
+
+  memcpy(code_sizes_to_pack, &d->m_huff_code_sizes[0][0], num_lit_codes);
+  memcpy(code_sizes_to_pack + num_lit_codes, &d->m_huff_code_sizes[1][0], num_dist_codes);
+  total_code_sizes_to_pack = num_lit_codes + num_dist_codes; num_packed_code_sizes = 0; rle_z_count = 0; rle_repeat_count = 0;
+
+  memset(&d->m_huff_count[2][0], 0, sizeof(d->m_huff_count[2][0]) * TDEFL_MAX_HUFF_SYMBOLS_2);
+  for (i = 0; i < total_code_sizes_to_pack; i++)
+  {
+    mz_uint8 code_size = code_sizes_to_pack[i];
+    if (!code_size)
+    {
+      TDEFL_RLE_PREV_CODE_SIZE();
+      if (++rle_z_count == 138) { TDEFL_RLE_ZERO_CODE_SIZE(); }
+    }
+    else
+    {
+      TDEFL_RLE_ZERO_CODE_SIZE();
+      if (code_size != prev_code_size)
+      {
+        TDEFL_RLE_PREV_CODE_SIZE();
+        d->m_huff_count[2][code_size] = (mz_uint16)(d->m_huff_count[2][code_size] + 1); packed_code_sizes[num_packed_code_sizes++] = code_size;
+      }
+      else if (++rle_repeat_count == 6)
+      {
+        TDEFL_RLE_PREV_CODE_SIZE();
+      }
+    }
+    prev_code_size = code_size;
+  }
+  if (rle_repeat_count) { TDEFL_RLE_PREV_CODE_SIZE(); } else { TDEFL_RLE_ZERO_CODE_SIZE(); }
+
+  tdefl_optimize_huffman_table(d, 2, TDEFL_MAX_HUFF_SYMBOLS_2, 7, MZ_FALSE);
+
+  TDEFL_PUT_BITS(2, 2);
+
+  TDEFL_PUT_BITS(num_lit_codes - 257, 5);
+  TDEFL_PUT_BITS(num_dist_codes - 1, 5);
+
+  for (num_bit_lengths = 18; num_bit_lengths >= 0; num_bit_lengths--) if (d->m_huff_code_sizes[2][s_tdefl_packed_code_size_syms_swizzle[num_bit_lengths]]) break;
+  num_bit_lengths = MZ_MAX(4, (num_bit_lengths + 1)); TDEFL_PUT_BITS(num_bit_lengths - 4, 4);
+  for (i = 0; (int)i < num_bit_lengths; i++) TDEFL_PUT_BITS(d->m_huff_code_sizes[2][s_tdefl_packed_code_size_syms_swizzle[i]], 3);
+
+  for (packed_code_sizes_index = 0; packed_code_sizes_index < num_packed_code_sizes; )
+  {
+    mz_uint code = packed_code_sizes[packed_code_sizes_index++]; MZ_ASSERT(code < TDEFL_MAX_HUFF_SYMBOLS_2);
+    TDEFL_PUT_BITS(d->m_huff_codes[2][code], d->m_huff_code_sizes[2][code]);
+    if (code >= 16) TDEFL_PUT_BITS(packed_code_sizes[packed_code_sizes_index++], "\02\03\07"[code - 16]);
+  }
+}
+
+static void tdefl_start_static_block(tdefl_compressor *d)
+{
+  mz_uint i;
+  mz_uint8 *p = &d->m_huff_code_sizes[0][0];
+
+  for (i = 0; i <= 143; ++i) *p++ = 8;
+  for ( ; i <= 255; ++i) *p++ = 9;
+  for ( ; i <= 279; ++i) *p++ = 7;
+  for ( ; i <= 287; ++i) *p++ = 8;
+
+  memset(d->m_huff_code_sizes[1], 5, 32);
+
+  tdefl_optimize_huffman_table(d, 0, 288, 15, MZ_TRUE);
+  tdefl_optimize_huffman_table(d, 1, 32, 15, MZ_TRUE);
+
+  TDEFL_PUT_BITS(1, 2);
+}
+
+static const mz_uint mz_bitmasks[17] = { 0x0000, 0x0001, 0x0003, 0x0007, 0x000F, 0x001F, 0x003F, 0x007F, 0x00FF, 0x01FF, 0x03FF, 0x07FF, 0x0FFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF };
+
+#if MINIZ_USE_UNALIGNED_LOADS_AND_STORES && MINIZ_LITTLE_ENDIAN && MINIZ_HAS_64BIT_REGISTERS
+static mz_bool tdefl_compress_lz_codes(tdefl_compressor *d)
+{
+  mz_uint flags;
+  mz_uint8 *pLZ_codes;
+  mz_uint8 *pOutput_buf = d->m_pOutput_buf;
+  mz_uint8 *pLZ_code_buf_end = d->m_pLZ_code_buf;
+  mz_uint64 bit_buffer = d->m_bit_buffer;
+  mz_uint bits_in = d->m_bits_in;
+
+#define TDEFL_PUT_BITS_FAST(b, l) { bit_buffer |= (((mz_uint64)(b)) << bits_in); bits_in += (l); }
+
+  flags = 1;
+  for (pLZ_codes = d->m_lz_code_buf; pLZ_codes < pLZ_code_buf_end; flags >>= 1)
+  {
+    if (flags == 1)
+      flags = *pLZ_codes++ | 0x100;
+
+    if (flags & 1)
+    {
+      mz_uint s0, s1, n0, n1, sym, num_extra_bits;
+      mz_uint match_len = pLZ_codes[0], match_dist = *(const mz_uint16 *)(pLZ_codes + 1); pLZ_codes += 3;
+
+      MZ_ASSERT(d->m_huff_code_sizes[0][s_tdefl_len_sym[match_len]]);
+      TDEFL_PUT_BITS_FAST(d->m_huff_codes[0][s_tdefl_len_sym[match_len]], d->m_huff_code_sizes[0][s_tdefl_len_sym[match_len]]);
+      TDEFL_PUT_BITS_FAST(match_len & mz_bitmasks[s_tdefl_len_extra[match_len]], s_tdefl_len_extra[match_len]);
+
+      // This sequence coaxes MSVC into using cmov's vs. jmp's.
+      s0 = s_tdefl_small_dist_sym[match_dist & 511];
+      n0 = s_tdefl_small_dist_extra[match_dist & 511];
+      s1 = s_tdefl_large_dist_sym[match_dist >> 8];
+      n1 = s_tdefl_large_dist_extra[match_dist >> 8];
+      sym = (match_dist < 512) ? s0 : s1;
+      num_extra_bits = (match_dist < 512) ? n0 : n1;
+
+      MZ_ASSERT(d->m_huff_code_sizes[1][sym]);
+      TDEFL_PUT_BITS_FAST(d->m_huff_codes[1][sym], d->m_huff_code_sizes[1][sym]);
+      TDEFL_PUT_BITS_FAST(match_dist & mz_bitmasks[num_extra_bits], num_extra_bits);
+    }
+    else
+    {
+      mz_uint lit = *pLZ_codes++;
+      MZ_ASSERT(d->m_huff_code_sizes[0][lit]);
+      TDEFL_PUT_BITS_FAST(d->m_huff_codes[0][lit], d->m_huff_code_sizes[0][lit]);
+
+      if (((flags & 2) == 0) && (pLZ_codes < pLZ_code_buf_end))
+      {
+        flags >>= 1;
+        lit = *pLZ_codes++;
+        MZ_ASSERT(d->m_huff_code_sizes[0][lit]);
+        TDEFL_PUT_BITS_FAST(d->m_huff_codes[0][lit], d->m_huff_code_sizes[0][lit]);
+
+        if (((flags & 2) == 0) && (pLZ_codes < pLZ_code_buf_end))
+        {
+          flags >>= 1;
+          lit = *pLZ_codes++;
+          MZ_ASSERT(d->m_huff_code_sizes[0][lit]);
+          TDEFL_PUT_BITS_FAST(d->m_huff_codes[0][lit], d->m_huff_code_sizes[0][lit]);
+        }
+      }
+    }
+
+    if (pOutput_buf >= d->m_pOutput_buf_end)
+      return MZ_FALSE;
+
+    *(mz_uint64*)pOutput_buf = bit_buffer;
+    pOutput_buf += (bits_in >> 3);
+    bit_buffer >>= (bits_in & ~7);
+    bits_in &= 7;
+  }
+
+#undef TDEFL_PUT_BITS_FAST
+
+  d->m_pOutput_buf = pOutput_buf;
+  d->m_bits_in = 0;
+  d->m_bit_buffer = 0;
+
+  while (bits_in)
+  {
+    mz_uint32 n = MZ_MIN(bits_in, 16);
+    TDEFL_PUT_BITS((mz_uint)bit_buffer & mz_bitmasks[n], n);
+    bit_buffer >>= n;
+    bits_in -= n;
+  }
+
+  TDEFL_PUT_BITS(d->m_huff_codes[0][256], d->m_huff_code_sizes[0][256]);
+
+  return (d->m_pOutput_buf < d->m_pOutput_buf_end);
+}
+#else
+static mz_bool tdefl_compress_lz_codes(tdefl_compressor *d)
+{
+  mz_uint flags;
+  mz_uint8 *pLZ_codes;
+
+  flags = 1;
+  for (pLZ_codes = d->m_lz_code_buf; pLZ_codes < d->m_pLZ_code_buf; flags >>= 1)
+  {
+    if (flags == 1)
+      flags = *pLZ_codes++ | 0x100;
+    if (flags & 1)
+    {
+      mz_uint sym, num_extra_bits;
+      mz_uint match_len = pLZ_codes[0], match_dist = (pLZ_codes[1] | (pLZ_codes[2] << 8)); pLZ_codes += 3;
+
+      MZ_ASSERT(d->m_huff_code_sizes[0][s_tdefl_len_sym[match_len]]);
+      TDEFL_PUT_BITS(d->m_huff_codes[0][s_tdefl_len_sym[match_len]], d->m_huff_code_sizes[0][s_tdefl_len_sym[match_len]]);
+      TDEFL_PUT_BITS(match_len & mz_bitmasks[s_tdefl_len_extra[match_len]], s_tdefl_len_extra[match_len]);
+
+      if (match_dist < 512)
+      {
+        sym = s_tdefl_small_dist_sym[match_dist]; num_extra_bits = s_tdefl_small_dist_extra[match_dist];
+      }
+      else
+      {
+        sym = s_tdefl_large_dist_sym[match_dist >> 8]; num_extra_bits = s_tdefl_large_dist_extra[match_dist >> 8];
+      }
+      MZ_ASSERT(d->m_huff_code_sizes[1][sym]);
+      TDEFL_PUT_BITS(d->m_huff_codes[1][sym], d->m_huff_code_sizes[1][sym]);
+      TDEFL_PUT_BITS(match_dist & mz_bitmasks[num_extra_bits], num_extra_bits);
+    }
+    else
+    {
+      mz_uint lit = *pLZ_codes++;
+      MZ_ASSERT(d->m_huff_code_sizes[0][lit]);
+      TDEFL_PUT_BITS(d->m_huff_codes[0][lit], d->m_huff_code_sizes[0][lit]);
+    }
+  }
+
+  TDEFL_PUT_BITS(d->m_huff_codes[0][256], d->m_huff_code_sizes[0][256]);
+
+  return (d->m_pOutput_buf < d->m_pOutput_buf_end);
+}
+#endif // MINIZ_USE_UNALIGNED_LOADS_AND_STORES && MINIZ_LITTLE_ENDIAN && MINIZ_HAS_64BIT_REGISTERS
+
+static mz_bool tdefl_compress_block(tdefl_compressor *d, mz_bool static_block)
+{
+  if (static_block)
+    tdefl_start_static_block(d);
+  else
+    tdefl_start_dynamic_block(d);
+  return tdefl_compress_lz_codes(d);
+}
+
+static int tdefl_flush_block(tdefl_compressor *d, int flush)
+{
+  mz_uint saved_bit_buf, saved_bits_in;
+  mz_uint8 *pSaved_output_buf;
+  mz_bool comp_block_succeeded = MZ_FALSE;
+  int n, use_raw_block = ((d->m_flags & TDEFL_FORCE_ALL_RAW_BLOCKS) != 0) && (d->m_lookahead_pos - d->m_lz_code_buf_dict_pos) <= d->m_dict_size;
+  mz_uint8 *pOutput_buf_start = ((d->m_pPut_buf_func == NULL) && ((*d->m_pOut_buf_size - d->m_out_buf_ofs) >= TDEFL_OUT_BUF_SIZE)) ? ((mz_uint8 *)d->m_pOut_buf + d->m_out_buf_ofs) : d->m_output_buf;
+
+  d->m_pOutput_buf = pOutput_buf_start;
+  d->m_pOutput_buf_end = d->m_pOutput_buf + TDEFL_OUT_BUF_SIZE - 16;
+
+  MZ_ASSERT(!d->m_output_flush_remaining);
+  d->m_output_flush_ofs = 0;
+  d->m_output_flush_remaining = 0;
+
+  *d->m_pLZ_flags = (mz_uint8)(*d->m_pLZ_flags >> d->m_num_flags_left);
+  d->m_pLZ_code_buf -= (d->m_num_flags_left == 8);
+
+  if ((d->m_flags & TDEFL_WRITE_ZLIB_HEADER) && (!d->m_block_index))
+  {
+    TDEFL_PUT_BITS(0x78, 8); TDEFL_PUT_BITS(0x01, 8);
+  }
+
+  TDEFL_PUT_BITS(flush == TDEFL_FINISH, 1);
+
+  pSaved_output_buf = d->m_pOutput_buf; saved_bit_buf = d->m_bit_buffer; saved_bits_in = d->m_bits_in;
+
+  if (!use_raw_block)
+    comp_block_succeeded = tdefl_compress_block(d, (d->m_flags & TDEFL_FORCE_ALL_STATIC_BLOCKS) || (d->m_total_lz_bytes < 48));
+
+  // If the block gets expanded, forget the current contents of the output buffer and send a raw block instead.
+  if ( ((use_raw_block) || ((d->m_total_lz_bytes) && ((d->m_pOutput_buf - pSaved_output_buf + 1U) >= d->m_total_lz_bytes))) &&
+       ((d->m_lookahead_pos - d->m_lz_code_buf_dict_pos) <= d->m_dict_size) )
+  {
+    mz_uint i; d->m_pOutput_buf = pSaved_output_buf; d->m_bit_buffer = saved_bit_buf, d->m_bits_in = saved_bits_in;
+    TDEFL_PUT_BITS(0, 2);
+    if (d->m_bits_in) { TDEFL_PUT_BITS(0, 8 - d->m_bits_in); }
+    for (i = 2; i; --i, d->m_total_lz_bytes ^= 0xFFFF)
+    {
+      TDEFL_PUT_BITS(d->m_total_lz_bytes & 0xFFFF, 16);
+    }
+    for (i = 0; i < d->m_total_lz_bytes; ++i)
+    {
+      TDEFL_PUT_BITS(d->m_dict[(d->m_lz_code_buf_dict_pos + i) & TDEFL_LZ_DICT_SIZE_MASK], 8);
+    }
+  }
+  // Check for the extremely unlikely (if not impossible) case of the compressed block not fitting into the output buffer when using dynamic codes.
+  else if (!comp_block_succeeded)
+  {
+    d->m_pOutput_buf = pSaved_output_buf; d->m_bit_buffer = saved_bit_buf, d->m_bits_in = saved_bits_in;
+    tdefl_compress_block(d, MZ_TRUE);
+  }
+
+  if (flush)
+  {
+    if (flush == TDEFL_FINISH)
+    {
+      if (d->m_bits_in) { TDEFL_PUT_BITS(0, 8 - d->m_bits_in); }
+      if (d->m_flags & TDEFL_WRITE_ZLIB_HEADER) { mz_uint i, a = d->m_adler32; for (i = 0; i < 4; i++) { TDEFL_PUT_BITS((a >> 24) & 0xFF, 8); a <<= 8; } }
+    }
+    else
+    {
+      mz_uint i, z = 0; TDEFL_PUT_BITS(0, 3); if (d->m_bits_in) { TDEFL_PUT_BITS(0, 8 - d->m_bits_in); } for (i = 2; i; --i, z ^= 0xFFFF) { TDEFL_PUT_BITS(z & 0xFFFF, 16); }
+    }
+  }
+
+  MZ_ASSERT(d->m_pOutput_buf < d->m_pOutput_buf_end);
+
+  memset(&d->m_huff_count[0][0], 0, sizeof(d->m_huff_count[0][0]) * TDEFL_MAX_HUFF_SYMBOLS_0);
+  memset(&d->m_huff_count[1][0], 0, sizeof(d->m_huff_count[1][0]) * TDEFL_MAX_HUFF_SYMBOLS_1);
+
+  d->m_pLZ_code_buf = d->m_lz_code_buf + 1; d->m_pLZ_flags = d->m_lz_code_buf; d->m_num_flags_left = 8; d->m_lz_code_buf_dict_pos += d->m_total_lz_bytes; d->m_total_lz_bytes = 0; d->m_block_index++;
+
+  if ((n = (int)(d->m_pOutput_buf - pOutput_buf_start)) != 0)
+  {
+    if (d->m_pPut_buf_func)
+    {
+      *d->m_pIn_buf_size = d->m_pSrc - (const mz_uint8 *)d->m_pIn_buf;
+      if (!(*d->m_pPut_buf_func)(d->m_output_buf, n, d->m_pPut_buf_user))
+        return (d->m_prev_return_status = TDEFL_STATUS_PUT_BUF_FAILED);
+    }
+    else if (pOutput_buf_start == d->m_output_buf)
+    {
+      int bytes_to_copy = (int)MZ_MIN((size_t)n, (size_t)(*d->m_pOut_buf_size - d->m_out_buf_ofs));
+      memcpy((mz_uint8 *)d->m_pOut_buf + d->m_out_buf_ofs, d->m_output_buf, bytes_to_copy);
+      d->m_out_buf_ofs += bytes_to_copy;
+      if ((n -= bytes_to_copy) != 0)
+      {
+        d->m_output_flush_ofs = bytes_to_copy;
+        d->m_output_flush_remaining = n;
+      }
+    }
+    else
+    {
+      d->m_out_buf_ofs += n;
+    }
+  }
+
+  return d->m_output_flush_remaining;
+}
+
+#if MINIZ_USE_UNALIGNED_LOADS_AND_STORES
+#define TDEFL_READ_UNALIGNED_WORD(p) *(const mz_uint16*)(p)
+static MZ_FORCEINLINE void tdefl_find_match(tdefl_compressor *d, mz_uint lookahead_pos, mz_uint max_dist, mz_uint max_match_len, mz_uint *pMatch_dist, mz_uint *pMatch_len)
+{
+  mz_uint dist, pos = lookahead_pos & TDEFL_LZ_DICT_SIZE_MASK, match_len = *pMatch_len, probe_pos = pos, next_probe_pos, probe_len;
+  mz_uint num_probes_left = d->m_max_probes[match_len >= 32];
+  const mz_uint16 *s = (const mz_uint16*)(d->m_dict + pos), *p, *q;
+  mz_uint16 c01 = TDEFL_READ_UNALIGNED_WORD(&d->m_dict[pos + match_len - 1]), s01 = TDEFL_READ_UNALIGNED_WORD(s);
+  MZ_ASSERT(max_match_len <= TDEFL_MAX_MATCH_LEN); if (max_match_len <= match_len) return;
+  for ( ; ; )
+  {
+    for ( ; ; )
+    {
+      if (--num_probes_left == 0) return;
+      #define TDEFL_PROBE \
+        next_probe_pos = d->m_next[probe_pos]; \
+        if ((!next_probe_pos) || ((dist = (mz_uint16)(lookahead_pos - next_probe_pos)) > max_dist)) return; \
+        probe_pos = next_probe_pos & TDEFL_LZ_DICT_SIZE_MASK; \
+        if (TDEFL_READ_UNALIGNED_WORD(&d->m_dict[probe_pos + match_len - 1]) == c01) break;
+      TDEFL_PROBE; TDEFL_PROBE; TDEFL_PROBE;
+    }
+    if (!dist) break; q = (const mz_uint16*)(d->m_dict + probe_pos); if (TDEFL_READ_UNALIGNED_WORD(q) != s01) continue; p = s; probe_len = 32;
+    do { } while ( (TDEFL_READ_UNALIGNED_WORD(++p) == TDEFL_READ_UNALIGNED_WORD(++q)) && (TDEFL_READ_UNALIGNED_WORD(++p) == TDEFL_READ_UNALIGNED_WORD(++q)) &&
+                   (TDEFL_READ_UNALIGNED_WORD(++p) == TDEFL_READ_UNALIGNED_WORD(++q)) && (TDEFL_READ_UNALIGNED_WORD(++p) == TDEFL_READ_UNALIGNED_WORD(++q)) && (--probe_len > 0) );
+    if (!probe_len)
+    {
+      *pMatch_dist = dist; *pMatch_len = MZ_MIN(max_match_len, TDEFL_MAX_MATCH_LEN); break;
+    }
+    else if ((probe_len = ((mz_uint)(p - s) * 2) + (mz_uint)(*(const mz_uint8*)p == *(const mz_uint8*)q)) > match_len)
+    {
+      *pMatch_dist = dist; if ((*pMatch_len = match_len = MZ_MIN(max_match_len, probe_len)) == max_match_len) break;
+      c01 = TDEFL_READ_UNALIGNED_WORD(&d->m_dict[pos + match_len - 1]);
+    }
+  }
+}
+#else
+static MZ_FORCEINLINE void tdefl_find_match(tdefl_compressor *d, mz_uint lookahead_pos, mz_uint max_dist, mz_uint max_match_len, mz_uint *pMatch_dist, mz_uint *pMatch_len)
+{
+  mz_uint dist, pos = lookahead_pos & TDEFL_LZ_DICT_SIZE_MASK, match_len = *pMatch_len, probe_pos = pos, next_probe_pos, probe_len;
+  mz_uint num_probes_left = d->m_max_probes[match_len >= 32];
+  const mz_uint8 *s = d->m_dict + pos, *p, *q;
+  mz_uint8 c0 = d->m_dict[pos + match_len], c1 = d->m_dict[pos + match_len - 1];
+  MZ_ASSERT(max_match_len <= TDEFL_MAX_MATCH_LEN); if (max_match_len <= match_len) return;
+  for ( ; ; )
+  {
+    for ( ; ; )
+    {
+      if (--num_probes_left == 0) return;
+      #define TDEFL_PROBE \
+        next_probe_pos = d->m_next[probe_pos]; \
+        if ((!next_probe_pos) || ((dist = (mz_uint16)(lookahead_pos - next_probe_pos)) > max_dist)) return; \
+        probe_pos = next_probe_pos & TDEFL_LZ_DICT_SIZE_MASK; \
+        if ((d->m_dict[probe_pos + match_len] == c0) && (d->m_dict[probe_pos + match_len - 1] == c1)) break;
+      TDEFL_PROBE; TDEFL_PROBE; TDEFL_PROBE;
+    }
+    if (!dist) break; p = s; q = d->m_dict + probe_pos; for (probe_len = 0; probe_len < max_match_len; probe_len++) if (*p++ != *q++) break;
+    if (probe_len > match_len)
+    {
+      *pMatch_dist = dist; if ((*pMatch_len = match_len = probe_len) == max_match_len) return;
+      c0 = d->m_dict[pos + match_len]; c1 = d->m_dict[pos + match_len - 1];
+    }
+  }
+}
+#endif // #if MINIZ_USE_UNALIGNED_LOADS_AND_STORES
+
+#if MINIZ_USE_UNALIGNED_LOADS_AND_STORES && MINIZ_LITTLE_ENDIAN
+static mz_bool tdefl_compress_fast(tdefl_compressor *d)
+{
+  // Faster, minimally featured LZRW1-style match+parse loop with better register utilization. Intended for applications where raw throughput is valued more highly than ratio.
+  mz_uint lookahead_pos = d->m_lookahead_pos, lookahead_size = d->m_lookahead_size, dict_size = d->m_dict_size, total_lz_bytes = d->m_total_lz_bytes, num_flags_left = d->m_num_flags_left;
+  mz_uint8 *pLZ_code_buf = d->m_pLZ_code_buf, *pLZ_flags = d->m_pLZ_flags;
+  mz_uint cur_pos = lookahead_pos & TDEFL_LZ_DICT_SIZE_MASK;
+
+  while ((d->m_src_buf_left) || ((d->m_flush) && (lookahead_size)))
+  {
+    const mz_uint TDEFL_COMP_FAST_LOOKAHEAD_SIZE = 4096;
+    mz_uint dst_pos = (lookahead_pos + lookahead_size) & TDEFL_LZ_DICT_SIZE_MASK;
+    mz_uint num_bytes_to_process = (mz_uint)MZ_MIN(d->m_src_buf_left, TDEFL_COMP_FAST_LOOKAHEAD_SIZE - lookahead_size);
+    d->m_src_buf_left -= num_bytes_to_process;
+    lookahead_size += num_bytes_to_process;
+
+    while (num_bytes_to_process)
+    {
+      mz_uint32 n = MZ_MIN(TDEFL_LZ_DICT_SIZE - dst_pos, num_bytes_to_process);
+      memcpy(d->m_dict + dst_pos, d->m_pSrc, n);
+      if (dst_pos < (TDEFL_MAX_MATCH_LEN - 1))
+        memcpy(d->m_dict + TDEFL_LZ_DICT_SIZE + dst_pos, d->m_pSrc, MZ_MIN(n, (TDEFL_MAX_MATCH_LEN - 1) - dst_pos));
+      d->m_pSrc += n;
+      dst_pos = (dst_pos + n) & TDEFL_LZ_DICT_SIZE_MASK;
+      num_bytes_to_process -= n;
+    }
+
+    dict_size = MZ_MIN(TDEFL_LZ_DICT_SIZE - lookahead_size, dict_size);
+    if ((!d->m_flush) && (lookahead_size < TDEFL_COMP_FAST_LOOKAHEAD_SIZE)) break;
+
+    while (lookahead_size >= 4)
+    {
+      mz_uint cur_match_dist, cur_match_len = 1;
+      mz_uint8 *pCur_dict = d->m_dict + cur_pos;
+      mz_uint first_trigram = (*(const mz_uint32 *)pCur_dict) & 0xFFFFFF;
+      mz_uint hash = (first_trigram ^ (first_trigram >> (24 - (TDEFL_LZ_HASH_BITS - 8)))) & TDEFL_LEVEL1_HASH_SIZE_MASK;
+      mz_uint probe_pos = d->m_hash[hash];
+      d->m_hash[hash] = (mz_uint16)lookahead_pos;
+
+      if (((cur_match_dist = (mz_uint16)(lookahead_pos - probe_pos)) <= dict_size) && ((*(const mz_uint32 *)(d->m_dict + (probe_pos &= TDEFL_LZ_DICT_SIZE_MASK)) & 0xFFFFFF) == first_trigram))
+      {
+        const mz_uint16 *p = (const mz_uint16 *)pCur_dict;
+        const mz_uint16 *q = (const mz_uint16 *)(d->m_dict + probe_pos);
+        mz_uint32 probe_len = 32;
+        do { } while ( (TDEFL_READ_UNALIGNED_WORD(++p) == TDEFL_READ_UNALIGNED_WORD(++q)) && (TDEFL_READ_UNALIGNED_WORD(++p) == TDEFL_READ_UNALIGNED_WORD(++q)) &&
+          (TDEFL_READ_UNALIGNED_WORD(++p) == TDEFL_READ_UNALIGNED_WORD(++q)) && (TDEFL_READ_UNALIGNED_WORD(++p) == TDEFL_READ_UNALIGNED_WORD(++q)) && (--probe_len > 0) );
+        cur_match_len = ((mz_uint)(p - (const mz_uint16 *)pCur_dict) * 2) + (mz_uint)(*(const mz_uint8 *)p == *(const mz_uint8 *)q);
+        if (!probe_len)
+          cur_match_len = cur_match_dist ? TDEFL_MAX_MATCH_LEN : 0;
+
+        if ((cur_match_len < TDEFL_MIN_MATCH_LEN) || ((cur_match_len == TDEFL_MIN_MATCH_LEN) && (cur_match_dist >= 8U*1024U)))
+        {
+          cur_match_len = 1;
+          *pLZ_code_buf++ = (mz_uint8)first_trigram;
+          *pLZ_flags = (mz_uint8)(*pLZ_flags >> 1);
+          d->m_huff_count[0][(mz_uint8)first_trigram]++;
+        }
+        else
+        {
+          mz_uint32 s0, s1;
+          cur_match_len = MZ_MIN(cur_match_len, lookahead_size);
+
+          MZ_ASSERT((cur_match_len >= TDEFL_MIN_MATCH_LEN) && (cur_match_dist >= 1) && (cur_match_dist <= TDEFL_LZ_DICT_SIZE));
+
+          cur_match_dist--;
+
+          pLZ_code_buf[0] = (mz_uint8)(cur_match_len - TDEFL_MIN_MATCH_LEN);
+          *(mz_uint16 *)(&pLZ_code_buf[1]) = (mz_uint16)cur_match_dist;
+          pLZ_code_buf += 3;
+          *pLZ_flags = (mz_uint8)((*pLZ_flags >> 1) | 0x80);
+
+          s0 = s_tdefl_small_dist_sym[cur_match_dist & 511];
+          s1 = s_tdefl_large_dist_sym[cur_match_dist >> 8];
+          d->m_huff_count[1][(cur_match_dist < 512) ? s0 : s1]++;
+
+          d->m_huff_count[0][s_tdefl_len_sym[cur_match_len - TDEFL_MIN_MATCH_LEN]]++;
+        }
+      }
+      else
+      {
+        *pLZ_code_buf++ = (mz_uint8)first_trigram;
+        *pLZ_flags = (mz_uint8)(*pLZ_flags >> 1);
+        d->m_huff_count[0][(mz_uint8)first_trigram]++;
+      }
+
+      if (--num_flags_left == 0) { num_flags_left = 8; pLZ_flags = pLZ_code_buf++; }
+
+      total_lz_bytes += cur_match_len;
+      lookahead_pos += cur_match_len;
+      dict_size = MZ_MIN(dict_size + cur_match_len, TDEFL_LZ_DICT_SIZE);
+      cur_pos = (cur_pos + cur_match_len) & TDEFL_LZ_DICT_SIZE_MASK;
+      MZ_ASSERT(lookahead_size >= cur_match_len);
+      lookahead_size -= cur_match_len;
+
+      if (pLZ_code_buf > &d->m_lz_code_buf[TDEFL_LZ_CODE_BUF_SIZE - 8])
+      {
+        int n;
+        d->m_lookahead_pos = lookahead_pos; d->m_lookahead_size = lookahead_size; d->m_dict_size = dict_size;
+        d->m_total_lz_bytes = total_lz_bytes; d->m_pLZ_code_buf = pLZ_code_buf; d->m_pLZ_flags = pLZ_flags; d->m_num_flags_left = num_flags_left;
+        if ((n = tdefl_flush_block(d, 0)) != 0)
+          return (n < 0) ? MZ_FALSE : MZ_TRUE;
+        total_lz_bytes = d->m_total_lz_bytes; pLZ_code_buf = d->m_pLZ_code_buf; pLZ_flags = d->m_pLZ_flags; num_flags_left = d->m_num_flags_left;
+      }
+    }
+
+    while (lookahead_size)
+    {
+      mz_uint8 lit = d->m_dict[cur_pos];
+
+      total_lz_bytes++;
+      *pLZ_code_buf++ = lit;
+      *pLZ_flags = (mz_uint8)(*pLZ_flags >> 1);
+      if (--num_flags_left == 0) { num_flags_left = 8; pLZ_flags = pLZ_code_buf++; }
+
+      d->m_huff_count[0][lit]++;
+
+      lookahead_pos++;
+      dict_size = MZ_MIN(dict_size + 1, TDEFL_LZ_DICT_SIZE);
+      cur_pos = (cur_pos + 1) & TDEFL_LZ_DICT_SIZE_MASK;
+      lookahead_size--;
+
+      if (pLZ_code_buf > &d->m_lz_code_buf[TDEFL_LZ_CODE_BUF_SIZE - 8])
+      {
+        int n;
+        d->m_lookahead_pos = lookahead_pos; d->m_lookahead_size = lookahead_size; d->m_dict_size = dict_size;
+        d->m_total_lz_bytes = total_lz_bytes; d->m_pLZ_code_buf = pLZ_code_buf; d->m_pLZ_flags = pLZ_flags; d->m_num_flags_left = num_flags_left;
+        if ((n = tdefl_flush_block(d, 0)) != 0)
+          return (n < 0) ? MZ_FALSE : MZ_TRUE;
+        total_lz_bytes = d->m_total_lz_bytes; pLZ_code_buf = d->m_pLZ_code_buf; pLZ_flags = d->m_pLZ_flags; num_flags_left = d->m_num_flags_left;
+      }
+    }
+  }
+
+  d->m_lookahead_pos = lookahead_pos; d->m_lookahead_size = lookahead_size; d->m_dict_size = dict_size;
+  d->m_total_lz_bytes = total_lz_bytes; d->m_pLZ_code_buf = pLZ_code_buf; d->m_pLZ_flags = pLZ_flags; d->m_num_flags_left = num_flags_left;
+  return MZ_TRUE;
+}
+#endif // MINIZ_USE_UNALIGNED_LOADS_AND_STORES && MINIZ_LITTLE_ENDIAN
+
+static MZ_FORCEINLINE void tdefl_record_literal(tdefl_compressor *d, mz_uint8 lit)
+{
+  d->m_total_lz_bytes++;
+  *d->m_pLZ_code_buf++ = lit;
+  *d->m_pLZ_flags = (mz_uint8)(*d->m_pLZ_flags >> 1); if (--d->m_num_flags_left == 0) { d->m_num_flags_left = 8; d->m_pLZ_flags = d->m_pLZ_code_buf++; }
+  d->m_huff_count[0][lit]++;
+}
+
+static MZ_FORCEINLINE void tdefl_record_match(tdefl_compressor *d, mz_uint match_len, mz_uint match_dist)
+{
+  mz_uint32 s0, s1;
+
+  MZ_ASSERT((match_len >= TDEFL_MIN_MATCH_LEN) && (match_dist >= 1) && (match_dist <= TDEFL_LZ_DICT_SIZE));
+
+  d->m_total_lz_bytes += match_len;
+
+  d->m_pLZ_code_buf[0] = (mz_uint8)(match_len - TDEFL_MIN_MATCH_LEN);
+
+  match_dist -= 1;
+  d->m_pLZ_code_buf[1] = (mz_uint8)(match_dist & 0xFF);
+  d->m_pLZ_code_buf[2] = (mz_uint8)(match_dist >> 8); d->m_pLZ_code_buf += 3;
+
+  *d->m_pLZ_flags = (mz_uint8)((*d->m_pLZ_flags >> 1) | 0x80); if (--d->m_num_flags_left == 0) { d->m_num_flags_left = 8; d->m_pLZ_flags = d->m_pLZ_code_buf++; }
+
+  s0 = s_tdefl_small_dist_sym[match_dist & 511]; s1 = s_tdefl_large_dist_sym[(match_dist >> 8) & 127];
+  d->m_huff_count[1][(match_dist < 512) ? s0 : s1]++;
+
+  if (match_len >= TDEFL_MIN_MATCH_LEN) d->m_huff_count[0][s_tdefl_len_sym[match_len - TDEFL_MIN_MATCH_LEN]]++;
+}
+
+static mz_bool tdefl_compress_normal(tdefl_compressor *d)
+{
+  const mz_uint8 *pSrc = d->m_pSrc; size_t src_buf_left = d->m_src_buf_left;
+  tdefl_flush flush = d->m_flush;
+
+  while ((src_buf_left) || ((flush) && (d->m_lookahead_size)))
+  {
+    mz_uint len_to_move, cur_match_dist, cur_match_len, cur_pos;
+    // Update dictionary and hash chains. Keeps the lookahead size equal to TDEFL_MAX_MATCH_LEN.
+    if ((d->m_lookahead_size + d->m_dict_size) >= (TDEFL_MIN_MATCH_LEN - 1))
+    {
+      mz_uint dst_pos = (d->m_lookahead_pos + d->m_lookahead_size) & TDEFL_LZ_DICT_SIZE_MASK, ins_pos = d->m_lookahead_pos + d->m_lookahead_size - 2;
+      mz_uint hash = (d->m_dict[ins_pos & TDEFL_LZ_DICT_SIZE_MASK] << TDEFL_LZ_HASH_SHIFT) ^ d->m_dict[(ins_pos + 1) & TDEFL_LZ_DICT_SIZE_MASK];
+      mz_uint num_bytes_to_process = (mz_uint)MZ_MIN(src_buf_left, TDEFL_MAX_MATCH_LEN - d->m_lookahead_size);
+      const mz_uint8 *pSrc_end = pSrc + num_bytes_to_process;
+      src_buf_left -= num_bytes_to_process;
+      d->m_lookahead_size += num_bytes_to_process;
+      while (pSrc != pSrc_end)
+      {
+        mz_uint8 c = *pSrc++; d->m_dict[dst_pos] = c; if (dst_pos < (TDEFL_MAX_MATCH_LEN - 1)) d->m_dict[TDEFL_LZ_DICT_SIZE + dst_pos] = c;
+        hash = ((hash << TDEFL_LZ_HASH_SHIFT) ^ c) & (TDEFL_LZ_HASH_SIZE - 1);
+        d->m_next[ins_pos & TDEFL_LZ_DICT_SIZE_MASK] = d->m_hash[hash]; d->m_hash[hash] = (mz_uint16)(ins_pos);
+        dst_pos = (dst_pos + 1) & TDEFL_LZ_DICT_SIZE_MASK; ins_pos++;
+      }
+    }
+    else
+    {
+      while ((src_buf_left) && (d->m_lookahead_size < TDEFL_MAX_MATCH_LEN))
+      {
+        mz_uint8 c = *pSrc++;
+        mz_uint dst_pos = (d->m_lookahead_pos + d->m_lookahead_size) & TDEFL_LZ_DICT_SIZE_MASK;
+        src_buf_left--;
+        d->m_dict[dst_pos] = c;
+        if (dst_pos < (TDEFL_MAX_MATCH_LEN - 1))
+          d->m_dict[TDEFL_LZ_DICT_SIZE + dst_pos] = c;
+        if ((++d->m_lookahead_size + d->m_dict_size) >= TDEFL_MIN_MATCH_LEN)
+        {
+          mz_uint ins_pos = d->m_lookahead_pos + (d->m_lookahead_size - 1) - 2;
+          mz_uint hash = ((d->m_dict[ins_pos & TDEFL_LZ_DICT_SIZE_MASK] << (TDEFL_LZ_HASH_SHIFT * 2)) ^ (d->m_dict[(ins_pos + 1) & TDEFL_LZ_DICT_SIZE_MASK] << TDEFL_LZ_HASH_SHIFT) ^ c) & (TDEFL_LZ_HASH_SIZE - 1);
+          d->m_next[ins_pos & TDEFL_LZ_DICT_SIZE_MASK] = d->m_hash[hash]; d->m_hash[hash] = (mz_uint16)(ins_pos);
+        }
+      }
+    }
+    d->m_dict_size = MZ_MIN(TDEFL_LZ_DICT_SIZE - d->m_lookahead_size, d->m_dict_size);
+    if ((!flush) && (d->m_lookahead_size < TDEFL_MAX_MATCH_LEN))
+      break;
+
+    // Simple lazy/greedy parsing state machine.
+    len_to_move = 1; cur_match_dist = 0; cur_match_len = d->m_saved_match_len ? d->m_saved_match_len : (TDEFL_MIN_MATCH_LEN - 1); cur_pos = d->m_lookahead_pos & TDEFL_LZ_DICT_SIZE_MASK;
+    if (d->m_flags & (TDEFL_RLE_MATCHES | TDEFL_FORCE_ALL_RAW_BLOCKS))
+    {
+      if ((d->m_dict_size) && (!(d->m_flags & TDEFL_FORCE_ALL_RAW_BLOCKS)))
+      {
+        mz_uint8 c = d->m_dict[(cur_pos - 1) & TDEFL_LZ_DICT_SIZE_MASK];
+        cur_match_len = 0; while (cur_match_len < d->m_lookahead_size) { if (d->m_dict[cur_pos + cur_match_len] != c) break; cur_match_len++; }
+        if (cur_match_len < TDEFL_MIN_MATCH_LEN) cur_match_len = 0; else cur_match_dist = 1;
+      }
+    }
+    else
+    {
+      tdefl_find_match(d, d->m_lookahead_pos, d->m_dict_size, d->m_lookahead_size, &cur_match_dist, &cur_match_len);
+    }
+    if (((cur_match_len == TDEFL_MIN_MATCH_LEN) && (cur_match_dist >= 8U*1024U)) || (cur_pos == cur_match_dist) || ((d->m_flags & TDEFL_FILTER_MATCHES) && (cur_match_len <= 5)))
+    {
+      cur_match_dist = cur_match_len = 0;
+    }
+    if (d->m_saved_match_len)
+    {
+      if (cur_match_len > d->m_saved_match_len)
+      {
+        tdefl_record_literal(d, (mz_uint8)d->m_saved_lit);
+        if (cur_match_len >= 128)
+        {
+          tdefl_record_match(d, cur_match_len, cur_match_dist);
+          d->m_saved_match_len = 0; len_to_move = cur_match_len;
+        }
+        else
+        {
+          d->m_saved_lit = d->m_dict[cur_pos]; d->m_saved_match_dist = cur_match_dist; d->m_saved_match_len = cur_match_len;
+        }
+      }
+      else
+      {
+        tdefl_record_match(d, d->m_saved_match_len, d->m_saved_match_dist);
+        len_to_move = d->m_saved_match_len - 1; d->m_saved_match_len = 0;
+      }
+    }
+    else if (!cur_match_dist)
+      tdefl_record_literal(d, d->m_dict[MZ_MIN(cur_pos, sizeof(d->m_dict) - 1)]);
+    else if ((d->m_greedy_parsing) || (d->m_flags & TDEFL_RLE_MATCHES) || (cur_match_len >= 128))
+    {
+      tdefl_record_match(d, cur_match_len, cur_match_dist);
+      len_to_move = cur_match_len;
+    }
+    else
+    {
+      d->m_saved_lit = d->m_dict[MZ_MIN(cur_pos, sizeof(d->m_dict) - 1)]; d->m_saved_match_dist = cur_match_dist; d->m_saved_match_len = cur_match_len;
+    }
+    // Move the lookahead forward by len_to_move bytes.
+    d->m_lookahead_pos += len_to_move;
+    MZ_ASSERT(d->m_lookahead_size >= len_to_move);
+    d->m_lookahead_size -= len_to_move;
+    d->m_dict_size = MZ_MIN(d->m_dict_size + len_to_move, TDEFL_LZ_DICT_SIZE);
+    // Check if it's time to flush the current LZ codes to the internal output buffer.
+    if ( (d->m_pLZ_code_buf > &d->m_lz_code_buf[TDEFL_LZ_CODE_BUF_SIZE - 8]) ||
+         ( (d->m_total_lz_bytes > 31*1024) && (((((mz_uint)(d->m_pLZ_code_buf - d->m_lz_code_buf) * 115) >> 7) >= d->m_total_lz_bytes) || (d->m_flags & TDEFL_FORCE_ALL_RAW_BLOCKS))) )
+    {
+      int n;
+      d->m_pSrc = pSrc; d->m_src_buf_left = src_buf_left;
+      if ((n = tdefl_flush_block(d, 0)) != 0)
+        return (n < 0) ? MZ_FALSE : MZ_TRUE;
+    }
+  }
+
+  d->m_pSrc = pSrc; d->m_src_buf_left = src_buf_left;
+  return MZ_TRUE;
+}
+
+static tdefl_status tdefl_flush_output_buffer(tdefl_compressor *d)
+{
+  if (d->m_pIn_buf_size)
+  {
+    *d->m_pIn_buf_size = d->m_pSrc - (const mz_uint8 *)d->m_pIn_buf;
+  }
+
+  if (d->m_pOut_buf_size)
+  {
+    size_t n = MZ_MIN(*d->m_pOut_buf_size - d->m_out_buf_ofs, d->m_output_flush_remaining);
+    memcpy((mz_uint8 *)d->m_pOut_buf + d->m_out_buf_ofs, d->m_output_buf + d->m_output_flush_ofs, n);
+    d->m_output_flush_ofs += (mz_uint)n;
+    d->m_output_flush_remaining -= (mz_uint)n;
+    d->m_out_buf_ofs += n;
+
+    *d->m_pOut_buf_size = d->m_out_buf_ofs;
+  }
+
+  return (d->m_finished && !d->m_output_flush_remaining) ? TDEFL_STATUS_DONE : TDEFL_STATUS_OKAY;
+}
+
+tdefl_status tdefl_compress(tdefl_compressor *d, const void *pIn_buf, size_t *pIn_buf_size, void *pOut_buf, size_t *pOut_buf_size, tdefl_flush flush)
+{
+  if (!d)
+  {
+    if (pIn_buf_size) *pIn_buf_size = 0;
+    if (pOut_buf_size) *pOut_buf_size = 0;
+    return TDEFL_STATUS_BAD_PARAM;
+  }
+
+  d->m_pIn_buf = pIn_buf; d->m_pIn_buf_size = pIn_buf_size;
+  d->m_pOut_buf = pOut_buf; d->m_pOut_buf_size = pOut_buf_size;
+  d->m_pSrc = (const mz_uint8 *)(pIn_buf); d->m_src_buf_left = pIn_buf_size ? *pIn_buf_size : 0;
+  d->m_out_buf_ofs = 0;
+  d->m_flush = flush;
+
+  if ( ((d->m_pPut_buf_func != NULL) == ((pOut_buf != NULL) || (pOut_buf_size != NULL))) || (d->m_prev_return_status != TDEFL_STATUS_OKAY) ||
+        (d->m_wants_to_finish && (flush != TDEFL_FINISH)) || (pIn_buf_size && *pIn_buf_size && !pIn_buf) || (pOut_buf_size && *pOut_buf_size && !pOut_buf) )
+  {
+    if (pIn_buf_size) *pIn_buf_size = 0;
+    if (pOut_buf_size) *pOut_buf_size = 0;
+    return (d->m_prev_return_status = TDEFL_STATUS_BAD_PARAM);
+  }
+  d->m_wants_to_finish |= (flush == TDEFL_FINISH);
+
+  if ((d->m_output_flush_remaining) || (d->m_finished))
+    return (d->m_prev_return_status = tdefl_flush_output_buffer(d));
+
+#if MINIZ_USE_UNALIGNED_LOADS_AND_STORES && MINIZ_LITTLE_ENDIAN
+  if (((d->m_flags & TDEFL_MAX_PROBES_MASK) == 1) &&
+      ((d->m_flags & TDEFL_GREEDY_PARSING_FLAG) != 0) &&
+      ((d->m_flags & (TDEFL_FILTER_MATCHES | TDEFL_FORCE_ALL_RAW_BLOCKS | TDEFL_RLE_MATCHES)) == 0))
+  {
+    if (!tdefl_compress_fast(d))
+      return d->m_prev_return_status;
+  }
+  else
+#endif // #if MINIZ_USE_UNALIGNED_LOADS_AND_STORES && MINIZ_LITTLE_ENDIAN
+  {
+    if (!tdefl_compress_normal(d))
+      return d->m_prev_return_status;
+  }
+
+  if ((d->m_flags & (TDEFL_WRITE_ZLIB_HEADER | TDEFL_COMPUTE_ADLER32)) && (pIn_buf))
+    d->m_adler32 = (mz_uint32)mz_adler32(d->m_adler32, (const mz_uint8 *)pIn_buf, d->m_pSrc - (const mz_uint8 *)pIn_buf);
+
+  if ((flush) && (!d->m_lookahead_size) && (!d->m_src_buf_left) && (!d->m_output_flush_remaining))
+  {
+    if (tdefl_flush_block(d, flush) < 0)
+      return d->m_prev_return_status;
+    d->m_finished = (flush == TDEFL_FINISH);
+    if (flush == TDEFL_FULL_FLUSH) { MZ_CLEAR_OBJ(d->m_hash); MZ_CLEAR_OBJ(d->m_next); d->m_dict_size = 0; }
+  }
+
+  return (d->m_prev_return_status = tdefl_flush_output_buffer(d));
+}
+
+tdefl_status tdefl_compress_buffer(tdefl_compressor *d, const void *pIn_buf, size_t in_buf_size, tdefl_flush flush)
+{
+  MZ_ASSERT(d->m_pPut_buf_func); return tdefl_compress(d, pIn_buf, &in_buf_size, NULL, NULL, flush);
+}
+
+tdefl_status tdefl_init(tdefl_compressor *d, tdefl_put_buf_func_ptr pPut_buf_func, void *pPut_buf_user, int flags)
+{
+  d->m_pPut_buf_func = pPut_buf_func; d->m_pPut_buf_user = pPut_buf_user;
+  d->m_flags = (mz_uint)(flags); d->m_max_probes[0] = 1 + ((flags & 0xFFF) + 2) / 3; d->m_greedy_parsing = (flags & TDEFL_GREEDY_PARSING_FLAG) != 0;
+  d->m_max_probes[1] = 1 + (((flags & 0xFFF) >> 2) + 2) / 3;
+  if (!(flags & TDEFL_NONDETERMINISTIC_PARSING_FLAG)) MZ_CLEAR_OBJ(d->m_hash);
+  d->m_lookahead_pos = d->m_lookahead_size = d->m_dict_size = d->m_total_lz_bytes = d->m_lz_code_buf_dict_pos = d->m_bits_in = 0;
+  d->m_output_flush_ofs = d->m_output_flush_remaining = d->m_finished = d->m_block_index = d->m_bit_buffer = d->m_wants_to_finish = 0;
+  d->m_pLZ_code_buf = d->m_lz_code_buf + 1; d->m_pLZ_flags = d->m_lz_code_buf; d->m_num_flags_left = 8;
+  d->m_pOutput_buf = d->m_output_buf; d->m_pOutput_buf_end = d->m_output_buf; d->m_prev_return_status = TDEFL_STATUS_OKAY;
+  d->m_saved_match_dist = d->m_saved_match_len = d->m_saved_lit = 0; d->m_adler32 = 1;
+  d->m_pIn_buf = NULL; d->m_pOut_buf = NULL;
+  d->m_pIn_buf_size = NULL; d->m_pOut_buf_size = NULL;
+  d->m_flush = TDEFL_NO_FLUSH; d->m_pSrc = NULL; d->m_src_buf_left = 0; d->m_out_buf_ofs = 0;
+  memset(&d->m_huff_count[0][0], 0, sizeof(d->m_huff_count[0][0]) * TDEFL_MAX_HUFF_SYMBOLS_0);
+  memset(&d->m_huff_count[1][0], 0, sizeof(d->m_huff_count[1][0]) * TDEFL_MAX_HUFF_SYMBOLS_1);
+  return TDEFL_STATUS_OKAY;
+}
+
+tdefl_status tdefl_get_prev_return_status(tdefl_compressor *d)
+{
+  return d->m_prev_return_status;
+}
+
+mz_uint32 tdefl_get_adler32(tdefl_compressor *d)
+{
+  return d->m_adler32;
+}
+
+mz_bool tdefl_compress_mem_to_output(const void *pBuf, size_t buf_len, tdefl_put_buf_func_ptr pPut_buf_func, void *pPut_buf_user, int flags)
+{
+  tdefl_compressor *pComp; mz_bool succeeded; if (((buf_len) && (!pBuf)) || (!pPut_buf_func)) return MZ_FALSE;
+  pComp = (tdefl_compressor*)MZ_MALLOC(sizeof(tdefl_compressor)); if (!pComp) return MZ_FALSE;
+  succeeded = (tdefl_init(pComp, pPut_buf_func, pPut_buf_user, flags) == TDEFL_STATUS_OKAY);
+  succeeded = succeeded && (tdefl_compress_buffer(pComp, pBuf, buf_len, TDEFL_FINISH) == TDEFL_STATUS_DONE);
+  MZ_FREE(pComp); return succeeded;
+}
+
+typedef struct
+{
+  size_t m_size, m_capacity;
+  mz_uint8 *m_pBuf;
+  mz_bool m_expandable;
+} tdefl_output_buffer;
+
+static mz_bool tdefl_output_buffer_putter(const void *pBuf, int len, void *pUser)
+{
+  tdefl_output_buffer *p = (tdefl_output_buffer *)pUser;
+  size_t new_size = p->m_size + len;
+  if (new_size > p->m_capacity)
+  {
+    size_t new_capacity = p->m_capacity; mz_uint8 *pNew_buf; if (!p->m_expandable) return MZ_FALSE;
+    do { new_capacity = MZ_MAX(128U, new_capacity << 1U); } while (new_size > new_capacity);
+    pNew_buf = (mz_uint8*)MZ_REALLOC(p->m_pBuf, new_capacity); if (!pNew_buf) return MZ_FALSE;
+    p->m_pBuf = pNew_buf; p->m_capacity = new_capacity;
+  }
+  memcpy((mz_uint8*)p->m_pBuf + p->m_size, pBuf, len); p->m_size = new_size;
+  return MZ_TRUE;
+}
+
+void *tdefl_compress_mem_to_heap(const void *pSrc_buf, size_t src_buf_len, size_t *pOut_len, int flags)
+{
+  tdefl_output_buffer out_buf; MZ_CLEAR_OBJ(out_buf);
+  if (!pOut_len) return MZ_FALSE; else *pOut_len = 0;
+  out_buf.m_expandable = MZ_TRUE;
+  if (!tdefl_compress_mem_to_output(pSrc_buf, src_buf_len, tdefl_output_buffer_putter, &out_buf, flags)) return NULL;
+  *pOut_len = out_buf.m_size; return out_buf.m_pBuf;
+}
+
+size_t tdefl_compress_mem_to_mem(void *pOut_buf, size_t out_buf_len, const void *pSrc_buf, size_t src_buf_len, int flags)
+{
+  tdefl_output_buffer out_buf; MZ_CLEAR_OBJ(out_buf);
+  if (!pOut_buf) return 0;
+  out_buf.m_pBuf = (mz_uint8*)pOut_buf; out_buf.m_capacity = out_buf_len;
+  if (!tdefl_compress_mem_to_output(pSrc_buf, src_buf_len, tdefl_output_buffer_putter, &out_buf, flags)) return 0;
+  return out_buf.m_size;
+}
+
+#ifndef MINIZ_NO_ZLIB_APIS
+static const mz_uint s_tdefl_num_probes[11] = { 0, 1, 6, 32,  16, 32, 128, 256,  512, 768, 1500 };
+
+// level may actually range from [0,10] (10 is a "hidden" max level, where we want a bit more compression and it's fine if throughput to fall off a cliff on some files).
+mz_uint tdefl_create_comp_flags_from_zip_params(int level, int window_bits, int strategy)
+{
+  mz_uint comp_flags = s_tdefl_num_probes[(level >= 0) ? MZ_MIN(10, level) : MZ_DEFAULT_LEVEL] | ((level <= 3) ? TDEFL_GREEDY_PARSING_FLAG : 0);
+  if (window_bits > 0) comp_flags |= TDEFL_WRITE_ZLIB_HEADER;
+
+  if (!level) comp_flags |= TDEFL_FORCE_ALL_RAW_BLOCKS;
+  else if (strategy == MZ_FILTERED) comp_flags |= TDEFL_FILTER_MATCHES;
+  else if (strategy == MZ_HUFFMAN_ONLY) comp_flags &= ~TDEFL_MAX_PROBES_MASK;
+  else if (strategy == MZ_FIXED) comp_flags |= TDEFL_FORCE_ALL_STATIC_BLOCKS;
+  else if (strategy == MZ_RLE) comp_flags |= TDEFL_RLE_MATCHES;
+
+  return comp_flags;
+}
+#endif //MINIZ_NO_ZLIB_APIS
+
+#ifdef _MSC_VER
+#pragma warning (push)
+#pragma warning (disable:4204) // nonstandard extension used : non-constant aggregate initializer (also supported by GNU C and C99, so no big deal)
+#endif
+
+// Simple PNG writer function by Alex Evans, 2011. Released into the public domain: https://gist.github.com/908299, more context at
+// http://altdevblogaday.org/2011/04/06/a-smaller-jpg-encoder/.
+// This is actually a modification of Alex's original code so PNG files generated by this function pass pngcheck.
+void *tdefl_write_image_to_png_file_in_memory_ex(const void *pImage, int w, int h, int num_chans, size_t *pLen_out, mz_uint level, mz_bool flip)
+{
+  // Using a local copy of this array here in case MINIZ_NO_ZLIB_APIS was defined.
+  static const mz_uint s_tdefl_png_num_probes[11] = { 0, 1, 6, 32,  16, 32, 128, 256,  512, 768, 1500 };
+  tdefl_compressor *pComp = (tdefl_compressor *)MZ_MALLOC(sizeof(tdefl_compressor)); tdefl_output_buffer out_buf; int i, bpl = w * num_chans, y, z; mz_uint32 c; *pLen_out = 0;
+  if (!pComp) return NULL;
+  MZ_CLEAR_OBJ(out_buf); out_buf.m_expandable = MZ_TRUE; out_buf.m_capacity = 57+MZ_MAX(64, (1+bpl)*h); if (NULL == (out_buf.m_pBuf = (mz_uint8*)MZ_MALLOC(out_buf.m_capacity))) { MZ_FREE(pComp); return NULL; }
+  // write dummy header
+  for (z = 41; z; --z) tdefl_output_buffer_putter(&z, 1, &out_buf);
+  // compress image data
+  tdefl_init(pComp, tdefl_output_buffer_putter, &out_buf, s_tdefl_png_num_probes[MZ_MIN(10, level)] | TDEFL_WRITE_ZLIB_HEADER);
+  for (y = 0; y < h; ++y) { tdefl_compress_buffer(pComp, &z, 1, TDEFL_NO_FLUSH); tdefl_compress_buffer(pComp, (mz_uint8*)pImage + (flip ? (h - 1 - y) : y) * bpl, bpl, TDEFL_NO_FLUSH); }
+  if (tdefl_compress_buffer(pComp, NULL, 0, TDEFL_FINISH) != TDEFL_STATUS_DONE) { MZ_FREE(pComp); MZ_FREE(out_buf.m_pBuf); return NULL; }
+  // write real header
+  *pLen_out = out_buf.m_size-41;
+  {
+    static const mz_uint8 chans[] = {0x00, 0x00, 0x04, 0x02, 0x06};
+    mz_uint8 pnghdr[41]={0x89,0x50,0x4e,0x47,0x0d,0x0a,0x1a,0x0a,0x00,0x00,0x00,0x0d,0x49,0x48,0x44,0x52,
+      0,0,(mz_uint8)(w>>8),(mz_uint8)w,0,0,(mz_uint8)(h>>8),(mz_uint8)h,8,chans[num_chans],0,0,0,0,0,0,0,
+      (mz_uint8)(*pLen_out>>24),(mz_uint8)(*pLen_out>>16),(mz_uint8)(*pLen_out>>8),(mz_uint8)*pLen_out,0x49,0x44,0x41,0x54};
+    c=(mz_uint32)mz_crc32(MZ_CRC32_INIT,pnghdr+12,17); for (i=0; i<4; ++i, c<<=8) ((mz_uint8*)(pnghdr+29))[i]=(mz_uint8)(c>>24);
+    memcpy(out_buf.m_pBuf, pnghdr, 41);
+  }
+  // write footer (IDAT CRC-32, followed by IEND chunk)
+  if (!tdefl_output_buffer_putter("\0\0\0\0\0\0\0\0\x49\x45\x4e\x44\xae\x42\x60\x82", 16, &out_buf)) { *pLen_out = 0; MZ_FREE(pComp); MZ_FREE(out_buf.m_pBuf); return NULL; }
+  c = (mz_uint32)mz_crc32(MZ_CRC32_INIT,out_buf.m_pBuf+41-4, *pLen_out+4); for (i=0; i<4; ++i, c<<=8) (out_buf.m_pBuf+out_buf.m_size-16)[i] = (mz_uint8)(c >> 24);
+  // compute final size of file, grab compressed data buffer and return
+  *pLen_out += 57; MZ_FREE(pComp); return out_buf.m_pBuf;
+}
+void *tdefl_write_image_to_png_file_in_memory(const void *pImage, int w, int h, int num_chans, size_t *pLen_out)
+{
+  // Level 6 corresponds to TDEFL_DEFAULT_MAX_PROBES or MZ_DEFAULT_LEVEL (but we can't depend on MZ_DEFAULT_LEVEL being available in case the zlib API's where #defined out)
+  return tdefl_write_image_to_png_file_in_memory_ex(pImage, w, h, num_chans, pLen_out, 6, MZ_FALSE);
+}
+
+#ifdef _MSC_VER
+#pragma warning (pop)
+#endif
+
+// ------------------- .ZIP archive reading
+
+#ifndef MINIZ_NO_ARCHIVE_APIS
+
+#ifdef MINIZ_NO_STDIO
+  #define MZ_FILE void *
+#else
+  #include <stdio.h>
+  #include <sys/stat.h>
+
+  #if defined(_MSC_VER) || defined(__MINGW64__)
+    static FILE *mz_fopen(const char *pFilename, const char *pMode)
+    {
+      FILE* pFile = NULL;
+      fopen_s(&pFile, pFilename, pMode);
+      return pFile;
+    }
+    static FILE *mz_freopen(const char *pPath, const char *pMode, FILE *pStream)
+    {
+      FILE* pFile = NULL;
+      if (freopen_s(&pFile, pPath, pMode, pStream))
+        return NULL;
+      return pFile;
+    }
+    #ifndef MINIZ_NO_TIME
+      #include <sys/utime.h>
+    #endif
+    #define MZ_FILE FILE
+    #define MZ_FOPEN mz_fopen
+    #define MZ_FCLOSE fclose
+    #define MZ_FREAD fread
+    #define MZ_FWRITE fwrite
+    #define MZ_FTELL64 _ftelli64
+    #define MZ_FSEEK64 _fseeki64
+    #define MZ_FILE_STAT_STRUCT _stat
+    #define MZ_FILE_STAT _stat
+    #define MZ_FFLUSH fflush
+    #define MZ_FREOPEN mz_freopen
+    #define MZ_DELETE_FILE remove
+  #elif defined(__MINGW32__)
+    #ifndef MINIZ_NO_TIME
+      #include <sys/utime.h>
+    #endif
+    #define MZ_FILE FILE
+    #define MZ_FOPEN(f, m) fopen(f, m)
+    #define MZ_FCLOSE fclose
+    #define MZ_FREAD fread
+    #define MZ_FWRITE fwrite
+    #define MZ_FTELL64 ftello64
+    #define MZ_FSEEK64 fseeko64
+    #define MZ_FILE_STAT_STRUCT _stat
+    #define MZ_FILE_STAT _stat
+    #define MZ_FFLUSH fflush
+    #define MZ_FREOPEN(f, m, s) freopen(f, m, s)
+    #define MZ_DELETE_FILE remove
+  #elif defined(__TINYC__)
+    #ifndef MINIZ_NO_TIME
+      #include <sys/utime.h>
+    #endif
+    #define MZ_FILE FILE
+    #define MZ_FOPEN(f, m) fopen(f, m)
+    #define MZ_FCLOSE fclose
+    #define MZ_FREAD fread
+    #define MZ_FWRITE fwrite
+    #define MZ_FTELL64 ftell
+    #define MZ_FSEEK64 fseek
+    #define MZ_FILE_STAT_STRUCT stat
+    #define MZ_FILE_STAT stat
+    #define MZ_FFLUSH fflush
+    #define MZ_FREOPEN(f, m, s) freopen(f, m, s)
+    #define MZ_DELETE_FILE remove
+  #elif defined(__GNUC__) && _LARGEFILE64_SOURCE
+    #ifndef MINIZ_NO_TIME
+      #include <utime.h>
+    #endif
+    #define MZ_FILE FILE
+    #define MZ_FOPEN(f, m) fopen64(f, m)
+    #define MZ_FCLOSE fclose
+    #define MZ_FREAD fread
+    #define MZ_FWRITE fwrite
+    #define MZ_FTELL64 ftello64
+    #define MZ_FSEEK64 fseeko64
+    #define MZ_FILE_STAT_STRUCT stat64
+    #define MZ_FILE_STAT stat64
+    #define MZ_FFLUSH fflush
+    #define MZ_FREOPEN(p, m, s) freopen64(p, m, s)
+    #define MZ_DELETE_FILE remove
+  #else
+    #ifndef MINIZ_NO_TIME
+      #include <utime.h>
+    #endif
+    #define MZ_FILE FILE
+    #define MZ_FOPEN(f, m) fopen(f, m)
+    #define MZ_FCLOSE fclose
+    #define MZ_FREAD fread
+    #define MZ_FWRITE fwrite
+    #define MZ_FTELL64 ftello
+    #define MZ_FSEEK64 fseeko
+    #define MZ_FILE_STAT_STRUCT stat
+    #define MZ_FILE_STAT stat
+    #define MZ_FFLUSH fflush
+    #define MZ_FREOPEN(f, m, s) freopen(f, m, s)
+    #define MZ_DELETE_FILE remove
+  #endif // #ifdef _MSC_VER
+#endif // #ifdef MINIZ_NO_STDIO
+
+#define MZ_TOLOWER(c) ((((c) >= 'A') && ((c) <= 'Z')) ? ((c) - 'A' + 'a') : (c))
+
+// Various ZIP archive enums. To completely avoid cross platform compiler alignment and platform endian issues, miniz.c doesn't use structs for any of this stuff.
+enum
+{
+  // ZIP archive identifiers and record sizes
+  MZ_ZIP_END_OF_CENTRAL_DIR_HEADER_SIG = 0x06054b50, MZ_ZIP_CENTRAL_DIR_HEADER_SIG = 0x02014b50, MZ_ZIP_LOCAL_DIR_HEADER_SIG = 0x04034b50,
+  MZ_ZIP_LOCAL_DIR_HEADER_SIZE = 30, MZ_ZIP_CENTRAL_DIR_HEADER_SIZE = 46, MZ_ZIP_END_OF_CENTRAL_DIR_HEADER_SIZE = 22,
+  // Central directory header record offsets
+  MZ_ZIP_CDH_SIG_OFS = 0, MZ_ZIP_CDH_VERSION_MADE_BY_OFS = 4, MZ_ZIP_CDH_VERSION_NEEDED_OFS = 6, MZ_ZIP_CDH_BIT_FLAG_OFS = 8,
+  MZ_ZIP_CDH_METHOD_OFS = 10, MZ_ZIP_CDH_FILE_TIME_OFS = 12, MZ_ZIP_CDH_FILE_DATE_OFS = 14, MZ_ZIP_CDH_CRC32_OFS = 16,
+  MZ_ZIP_CDH_COMPRESSED_SIZE_OFS = 20, MZ_ZIP_CDH_DECOMPRESSED_SIZE_OFS = 24, MZ_ZIP_CDH_FILENAME_LEN_OFS = 28, MZ_ZIP_CDH_EXTRA_LEN_OFS = 30,
+  MZ_ZIP_CDH_COMMENT_LEN_OFS = 32, MZ_ZIP_CDH_DISK_START_OFS = 34, MZ_ZIP_CDH_INTERNAL_ATTR_OFS = 36, MZ_ZIP_CDH_EXTERNAL_ATTR_OFS = 38, MZ_ZIP_CDH_LOCAL_HEADER_OFS = 42,
+  // Local directory header offsets
+  MZ_ZIP_LDH_SIG_OFS = 0, MZ_ZIP_LDH_VERSION_NEEDED_OFS = 4, MZ_ZIP_LDH_BIT_FLAG_OFS = 6, MZ_ZIP_LDH_METHOD_OFS = 8, MZ_ZIP_LDH_FILE_TIME_OFS = 10,
+  MZ_ZIP_LDH_FILE_DATE_OFS = 12, MZ_ZIP_LDH_CRC32_OFS = 14, MZ_ZIP_LDH_COMPRESSED_SIZE_OFS = 18, MZ_ZIP_LDH_DECOMPRESSED_SIZE_OFS = 22,
+  MZ_ZIP_LDH_FILENAME_LEN_OFS = 26, MZ_ZIP_LDH_EXTRA_LEN_OFS = 28,
+  // End of central directory offsets
+  MZ_ZIP_ECDH_SIG_OFS = 0, MZ_ZIP_ECDH_NUM_THIS_DISK_OFS = 4, MZ_ZIP_ECDH_NUM_DISK_CDIR_OFS = 6, MZ_ZIP_ECDH_CDIR_NUM_ENTRIES_ON_DISK_OFS = 8,
+  MZ_ZIP_ECDH_CDIR_TOTAL_ENTRIES_OFS = 10, MZ_ZIP_ECDH_CDIR_SIZE_OFS = 12, MZ_ZIP_ECDH_CDIR_OFS_OFS = 16, MZ_ZIP_ECDH_COMMENT_SIZE_OFS = 20,
+};
+
+typedef struct
+{
+  void *m_p;
+  size_t m_size, m_capacity;
+  mz_uint m_element_size;
+} mz_zip_array;
+
+struct mz_zip_internal_state_tag
+{
+  mz_zip_array m_central_dir;
+  mz_zip_array m_central_dir_offsets;
+  mz_zip_array m_sorted_central_dir_offsets;
+  MZ_FILE *m_pFile;
+  void *m_pMem;
+  size_t m_mem_size;
+  size_t m_mem_capacity;
+};
+
+#define MZ_ZIP_ARRAY_SET_ELEMENT_SIZE(array_ptr, element_size) (array_ptr)->m_element_size = element_size
+#define MZ_ZIP_ARRAY_ELEMENT(array_ptr, element_type, index) ((element_type *)((array_ptr)->m_p))[index]
+
+static MZ_FORCEINLINE void mz_zip_array_clear(mz_zip_archive *pZip, mz_zip_array *pArray)
+{
+  pZip->m_pFree(pZip->m_pAlloc_opaque, pArray->m_p);
+  memset(pArray, 0, sizeof(mz_zip_array));
+}
+
+static mz_bool mz_zip_array_ensure_capacity(mz_zip_archive *pZip, mz_zip_array *pArray, size_t min_new_capacity, mz_uint growing)
+{
+  void *pNew_p; size_t new_capacity = min_new_capacity; MZ_ASSERT(pArray->m_element_size); if (pArray->m_capacity >= min_new_capacity) return MZ_TRUE;
+  if (growing) { new_capacity = MZ_MAX(1, pArray->m_capacity); while (new_capacity < min_new_capacity) new_capacity *= 2; }
+  if (NULL == (pNew_p = pZip->m_pRealloc(pZip->m_pAlloc_opaque, pArray->m_p, pArray->m_element_size, new_capacity))) return MZ_FALSE;
+  pArray->m_p = pNew_p; pArray->m_capacity = new_capacity;
+  return MZ_TRUE;
+}
+
+static MZ_FORCEINLINE mz_bool mz_zip_array_reserve(mz_zip_archive *pZip, mz_zip_array *pArray, size_t new_capacity, mz_uint growing)
+{
+  if (new_capacity > pArray->m_capacity) { if (!mz_zip_array_ensure_capacity(pZip, pArray, new_capacity, growing)) return MZ_FALSE; }
+  return MZ_TRUE;
+}
+
+static MZ_FORCEINLINE mz_bool mz_zip_array_resize(mz_zip_archive *pZip, mz_zip_array *pArray, size_t new_size, mz_uint growing)
+{
+  if (new_size > pArray->m_capacity) { if (!mz_zip_array_ensure_capacity(pZip, pArray, new_size, growing)) return MZ_FALSE; }
+  pArray->m_size = new_size;
+  return MZ_TRUE;
+}
+
+static MZ_FORCEINLINE mz_bool mz_zip_array_ensure_room(mz_zip_archive *pZip, mz_zip_array *pArray, size_t n)
+{
+  return mz_zip_array_reserve(pZip, pArray, pArray->m_size + n, MZ_TRUE);
+}
+
+static MZ_FORCEINLINE mz_bool mz_zip_array_push_back(mz_zip_archive *pZip, mz_zip_array *pArray, const void *pElements, size_t n)
+{
+  size_t orig_size = pArray->m_size; if (!mz_zip_array_resize(pZip, pArray, orig_size + n, MZ_TRUE)) return MZ_FALSE;
+  memcpy((mz_uint8*)pArray->m_p + orig_size * pArray->m_element_size, pElements, n * pArray->m_element_size);
+  return MZ_TRUE;
+}
+
+#ifndef MINIZ_NO_TIME
+static time_t mz_zip_dos_to_time_t(int dos_time, int dos_date)
+{
+  struct tm tm;
+  memset(&tm, 0, sizeof(tm)); tm.tm_isdst = -1;
+  tm.tm_year = ((dos_date >> 9) & 127) + 1980 - 1900; tm.tm_mon = ((dos_date >> 5) & 15) - 1; tm.tm_mday = dos_date & 31;
+  tm.tm_hour = (dos_time >> 11) & 31; tm.tm_min = (dos_time >> 5) & 63; tm.tm_sec = (dos_time << 1) & 62;
+  return mktime(&tm);
+}
+
+static void mz_zip_time_to_dos_time(time_t time, mz_uint16 *pDOS_time, mz_uint16 *pDOS_date)
+{
+#ifdef _MSC_VER
+  struct tm tm_struct;
+  struct tm *tm = &tm_struct;
+  errno_t err = localtime_s(tm, &time);
+  if (err)
+  {
+    *pDOS_date = 0; *pDOS_time = 0;
+    return;
+  }
+#else
+  struct tm *tm = localtime(&time);
+#endif
+  *pDOS_time = (mz_uint16)(((tm->tm_hour) << 11) + ((tm->tm_min) << 5) + ((tm->tm_sec) >> 1));
+  *pDOS_date = (mz_uint16)(((tm->tm_year + 1900 - 1980) << 9) + ((tm->tm_mon + 1) << 5) + tm->tm_mday);
+}
+#endif
+
+#ifndef MINIZ_NO_STDIO
+static mz_bool mz_zip_get_file_modified_time(const char *pFilename, mz_uint16 *pDOS_time, mz_uint16 *pDOS_date)
+{
+#ifdef MINIZ_NO_TIME
+  (void)pFilename; *pDOS_date = *pDOS_time = 0;
+#else
+  struct MZ_FILE_STAT_STRUCT file_stat;
+  // On Linux with x86 glibc, this call will fail on large files (>= 0x80000000 bytes) unless you compiled with _LARGEFILE64_SOURCE. Argh.
+  if (MZ_FILE_STAT(pFilename, &file_stat) != 0)
+    return MZ_FALSE;
+  mz_zip_time_to_dos_time(file_stat.st_mtime, pDOS_time, pDOS_date);
+#endif // #ifdef MINIZ_NO_TIME
+  return MZ_TRUE;
+}
+
+#ifndef MINIZ_NO_TIME
+static mz_bool mz_zip_set_file_times(const char *pFilename, time_t access_time, time_t modified_time)
+{
+  struct utimbuf t; t.actime = access_time; t.modtime = modified_time;
+  return !utime(pFilename, &t);
+}
+#endif // #ifndef MINIZ_NO_TIME
+#endif // #ifndef MINIZ_NO_STDIO
+
+static mz_bool mz_zip_reader_init_internal(mz_zip_archive *pZip, mz_uint32 flags)
+{
+  (void)flags;
+  if ((!pZip) || (pZip->m_pState) || (pZip->m_zip_mode != MZ_ZIP_MODE_INVALID))
+    return MZ_FALSE;
+
+  if (!pZip->m_pAlloc) pZip->m_pAlloc = def_alloc_func;
+  if (!pZip->m_pFree) pZip->m_pFree = def_free_func;
+  if (!pZip->m_pRealloc) pZip->m_pRealloc = def_realloc_func;
+
+  pZip->m_zip_mode = MZ_ZIP_MODE_READING;
+  pZip->m_archive_size = 0;
+  pZip->m_central_directory_file_ofs = 0;
+  pZip->m_total_files = 0;
+
+  if (NULL == (pZip->m_pState = (mz_zip_internal_state *)pZip->m_pAlloc(pZip->m_pAlloc_opaque, 1, sizeof(mz_zip_internal_state))))
+    return MZ_FALSE;
+  memset(pZip->m_pState, 0, sizeof(mz_zip_internal_state));
+  MZ_ZIP_ARRAY_SET_ELEMENT_SIZE(&pZip->m_pState->m_central_dir, sizeof(mz_uint8));
+  MZ_ZIP_ARRAY_SET_ELEMENT_SIZE(&pZip->m_pState->m_central_dir_offsets, sizeof(mz_uint32));
+  MZ_ZIP_ARRAY_SET_ELEMENT_SIZE(&pZip->m_pState->m_sorted_central_dir_offsets, sizeof(mz_uint32));
+  return MZ_TRUE;
+}
+
+static MZ_FORCEINLINE mz_bool mz_zip_reader_filename_less(const mz_zip_array *pCentral_dir_array, const mz_zip_array *pCentral_dir_offsets, mz_uint l_index, mz_uint r_index)
+{
+  const mz_uint8 *pL = &MZ_ZIP_ARRAY_ELEMENT(pCentral_dir_array, mz_uint8, MZ_ZIP_ARRAY_ELEMENT(pCentral_dir_offsets, mz_uint32, l_index)), *pE;
+  const mz_uint8 *pR = &MZ_ZIP_ARRAY_ELEMENT(pCentral_dir_array, mz_uint8, MZ_ZIP_ARRAY_ELEMENT(pCentral_dir_offsets, mz_uint32, r_index));
+  mz_uint l_len = MZ_READ_LE16(pL + MZ_ZIP_CDH_FILENAME_LEN_OFS), r_len = MZ_READ_LE16(pR + MZ_ZIP_CDH_FILENAME_LEN_OFS);
+  mz_uint8 l = 0, r = 0;
+  pL += MZ_ZIP_CENTRAL_DIR_HEADER_SIZE; pR += MZ_ZIP_CENTRAL_DIR_HEADER_SIZE;
+  pE = pL + MZ_MIN(l_len, r_len);
+  while (pL < pE)
+  {
+    if ((l = MZ_TOLOWER(*pL)) != (r = MZ_TOLOWER(*pR)))
+      break;
+    pL++; pR++;
+  }
+  return (pL == pE) ? (l_len < r_len) : (l < r);
+}
+
+#define MZ_SWAP_UINT32(a, b) do { mz_uint32 t = a; a = b; b = t; } MZ_MACRO_END
+
+// Heap sort of lowercased filenames, used to help accelerate plain central directory searches by mz_zip_reader_locate_file(). (Could also use qsort(), but it could allocate memory.)
+static void mz_zip_reader_sort_central_dir_offsets_by_filename(mz_zip_archive *pZip)
+{
+  mz_zip_internal_state *pState = pZip->m_pState;
+  const mz_zip_array *pCentral_dir_offsets = &pState->m_central_dir_offsets;
+  const mz_zip_array *pCentral_dir = &pState->m_central_dir;
+  mz_uint32 *pIndices = &MZ_ZIP_ARRAY_ELEMENT(&pState->m_sorted_central_dir_offsets, mz_uint32, 0);
+  const int size = pZip->m_total_files;
+  int start = (size - 2) >> 1, end;
+  while (start >= 0)
+  {
+    int child, root = start;
+    for ( ; ; )
+    {
+      if ((child = (root << 1) + 1) >= size)
+        break;
+      child += (((child + 1) < size) && (mz_zip_reader_filename_less(pCentral_dir, pCentral_dir_offsets, pIndices[child], pIndices[child + 1])));
+      if (!mz_zip_reader_filename_less(pCentral_dir, pCentral_dir_offsets, pIndices[root], pIndices[child]))
+        break;
+      MZ_SWAP_UINT32(pIndices[root], pIndices[child]); root = child;
+    }
+    start--;
+  }
+
+  end = size - 1;
+  while (end > 0)
+  {
+    int child, root = 0;
+    MZ_SWAP_UINT32(pIndices[end], pIndices[0]);
+    for ( ; ; )
+    {
+      if ((child = (root << 1) + 1) >= end)
+        break;
+      child += (((child + 1) < end) && mz_zip_reader_filename_less(pCentral_dir, pCentral_dir_offsets, pIndices[child], pIndices[child + 1]));
+      if (!mz_zip_reader_filename_less(pCentral_dir, pCentral_dir_offsets, pIndices[root], pIndices[child]))
+        break;
+      MZ_SWAP_UINT32(pIndices[root], pIndices[child]); root = child;
+    }
+    end--;
+  }
+}
+
+static mz_bool mz_zip_reader_read_central_dir(mz_zip_archive *pZip, mz_uint32 flags)
+{
+  mz_uint cdir_size, num_this_disk, cdir_disk_index;
+  mz_uint64 cdir_ofs;
+  mz_int64 cur_file_ofs;
+  const mz_uint8 *p;
+  mz_uint32 buf_u32[4096 / sizeof(mz_uint32)]; mz_uint8 *pBuf = (mz_uint8 *)buf_u32;
+  mz_bool sort_central_dir = ((flags & MZ_ZIP_FLAG_DO_NOT_SORT_CENTRAL_DIRECTORY) == 0);
+  // Basic sanity checks - reject files which are too small, and check the first 4 bytes of the file to make sure a local header is there.
+  if (pZip->m_archive_size < MZ_ZIP_END_OF_CENTRAL_DIR_HEADER_SIZE)
+    return MZ_FALSE;
+  // Find the end of central directory record by scanning the file from the end towards the beginning.
+  cur_file_ofs = MZ_MAX((mz_int64)pZip->m_archive_size - (mz_int64)sizeof(buf_u32), 0);
+  for ( ; ; )
+  {
+    int i, n = (int)MZ_MIN(sizeof(buf_u32), pZip->m_archive_size - cur_file_ofs);
+    if (pZip->m_pRead(pZip->m_pIO_opaque, cur_file_ofs, pBuf, n) != (mz_uint)n)
+      return MZ_FALSE;
+    for (i = n - 4; i >= 0; --i)
+      if (MZ_READ_LE32(pBuf + i) == MZ_ZIP_END_OF_CENTRAL_DIR_HEADER_SIG)
+        break;
+    if (i >= 0)
+    {
+      cur_file_ofs += i;
+      break;
+    }
+    if ((!cur_file_ofs) || ((pZip->m_archive_size - cur_file_ofs) >= (0xFFFF + MZ_ZIP_END_OF_CENTRAL_DIR_HEADER_SIZE)))
+      return MZ_FALSE;
+    cur_file_ofs = MZ_MAX(cur_file_ofs - (sizeof(buf_u32) - 3), 0);
+  }
+  // Read and verify the end of central directory record.
+  if (pZip->m_pRead(pZip->m_pIO_opaque, cur_file_ofs, pBuf, MZ_ZIP_END_OF_CENTRAL_DIR_HEADER_SIZE) != MZ_ZIP_END_OF_CENTRAL_DIR_HEADER_SIZE)
+    return MZ_FALSE;
+  if ((MZ_READ_LE32(pBuf + MZ_ZIP_ECDH_SIG_OFS) != MZ_ZIP_END_OF_CENTRAL_DIR_HEADER_SIG) ||
+      ((pZip->m_total_files = MZ_READ_LE16(pBuf + MZ_ZIP_ECDH_CDIR_TOTAL_ENTRIES_OFS)) != MZ_READ_LE16(pBuf + MZ_ZIP_ECDH_CDIR_NUM_ENTRIES_ON_DISK_OFS)))
+    return MZ_FALSE;
+
+  num_this_disk = MZ_READ_LE16(pBuf + MZ_ZIP_ECDH_NUM_THIS_DISK_OFS);
+  cdir_disk_index = MZ_READ_LE16(pBuf + MZ_ZIP_ECDH_NUM_DISK_CDIR_OFS);
+  if (((num_this_disk | cdir_disk_index) != 0) && ((num_this_disk != 1) || (cdir_disk_index != 1)))
+    return MZ_FALSE;
+
+  if ((cdir_size = MZ_READ_LE32(pBuf + MZ_ZIP_ECDH_CDIR_SIZE_OFS)) < pZip->m_total_files * MZ_ZIP_CENTRAL_DIR_HEADER_SIZE)
+    return MZ_FALSE;
+
+  cdir_ofs = MZ_READ_LE32(pBuf + MZ_ZIP_ECDH_CDIR_OFS_OFS);
+  if ((cdir_ofs + (mz_uint64)cdir_size) > pZip->m_archive_size)
+    return MZ_FALSE;
+
+  pZip->m_central_directory_file_ofs = cdir_ofs;
+
+  if (pZip->m_total_files)
+  {
+     mz_uint i, n;
+
+    // Read the entire central directory into a heap block, and allocate another heap block to hold the unsorted central dir file record offsets, and another to hold the sorted indices.
+    if ((!mz_zip_array_resize(pZip, &pZip->m_pState->m_central_dir, cdir_size, MZ_FALSE)) ||
+        (!mz_zip_array_resize(pZip, &pZip->m_pState->m_central_dir_offsets, pZip->m_total_files, MZ_FALSE)))
+      return MZ_FALSE;
+
+    if (sort_central_dir)
+    {
+      if (!mz_zip_array_resize(pZip, &pZip->m_pState->m_sorted_central_dir_offsets, pZip->m_total_files, MZ_FALSE))
+        return MZ_FALSE;
+    }
+
+    if (pZip->m_pRead(pZip->m_pIO_opaque, cdir_ofs, pZip->m_pState->m_central_dir.m_p, cdir_size) != cdir_size)
+      return MZ_FALSE;
+
+    // Now create an index into the central directory file records, do some basic sanity checking on each record, and check for zip64 entries (which are not yet supported).
+    p = (const mz_uint8 *)pZip->m_pState->m_central_dir.m_p;
+    for (n = cdir_size, i = 0; i < pZip->m_total_files; ++i)
+    {
+      mz_uint total_header_size, comp_size, decomp_size, disk_index;
+      if ((n < MZ_ZIP_CENTRAL_DIR_HEADER_SIZE) || (MZ_READ_LE32(p) != MZ_ZIP_CENTRAL_DIR_HEADER_SIG))
+        return MZ_FALSE;
+      MZ_ZIP_ARRAY_ELEMENT(&pZip->m_pState->m_central_dir_offsets, mz_uint32, i) = (mz_uint32)(p - (const mz_uint8 *)pZip->m_pState->m_central_dir.m_p);
+      if (sort_central_dir)
+        MZ_ZIP_ARRAY_ELEMENT(&pZip->m_pState->m_sorted_central_dir_offsets, mz_uint32, i) = i;
+      comp_size = MZ_READ_LE32(p + MZ_ZIP_CDH_COMPRESSED_SIZE_OFS);
+      decomp_size = MZ_READ_LE32(p + MZ_ZIP_CDH_DECOMPRESSED_SIZE_OFS);
+      if (((!MZ_READ_LE32(p + MZ_ZIP_CDH_METHOD_OFS)) && (decomp_size != comp_size)) || (decomp_size && !comp_size) || (decomp_size == 0xFFFFFFFF) || (comp_size == 0xFFFFFFFF))
+        return MZ_FALSE;
+      disk_index = MZ_READ_LE16(p + MZ_ZIP_CDH_DISK_START_OFS);
+      if ((disk_index != num_this_disk) && (disk_index != 1))
+        return MZ_FALSE;
+      if (((mz_uint64)MZ_READ_LE32(p + MZ_ZIP_CDH_LOCAL_HEADER_OFS) + MZ_ZIP_LOCAL_DIR_HEADER_SIZE + comp_size) > pZip->m_archive_size)
+        return MZ_FALSE;
+      if ((total_header_size = MZ_ZIP_CENTRAL_DIR_HEADER_SIZE + MZ_READ_LE16(p + MZ_ZIP_CDH_FILENAME_LEN_OFS) + MZ_READ_LE16(p + MZ_ZIP_CDH_EXTRA_LEN_OFS) + MZ_READ_LE16(p + MZ_ZIP_CDH_COMMENT_LEN_OFS)) > n)
+        return MZ_FALSE;
+      n -= total_header_size; p += total_header_size;
+    }
+  }
+
+  if (sort_central_dir)
+    mz_zip_reader_sort_central_dir_offsets_by_filename(pZip);
+
+  return MZ_TRUE;
+}
+
+mz_bool mz_zip_reader_init(mz_zip_archive *pZip, mz_uint64 size, mz_uint32 flags)
+{
+  if ((!pZip) || (!pZip->m_pRead))
+    return MZ_FALSE;
+  if (!mz_zip_reader_init_internal(pZip, flags))
+    return MZ_FALSE;
+  pZip->m_archive_size = size;
+  if (!mz_zip_reader_read_central_dir(pZip, flags))
+  {
+    mz_zip_reader_end(pZip);
+    return MZ_FALSE;
+  }
+  return MZ_TRUE;
+}
+
+static size_t mz_zip_mem_read_func(void *pOpaque, mz_uint64 file_ofs, void *pBuf, size_t n)
+{
+  mz_zip_archive *pZip = (mz_zip_archive *)pOpaque;
+  size_t s = (file_ofs >= pZip->m_archive_size) ? 0 : (size_t)MZ_MIN(pZip->m_archive_size - file_ofs, n);
+  memcpy(pBuf, (const mz_uint8 *)pZip->m_pState->m_pMem + file_ofs, s);
+  return s;
+}
+
+mz_bool mz_zip_reader_init_mem(mz_zip_archive *pZip, const void *pMem, size_t size, mz_uint32 flags)
+{
+  if (!mz_zip_reader_init_internal(pZip, flags))
+    return MZ_FALSE;
+  pZip->m_archive_size = size;
+  pZip->m_pRead = mz_zip_mem_read_func;
+  pZip->m_pIO_opaque = pZip;
+#ifdef __cplusplus
+  pZip->m_pState->m_pMem = const_cast<void *>(pMem);
+#else
+  pZip->m_pState->m_pMem = (void *)pMem;
+#endif
+  pZip->m_pState->m_mem_size = size;
+  if (!mz_zip_reader_read_central_dir(pZip, flags))
+  {
+    mz_zip_reader_end(pZip);
+    return MZ_FALSE;
+  }
+  return MZ_TRUE;
+}
+
+#ifndef MINIZ_NO_STDIO
+static size_t mz_zip_file_read_func(void *pOpaque, mz_uint64 file_ofs, void *pBuf, size_t n)
+{
+  mz_zip_archive *pZip = (mz_zip_archive *)pOpaque;
+  mz_int64 cur_ofs = MZ_FTELL64(pZip->m_pState->m_pFile);
+  if (((mz_int64)file_ofs < 0) || (((cur_ofs != (mz_int64)file_ofs)) && (MZ_FSEEK64(pZip->m_pState->m_pFile, (mz_int64)file_ofs, SEEK_SET))))
+    return 0;
+  return MZ_FREAD(pBuf, 1, n, pZip->m_pState->m_pFile);
+}
+
+mz_bool mz_zip_reader_init_file(mz_zip_archive *pZip, const char *pFilename, mz_uint32 flags)
+{
+  mz_uint64 file_size;
+  MZ_FILE *pFile = MZ_FOPEN(pFilename, "rb");
+  if (!pFile)
+    return MZ_FALSE;
+  if (MZ_FSEEK64(pFile, 0, SEEK_END))
+  {
+    MZ_FCLOSE(pFile);
+    return MZ_FALSE;
+  }
+  file_size = MZ_FTELL64(pFile);
+  if (!mz_zip_reader_init_internal(pZip, flags))
+  {
+    MZ_FCLOSE(pFile);
+    return MZ_FALSE;
+  }
+  pZip->m_pRead = mz_zip_file_read_func;
+  pZip->m_pIO_opaque = pZip;
+  pZip->m_pState->m_pFile = pFile;
+  pZip->m_archive_size = file_size;
+  if (!mz_zip_reader_read_central_dir(pZip, flags))
+  {
+    mz_zip_reader_end(pZip);
+    return MZ_FALSE;
+  }
+  return MZ_TRUE;
+}
+#endif // #ifndef MINIZ_NO_STDIO
+
+mz_uint mz_zip_reader_get_num_files(mz_zip_archive *pZip)
+{
+  return pZip ? pZip->m_total_files : 0;
+}
+
+static MZ_FORCEINLINE const mz_uint8 *mz_zip_reader_get_cdh(mz_zip_archive *pZip, mz_uint file_index)
+{
+  if ((!pZip) || (!pZip->m_pState) || (file_index >= pZip->m_total_files) || (pZip->m_zip_mode != MZ_ZIP_MODE_READING))
+    return NULL;
+  return &MZ_ZIP_ARRAY_ELEMENT(&pZip->m_pState->m_central_dir, mz_uint8, MZ_ZIP_ARRAY_ELEMENT(&pZip->m_pState->m_central_dir_offsets, mz_uint32, file_index));
+}
+
+mz_bool mz_zip_reader_is_file_encrypted(mz_zip_archive *pZip, mz_uint file_index)
+{
+  mz_uint m_bit_flag;
+  const mz_uint8 *p = mz_zip_reader_get_cdh(pZip, file_index);
+  if (!p)
+    return MZ_FALSE;
+  m_bit_flag = MZ_READ_LE16(p + MZ_ZIP_CDH_BIT_FLAG_OFS);
+  return (m_bit_flag & 1);
+}
+
+mz_bool mz_zip_reader_is_file_a_directory(mz_zip_archive *pZip, mz_uint file_index)
+{
+  mz_uint filename_len, external_attr;
+  const mz_uint8 *p = mz_zip_reader_get_cdh(pZip, file_index);
+  if (!p)
+    return MZ_FALSE;
+
+  // First see if the filename ends with a '/' character.
+  filename_len = MZ_READ_LE16(p + MZ_ZIP_CDH_FILENAME_LEN_OFS);
+  if (filename_len)
+  {
+    if (*(p + MZ_ZIP_CENTRAL_DIR_HEADER_SIZE + filename_len - 1) == '/')
+      return MZ_TRUE;
+  }
+
+  // Bugfix: This code was also checking if the internal attribute was non-zero, which wasn't correct.
+  // Most/all zip writers (hopefully) set DOS file/directory attributes in the low 16-bits, so check for the DOS directory flag and ignore the source OS ID in the created by field.
+  // FIXME: Remove this check? Is it necessary - we already check the filename.
+  external_attr = MZ_READ_LE32(p + MZ_ZIP_CDH_EXTERNAL_ATTR_OFS);
+  if ((external_attr & 0x10) != 0)
+    return MZ_TRUE;
+
+  return MZ_FALSE;
+}
+
+mz_bool mz_zip_reader_file_stat(mz_zip_archive *pZip, mz_uint file_index, mz_zip_archive_file_stat *pStat)
+{
+  mz_uint n;
+  const mz_uint8 *p = mz_zip_reader_get_cdh(pZip, file_index);
+  if ((!p) || (!pStat))
+    return MZ_FALSE;
+
+  // Unpack the central directory record.
+  pStat->m_file_index = file_index;
+  pStat->m_central_dir_ofs = MZ_ZIP_ARRAY_ELEMENT(&pZip->m_pState->m_central_dir_offsets, mz_uint32, file_index);
+  pStat->m_version_made_by = MZ_READ_LE16(p + MZ_ZIP_CDH_VERSION_MADE_BY_OFS);
+  pStat->m_version_needed = MZ_READ_LE16(p + MZ_ZIP_CDH_VERSION_NEEDED_OFS);
+  pStat->m_bit_flag = MZ_READ_LE16(p + MZ_ZIP_CDH_BIT_FLAG_OFS);
+  pStat->m_method = MZ_READ_LE16(p + MZ_ZIP_CDH_METHOD_OFS);
+#ifndef MINIZ_NO_TIME
+  pStat->m_time = mz_zip_dos_to_time_t(MZ_READ_LE16(p + MZ_ZIP_CDH_FILE_TIME_OFS), MZ_READ_LE16(p + MZ_ZIP_CDH_FILE_DATE_OFS));
+#endif
+  pStat->m_crc32 = MZ_READ_LE32(p + MZ_ZIP_CDH_CRC32_OFS);
+  pStat->m_comp_size = MZ_READ_LE32(p + MZ_ZIP_CDH_COMPRESSED_SIZE_OFS);
+  pStat->m_uncomp_size = MZ_READ_LE32(p + MZ_ZIP_CDH_DECOMPRESSED_SIZE_OFS);
+  pStat->m_internal_attr = MZ_READ_LE16(p + MZ_ZIP_CDH_INTERNAL_ATTR_OFS);
+  pStat->m_external_attr = MZ_READ_LE32(p + MZ_ZIP_CDH_EXTERNAL_ATTR_OFS);
+  pStat->m_local_header_ofs = MZ_READ_LE32(p + MZ_ZIP_CDH_LOCAL_HEADER_OFS);
+
+  // Copy as much of the filename and comment as possible.
+  n = MZ_READ_LE16(p + MZ_ZIP_CDH_FILENAME_LEN_OFS); n = MZ_MIN(n, MZ_ZIP_MAX_ARCHIVE_FILENAME_SIZE - 1);
+  memcpy(pStat->m_filename, p + MZ_ZIP_CENTRAL_DIR_HEADER_SIZE, n); pStat->m_filename[n] = '\0';
+
+  n = MZ_READ_LE16(p + MZ_ZIP_CDH_COMMENT_LEN_OFS); n = MZ_MIN(n, MZ_ZIP_MAX_ARCHIVE_FILE_COMMENT_SIZE - 1);
+  pStat->m_comment_size = n;
+  memcpy(pStat->m_comment, p + MZ_ZIP_CENTRAL_DIR_HEADER_SIZE + MZ_READ_LE16(p + MZ_ZIP_CDH_FILENAME_LEN_OFS) + MZ_READ_LE16(p + MZ_ZIP_CDH_EXTRA_LEN_OFS), n); pStat->m_comment[n] = '\0';
+
+  return MZ_TRUE;
+}
+
+mz_uint mz_zip_reader_get_filename(mz_zip_archive *pZip, mz_uint file_index, char *pFilename, mz_uint filename_buf_size)
+{
+  mz_uint n;
+  const mz_uint8 *p = mz_zip_reader_get_cdh(pZip, file_index);
+  if (!p) { if (filename_buf_size) pFilename[0] = '\0'; return 0; }
+  n = MZ_READ_LE16(p + MZ_ZIP_CDH_FILENAME_LEN_OFS);
+  if (filename_buf_size)
+  {
+    n = MZ_MIN(n, filename_buf_size - 1);
+    memcpy(pFilename, p + MZ_ZIP_CENTRAL_DIR_HEADER_SIZE, n);
+    pFilename[n] = '\0';
+  }
+  return n + 1;
+}
+
+static MZ_FORCEINLINE mz_bool mz_zip_reader_string_equal(const char *pA, const char *pB, mz_uint len, mz_uint flags)
+{
+  mz_uint i;
+  if (flags & MZ_ZIP_FLAG_CASE_SENSITIVE)
+    return 0 == memcmp(pA, pB, len);
+  for (i = 0; i < len; ++i)
+    if (MZ_TOLOWER(pA[i]) != MZ_TOLOWER(pB[i]))
+      return MZ_FALSE;
+  return MZ_TRUE;
+}
+
+static MZ_FORCEINLINE int mz_zip_reader_filename_compare(const mz_zip_array *pCentral_dir_array, const mz_zip_array *pCentral_dir_offsets, mz_uint l_index, const char *pR, mz_uint r_len)
+{
+  const mz_uint8 *pL = &MZ_ZIP_ARRAY_ELEMENT(pCentral_dir_array, mz_uint8, MZ_ZIP_ARRAY_ELEMENT(pCentral_dir_offsets, mz_uint32, l_index)), *pE;
+  mz_uint l_len = MZ_READ_LE16(pL + MZ_ZIP_CDH_FILENAME_LEN_OFS);
+  mz_uint8 l = 0, r = 0;
+  pL += MZ_ZIP_CENTRAL_DIR_HEADER_SIZE;
+  pE = pL + MZ_MIN(l_len, r_len);
+  while (pL < pE)
+  {
+    if ((l = MZ_TOLOWER(*pL)) != (r = MZ_TOLOWER(*pR)))
+      break;
+    pL++; pR++;
+  }
+  return (pL == pE) ? (int)(l_len - r_len) : (l - r);
+}
+
+static int mz_zip_reader_locate_file_binary_search(mz_zip_archive *pZip, const char *pFilename)
+{
+  mz_zip_internal_state *pState = pZip->m_pState;
+  const mz_zip_array *pCentral_dir_offsets = &pState->m_central_dir_offsets;
+  const mz_zip_array *pCentral_dir = &pState->m_central_dir;
+  mz_uint32 *pIndices = &MZ_ZIP_ARRAY_ELEMENT(&pState->m_sorted_central_dir_offsets, mz_uint32, 0);
+  const int size = pZip->m_total_files;
+  const mz_uint filename_len = (mz_uint)strlen(pFilename);
+  int l = 0, h = size - 1;
+  while (l <= h)
+  {
+    int m = (l + h) >> 1, file_index = pIndices[m], comp = mz_zip_reader_filename_compare(pCentral_dir, pCentral_dir_offsets, file_index, pFilename, filename_len);
+    if (!comp)
+      return file_index;
+    else if (comp < 0)
+      l = m + 1;
+    else
+      h = m - 1;
+  }
+  return -1;
+}
+
+int mz_zip_reader_locate_file(mz_zip_archive *pZip, const char *pName, const char *pComment, mz_uint flags)
+{
+  mz_uint file_index; size_t name_len, comment_len;
+  if ((!pZip) || (!pZip->m_pState) || (!pName) || (pZip->m_zip_mode != MZ_ZIP_MODE_READING))
+    return -1;
+  if (((flags & (MZ_ZIP_FLAG_IGNORE_PATH | MZ_ZIP_FLAG_CASE_SENSITIVE)) == 0) && (!pComment) && (pZip->m_pState->m_sorted_central_dir_offsets.m_size))
+    return mz_zip_reader_locate_file_binary_search(pZip, pName);
+  name_len = strlen(pName); if (name_len > 0xFFFF) return -1;
+  comment_len = pComment ? strlen(pComment) : 0; if (comment_len > 0xFFFF) return -1;
+  for (file_index = 0; file_index < pZip->m_total_files; file_index++)
+  {
+    const mz_uint8 *pHeader = &MZ_ZIP_ARRAY_ELEMENT(&pZip->m_pState->m_central_dir, mz_uint8, MZ_ZIP_ARRAY_ELEMENT(&pZip->m_pState->m_central_dir_offsets, mz_uint32, file_index));
+    mz_uint filename_len = MZ_READ_LE16(pHeader + MZ_ZIP_CDH_FILENAME_LEN_OFS);
+    const char *pFilename = (const char *)pHeader + MZ_ZIP_CENTRAL_DIR_HEADER_SIZE;
+    if (filename_len < name_len)
+      continue;
+    if (comment_len)
+    {
+      mz_uint file_extra_len = MZ_READ_LE16(pHeader + MZ_ZIP_CDH_EXTRA_LEN_OFS), file_comment_len = MZ_READ_LE16(pHeader + MZ_ZIP_CDH_COMMENT_LEN_OFS);
+      const char *pFile_comment = pFilename + filename_len + file_extra_len;
+      if ((file_comment_len != comment_len) || (!mz_zip_reader_string_equal(pComment, pFile_comment, file_comment_len, flags)))
+        continue;
+    }
+    if ((flags & MZ_ZIP_FLAG_IGNORE_PATH) && (filename_len))
+    {
+      int ofs = filename_len - 1;
+      do
+      {
+        if ((pFilename[ofs] == '/') || (pFilename[ofs] == '\\') || (pFilename[ofs] == ':'))
+          break;
+      } while (--ofs >= 0);
+      ofs++;
+      pFilename += ofs; filename_len -= ofs;
+    }
+    if ((filename_len == name_len) && (mz_zip_reader_string_equal(pName, pFilename, filename_len, flags)))
+      return file_index;
+  }
+  return -1;
+}
+
+mz_bool mz_zip_reader_extract_to_mem_no_alloc(mz_zip_archive *pZip, mz_uint file_index, void *pBuf, size_t buf_size, mz_uint flags, void *pUser_read_buf, size_t user_read_buf_size)
+{
+  int status = TINFL_STATUS_DONE;
+  mz_uint64 needed_size, cur_file_ofs, comp_remaining, out_buf_ofs = 0, read_buf_size, read_buf_ofs = 0, read_buf_avail;
+  mz_zip_archive_file_stat file_stat;
+  void *pRead_buf;
+  mz_uint32 local_header_u32[(MZ_ZIP_LOCAL_DIR_HEADER_SIZE + sizeof(mz_uint32) - 1) / sizeof(mz_uint32)]; mz_uint8 *pLocal_header = (mz_uint8 *)local_header_u32;
+  tinfl_decompressor inflator;
+
+  if ((buf_size) && (!pBuf))
+    return MZ_FALSE;
+
+  if (!mz_zip_reader_file_stat(pZip, file_index, &file_stat))
+    return MZ_FALSE;
+
+  // Empty file, or a directory (but not always a directory - I've seen odd zips with directories that have compressed data which inflates to 0 bytes)
+  if (!file_stat.m_comp_size)
+    return MZ_TRUE;
+
+  // Entry is a subdirectory (I've seen old zips with dir entries which have compressed deflate data which inflates to 0 bytes, but these entries claim to uncompress to 512 bytes in the headers).
+  // I'm torn how to handle this case - should it fail instead?
+  if (mz_zip_reader_is_file_a_directory(pZip, file_index))
+    return MZ_TRUE;
+
+  // Encryption and patch files are not supported.
+  if (file_stat.m_bit_flag & (1 | 32))
+    return MZ_FALSE;
+
+  // This function only supports stored and deflate.
+  if ((!(flags & MZ_ZIP_FLAG_COMPRESSED_DATA)) && (file_stat.m_method != 0) && (file_stat.m_method != MZ_DEFLATED))
+    return MZ_FALSE;
+
+  // Ensure supplied output buffer is large enough.
+  needed_size = (flags & MZ_ZIP_FLAG_COMPRESSED_DATA) ? file_stat.m_comp_size : file_stat.m_uncomp_size;
+  if (buf_size < needed_size)
+    return MZ_FALSE;
+
+  // Read and parse the local directory entry.
+  cur_file_ofs = file_stat.m_local_header_ofs;
+  if (pZip->m_pRead(pZip->m_pIO_opaque, cur_file_ofs, pLocal_header, MZ_ZIP_LOCAL_DIR_HEADER_SIZE) != MZ_ZIP_LOCAL_DIR_HEADER_SIZE)
+    return MZ_FALSE;
+  if (MZ_READ_LE32(pLocal_header) != MZ_ZIP_LOCAL_DIR_HEADER_SIG)
+    return MZ_FALSE;
+
+  cur_file_ofs += MZ_ZIP_LOCAL_DIR_HEADER_SIZE + MZ_READ_LE16(pLocal_header + MZ_ZIP_LDH_FILENAME_LEN_OFS) + MZ_READ_LE16(pLocal_header + MZ_ZIP_LDH_EXTRA_LEN_OFS);
+  if ((cur_file_ofs + file_stat.m_comp_size) > pZip->m_archive_size)
+    return MZ_FALSE;
+
+  if ((flags & MZ_ZIP_FLAG_COMPRESSED_DATA) || (!file_stat.m_method))
+  {
+    // The file is stored or the caller has requested the compressed data.
+    if (pZip->m_pRead(pZip->m_pIO_opaque, cur_file_ofs, pBuf, (size_t)needed_size) != needed_size)
+      return MZ_FALSE;
+    return ((flags & MZ_ZIP_FLAG_COMPRESSED_DATA) != 0) || (mz_crc32(MZ_CRC32_INIT, (const mz_uint8 *)pBuf, (size_t)file_stat.m_uncomp_size) == file_stat.m_crc32);
+  }
+
+  // Decompress the file either directly from memory or from a file input buffer.
+  tinfl_init(&inflator);
+
+  if (pZip->m_pState->m_pMem)
+  {
+    // Read directly from the archive in memory.
+    pRead_buf = (mz_uint8 *)pZip->m_pState->m_pMem + cur_file_ofs;
+    read_buf_size = read_buf_avail = file_stat.m_comp_size;
+    comp_remaining = 0;
+  }
+  else if (pUser_read_buf)
+  {
+    // Use a user provided read buffer.
+    if (!user_read_buf_size)
+      return MZ_FALSE;
+    pRead_buf = (mz_uint8 *)pUser_read_buf;
+    read_buf_size = user_read_buf_size;
+    read_buf_avail = 0;
+    comp_remaining = file_stat.m_comp_size;
+  }
+  else
+  {
+    // Temporarily allocate a read buffer.
+    read_buf_size = MZ_MIN(file_stat.m_comp_size, MZ_ZIP_MAX_IO_BUF_SIZE);
+#ifdef _MSC_VER
+    if (((0, sizeof(size_t) == sizeof(mz_uint32))) && (read_buf_size > 0x7FFFFFFF))
+#else
+    if (((sizeof(size_t) == sizeof(mz_uint32))) && (read_buf_size > 0x7FFFFFFF))
+#endif
+      return MZ_FALSE;
+    if (NULL == (pRead_buf = pZip->m_pAlloc(pZip->m_pAlloc_opaque, 1, (size_t)read_buf_size)))
+      return MZ_FALSE;
+    read_buf_avail = 0;
+    comp_remaining = file_stat.m_comp_size;
+  }
+
+  do
+  {
+    size_t in_buf_size, out_buf_size = (size_t)(file_stat.m_uncomp_size - out_buf_ofs);
+    if ((!read_buf_avail) && (!pZip->m_pState->m_pMem))
+    {
+      read_buf_avail = MZ_MIN(read_buf_size, comp_remaining);
+      if (pZip->m_pRead(pZip->m_pIO_opaque, cur_file_ofs, pRead_buf, (size_t)read_buf_avail) != read_buf_avail)
+      {
+        status = TINFL_STATUS_FAILED;
+        break;
+      }
+      cur_file_ofs += read_buf_avail;
+      comp_remaining -= read_buf_avail;
+      read_buf_ofs = 0;
+    }
+    in_buf_size = (size_t)read_buf_avail;
+    status = tinfl_decompress(&inflator, (mz_uint8 *)pRead_buf + read_buf_ofs, &in_buf_size, (mz_uint8 *)pBuf, (mz_uint8 *)pBuf + out_buf_ofs, &out_buf_size, TINFL_FLAG_USING_NON_WRAPPING_OUTPUT_BUF | (comp_remaining ? TINFL_FLAG_HAS_MORE_INPUT : 0));
+    read_buf_avail -= in_buf_size;
+    read_buf_ofs += in_buf_size;
+    out_buf_ofs += out_buf_size;
+  } while (status == TINFL_STATUS_NEEDS_MORE_INPUT);
+
+  if (status == TINFL_STATUS_DONE)
+  {
+    // Make sure the entire file was decompressed, and check its CRC.
+    if ((out_buf_ofs != file_stat.m_uncomp_size) || (mz_crc32(MZ_CRC32_INIT, (const mz_uint8 *)pBuf, (size_t)file_stat.m_uncomp_size) != file_stat.m_crc32))
+      status = TINFL_STATUS_FAILED;
+  }
+
+  if ((!pZip->m_pState->m_pMem) && (!pUser_read_buf))
+    pZip->m_pFree(pZip->m_pAlloc_opaque, pRead_buf);
+
+  return status == TINFL_STATUS_DONE;
+}
+
+mz_bool mz_zip_reader_extract_file_to_mem_no_alloc(mz_zip_archive *pZip, const char *pFilename, void *pBuf, size_t buf_size, mz_uint flags, void *pUser_read_buf, size_t user_read_buf_size)
+{
+  int file_index = mz_zip_reader_locate_file(pZip, pFilename, NULL, flags);
+  if (file_index < 0)
+    return MZ_FALSE;
+  return mz_zip_reader_extract_to_mem_no_alloc(pZip, file_index, pBuf, buf_size, flags, pUser_read_buf, user_read_buf_size);
+}
+
+mz_bool mz_zip_reader_extract_to_mem(mz_zip_archive *pZip, mz_uint file_index, void *pBuf, size_t buf_size, mz_uint flags)
+{
+  return mz_zip_reader_extract_to_mem_no_alloc(pZip, file_index, pBuf, buf_size, flags, NULL, 0);
+}
+
+mz_bool mz_zip_reader_extract_file_to_mem(mz_zip_archive *pZip, const char *pFilename, void *pBuf, size_t buf_size, mz_uint flags)
+{
+  return mz_zip_reader_extract_file_to_mem_no_alloc(pZip, pFilename, pBuf, buf_size, flags, NULL, 0);
+}
+
+void *mz_zip_reader_extract_to_heap(mz_zip_archive *pZip, mz_uint file_index, size_t *pSize, mz_uint flags)
+{
+  mz_uint64 comp_size, uncomp_size, alloc_size;
+  const mz_uint8 *p = mz_zip_reader_get_cdh(pZip, file_index);
+  void *pBuf;
+
+  if (pSize)
+    *pSize = 0;
+  if (!p)
+    return NULL;
+
+  comp_size = MZ_READ_LE32(p + MZ_ZIP_CDH_COMPRESSED_SIZE_OFS);
+  uncomp_size = MZ_READ_LE32(p + MZ_ZIP_CDH_DECOMPRESSED_SIZE_OFS);
+
+  alloc_size = (flags & MZ_ZIP_FLAG_COMPRESSED_DATA) ? comp_size : uncomp_size;
+#ifdef _MSC_VER
+  if (((0, sizeof(size_t) == sizeof(mz_uint32))) && (alloc_size > 0x7FFFFFFF))
+#else
+  if (((sizeof(size_t) == sizeof(mz_uint32))) && (alloc_size > 0x7FFFFFFF))
+#endif
+    return NULL;
+  if (NULL == (pBuf = pZip->m_pAlloc(pZip->m_pAlloc_opaque, 1, (size_t)alloc_size)))
+    return NULL;
+
+  if (!mz_zip_reader_extract_to_mem(pZip, file_index, pBuf, (size_t)alloc_size, flags))
+  {
+    pZip->m_pFree(pZip->m_pAlloc_opaque, pBuf);
+    return NULL;
+  }
+
+  if (pSize) *pSize = (size_t)alloc_size;
+  return pBuf;
+}
+
+void *mz_zip_reader_extract_file_to_heap(mz_zip_archive *pZip, const char *pFilename, size_t *pSize, mz_uint flags)
+{
+  int file_index = mz_zip_reader_locate_file(pZip, pFilename, NULL, flags);
+  if (file_index < 0)
+  {
+    if (pSize) *pSize = 0;
+    return MZ_FALSE;
+  }
+  return mz_zip_reader_extract_to_heap(pZip, file_index, pSize, flags);
+}
+
+mz_bool mz_zip_reader_extract_to_callback(mz_zip_archive *pZip, mz_uint file_index, mz_file_write_func pCallback, void *pOpaque, mz_uint flags)
+{
+  int status = TINFL_STATUS_DONE; mz_uint file_crc32 = MZ_CRC32_INIT;
+  mz_uint64 read_buf_size, read_buf_ofs = 0, read_buf_avail, comp_remaining, out_buf_ofs = 0, cur_file_ofs;
+  mz_zip_archive_file_stat file_stat;
+  void *pRead_buf = NULL; void *pWrite_buf = NULL;
+  mz_uint32 local_header_u32[(MZ_ZIP_LOCAL_DIR_HEADER_SIZE + sizeof(mz_uint32) - 1) / sizeof(mz_uint32)]; mz_uint8 *pLocal_header = (mz_uint8 *)local_header_u32;
+
+  if (!mz_zip_reader_file_stat(pZip, file_index, &file_stat))
+    return MZ_FALSE;
+
+  // Empty file, or a directory (but not always a directory - I've seen odd zips with directories that have compressed data which inflates to 0 bytes)
+  if (!file_stat.m_comp_size)
+    return MZ_TRUE;
+
+  // Entry is a subdirectory (I've seen old zips with dir entries which have compressed deflate data which inflates to 0 bytes, but these entries claim to uncompress to 512 bytes in the headers).
+  // I'm torn how to handle this case - should it fail instead?
+  if (mz_zip_reader_is_file_a_directory(pZip, file_index))
+    return MZ_TRUE;
+
+  // Encryption and patch files are not supported.
+  if (file_stat.m_bit_flag & (1 | 32))
+    return MZ_FALSE;
+
+  // This function only supports stored and deflate.
+  if ((!(flags & MZ_ZIP_FLAG_COMPRESSED_DATA)) && (file_stat.m_method != 0) && (file_stat.m_method != MZ_DEFLATED))
+    return MZ_FALSE;
+
+  // Read and parse the local directory entry.
+  cur_file_ofs = file_stat.m_local_header_ofs;
+  if (pZip->m_pRead(pZip->m_pIO_opaque, cur_file_ofs, pLocal_header, MZ_ZIP_LOCAL_DIR_HEADER_SIZE) != MZ_ZIP_LOCAL_DIR_HEADER_SIZE)
+    return MZ_FALSE;
+  if (MZ_READ_LE32(pLocal_header) != MZ_ZIP_LOCAL_DIR_HEADER_SIG)
+    return MZ_FALSE;
+
+  cur_file_ofs += MZ_ZIP_LOCAL_DIR_HEADER_SIZE + MZ_READ_LE16(pLocal_header + MZ_ZIP_LDH_FILENAME_LEN_OFS) + MZ_READ_LE16(pLocal_header + MZ_ZIP_LDH_EXTRA_LEN_OFS);
+  if ((cur_file_ofs + file_stat.m_comp_size) > pZip->m_archive_size)
+    return MZ_FALSE;
+
+  // Decompress the file either directly from memory or from a file input buffer.
+  if (pZip->m_pState->m_pMem)
+  {
+    pRead_buf = (mz_uint8 *)pZip->m_pState->m_pMem + cur_file_ofs;
+    read_buf_size = read_buf_avail = file_stat.m_comp_size;
+    comp_remaining = 0;
+  }
+  else
+  {
+    read_buf_size = MZ_MIN(file_stat.m_comp_size, MZ_ZIP_MAX_IO_BUF_SIZE);
+    if (NULL == (pRead_buf = pZip->m_pAlloc(pZip->m_pAlloc_opaque, 1, (size_t)read_buf_size)))
+      return MZ_FALSE;
+    read_buf_avail = 0;
+    comp_remaining = file_stat.m_comp_size;
+  }
+
+  if ((flags & MZ_ZIP_FLAG_COMPRESSED_DATA) || (!file_stat.m_method))
+  {
+    // The file is stored or the caller has requested the compressed data.
+    if (pZip->m_pState->m_pMem)
+    {
+#ifdef _MSC_VER
+      if (((0, sizeof(size_t) == sizeof(mz_uint32))) && (file_stat.m_comp_size > 0xFFFFFFFF))
+#else
+      if (((sizeof(size_t) == sizeof(mz_uint32))) && (file_stat.m_comp_size > 0xFFFFFFFF))
+#endif
+        return MZ_FALSE;
+      if (pCallback(pOpaque, out_buf_ofs, pRead_buf, (size_t)file_stat.m_comp_size) != file_stat.m_comp_size)
+        status = TINFL_STATUS_FAILED;
+      else if (!(flags & MZ_ZIP_FLAG_COMPRESSED_DATA))
+        file_crc32 = (mz_uint32)mz_crc32(file_crc32, (const mz_uint8 *)pRead_buf, (size_t)file_stat.m_comp_size);
+      cur_file_ofs += file_stat.m_comp_size;
+      out_buf_ofs += file_stat.m_comp_size;
+      comp_remaining = 0;
+    }
+    else
+    {
+      while (comp_remaining)
+      {
+        read_buf_avail = MZ_MIN(read_buf_size, comp_remaining);
+        if (pZip->m_pRead(pZip->m_pIO_opaque, cur_file_ofs, pRead_buf, (size_t)read_buf_avail) != read_buf_avail)
+        {
+          status = TINFL_STATUS_FAILED;
+          break;
+        }
+
+        if (!(flags & MZ_ZIP_FLAG_COMPRESSED_DATA))
+          file_crc32 = (mz_uint32)mz_crc32(file_crc32, (const mz_uint8 *)pRead_buf, (size_t)read_buf_avail);
+
+        if (pCallback(pOpaque, out_buf_ofs, pRead_buf, (size_t)read_buf_avail) != read_buf_avail)
+        {
+          status = TINFL_STATUS_FAILED;
+          break;
+        }
+        cur_file_ofs += read_buf_avail;
+        out_buf_ofs += read_buf_avail;
+        comp_remaining -= read_buf_avail;
+      }
+    }
+  }
+  else
+  {
+    tinfl_decompressor inflator;
+    tinfl_init(&inflator);
+
+    if (NULL == (pWrite_buf = pZip->m_pAlloc(pZip->m_pAlloc_opaque, 1, TINFL_LZ_DICT_SIZE)))
+      status = TINFL_STATUS_FAILED;
+    else
+    {
+      do
+      {
+        mz_uint8 *pWrite_buf_cur = (mz_uint8 *)pWrite_buf + (out_buf_ofs & (TINFL_LZ_DICT_SIZE - 1));
+        size_t in_buf_size, out_buf_size = TINFL_LZ_DICT_SIZE - (out_buf_ofs & (TINFL_LZ_DICT_SIZE - 1));
+        if ((!read_buf_avail) && (!pZip->m_pState->m_pMem))
+        {
+          read_buf_avail = MZ_MIN(read_buf_size, comp_remaining);
+          if (pZip->m_pRead(pZip->m_pIO_opaque, cur_file_ofs, pRead_buf, (size_t)read_buf_avail) != read_buf_avail)
+          {
+            status = TINFL_STATUS_FAILED;
+            break;
+          }
+          cur_file_ofs += read_buf_avail;
+          comp_remaining -= read_buf_avail;
+          read_buf_ofs = 0;
+        }
+
+        in_buf_size = (size_t)read_buf_avail;
+        status = tinfl_decompress(&inflator, (const mz_uint8 *)pRead_buf + read_buf_ofs, &in_buf_size, (mz_uint8 *)pWrite_buf, pWrite_buf_cur, &out_buf_size, comp_remaining ? TINFL_FLAG_HAS_MORE_INPUT : 0);
+        read_buf_avail -= in_buf_size;
+        read_buf_ofs += in_buf_size;
+
+        if (out_buf_size)
+        {
+          if (pCallback(pOpaque, out_buf_ofs, pWrite_buf_cur, out_buf_size) != out_buf_size)
+          {
+            status = TINFL_STATUS_FAILED;
+            break;
+          }
+          file_crc32 = (mz_uint32)mz_crc32(file_crc32, pWrite_buf_cur, out_buf_size);
+          if ((out_buf_ofs += out_buf_size) > file_stat.m_uncomp_size)
+          {
+            status = TINFL_STATUS_FAILED;
+            break;
+          }
+        }
+      } while ((status == TINFL_STATUS_NEEDS_MORE_INPUT) || (status == TINFL_STATUS_HAS_MORE_OUTPUT));
+    }
+  }
+
+  if ((status == TINFL_STATUS_DONE) && (!(flags & MZ_ZIP_FLAG_COMPRESSED_DATA)))
+  {
+    // Make sure the entire file was decompressed, and check its CRC.
+    if ((out_buf_ofs != file_stat.m_uncomp_size) || (file_crc32 != file_stat.m_crc32))
+      status = TINFL_STATUS_FAILED;
+  }
+
+  if (!pZip->m_pState->m_pMem)
+    pZip->m_pFree(pZip->m_pAlloc_opaque, pRead_buf);
+  if (pWrite_buf)
+    pZip->m_pFree(pZip->m_pAlloc_opaque, pWrite_buf);
+
+  return status == TINFL_STATUS_DONE;
+}
+
+mz_bool mz_zip_reader_extract_file_to_callback(mz_zip_archive *pZip, const char *pFilename, mz_file_write_func pCallback, void *pOpaque, mz_uint flags)
+{
+  int file_index = mz_zip_reader_locate_file(pZip, pFilename, NULL, flags);
+  if (file_index < 0)
+    return MZ_FALSE;
+  return mz_zip_reader_extract_to_callback(pZip, file_index, pCallback, pOpaque, flags);
+}
+
+#ifndef MINIZ_NO_STDIO
+static size_t mz_zip_file_write_callback(void *pOpaque, mz_uint64 ofs, const void *pBuf, size_t n)
+{
+  (void)ofs; return MZ_FWRITE(pBuf, 1, n, (MZ_FILE*)pOpaque);
+}
+
+mz_bool mz_zip_reader_extract_to_file(mz_zip_archive *pZip, mz_uint file_index, const char *pDst_filename, mz_uint flags)
+{
+  mz_bool status;
+  mz_zip_archive_file_stat file_stat;
+  MZ_FILE *pFile;
+  if (!mz_zip_reader_file_stat(pZip, file_index, &file_stat))
+    return MZ_FALSE;
+  pFile = MZ_FOPEN(pDst_filename, "wb");
+  if (!pFile)
+    return MZ_FALSE;
+  status = mz_zip_reader_extract_to_callback(pZip, file_index, mz_zip_file_write_callback, pFile, flags);
+  if (MZ_FCLOSE(pFile) == EOF)
+    return MZ_FALSE;
+#ifndef MINIZ_NO_TIME
+  if (status)
+    mz_zip_set_file_times(pDst_filename, file_stat.m_time, file_stat.m_time);
+#endif
+  return status;
+}
+#endif // #ifndef MINIZ_NO_STDIO
+
+mz_bool mz_zip_reader_end(mz_zip_archive *pZip)
+{
+  if ((!pZip) || (!pZip->m_pState) || (!pZip->m_pAlloc) || (!pZip->m_pFree) || (pZip->m_zip_mode != MZ_ZIP_MODE_READING))
+    return MZ_FALSE;
+
+  if (pZip->m_pState)
+  {
+    mz_zip_internal_state *pState = pZip->m_pState; pZip->m_pState = NULL;
+    mz_zip_array_clear(pZip, &pState->m_central_dir);
+    mz_zip_array_clear(pZip, &pState->m_central_dir_offsets);
+    mz_zip_array_clear(pZip, &pState->m_sorted_central_dir_offsets);
+
+#ifndef MINIZ_NO_STDIO
+    if (pState->m_pFile)
+    {
+      MZ_FCLOSE(pState->m_pFile);
+      pState->m_pFile = NULL;
+    }
+#endif // #ifndef MINIZ_NO_STDIO
+
+    pZip->m_pFree(pZip->m_pAlloc_opaque, pState);
+  }
+  pZip->m_zip_mode = MZ_ZIP_MODE_INVALID;
+
+  return MZ_TRUE;
+}
+
+#ifndef MINIZ_NO_STDIO
+mz_bool mz_zip_reader_extract_file_to_file(mz_zip_archive *pZip, const char *pArchive_filename, const char *pDst_filename, mz_uint flags)
+{
+  int file_index = mz_zip_reader_locate_file(pZip, pArchive_filename, NULL, flags);
+  if (file_index < 0)
+    return MZ_FALSE;
+  return mz_zip_reader_extract_to_file(pZip, file_index, pDst_filename, flags);
+}
+#endif
+
+// ------------------- .ZIP archive writing
+
+#ifndef MINIZ_NO_ARCHIVE_WRITING_APIS
+
+static void mz_write_le16(mz_uint8 *p, mz_uint16 v) { p[0] = (mz_uint8)v; p[1] = (mz_uint8)(v >> 8); }
+static void mz_write_le32(mz_uint8 *p, mz_uint32 v) { p[0] = (mz_uint8)v; p[1] = (mz_uint8)(v >> 8); p[2] = (mz_uint8)(v >> 16); p[3] = (mz_uint8)(v >> 24); }
+#define MZ_WRITE_LE16(p, v) mz_write_le16((mz_uint8 *)(p), (mz_uint16)(v))
+#define MZ_WRITE_LE32(p, v) mz_write_le32((mz_uint8 *)(p), (mz_uint32)(v))
+
+mz_bool mz_zip_writer_init(mz_zip_archive *pZip, mz_uint64 existing_size)
+{
+  if ((!pZip) || (pZip->m_pState) || (!pZip->m_pWrite) || (pZip->m_zip_mode != MZ_ZIP_MODE_INVALID))
+    return MZ_FALSE;
+
+  if (pZip->m_file_offset_alignment)
+  {
+    // Ensure user specified file offset alignment is a power of 2.
+    if (pZip->m_file_offset_alignment & (pZip->m_file_offset_alignment - 1))
+      return MZ_FALSE;
+  }
+
+  if (!pZip->m_pAlloc) pZip->m_pAlloc = def_alloc_func;
+  if (!pZip->m_pFree) pZip->m_pFree = def_free_func;
+  if (!pZip->m_pRealloc) pZip->m_pRealloc = def_realloc_func;
+
+  pZip->m_zip_mode = MZ_ZIP_MODE_WRITING;
+  pZip->m_archive_size = existing_size;
+  pZip->m_central_directory_file_ofs = 0;
+  pZip->m_total_files = 0;
+
+  if (NULL == (pZip->m_pState = (mz_zip_internal_state *)pZip->m_pAlloc(pZip->m_pAlloc_opaque, 1, sizeof(mz_zip_internal_state))))
+    return MZ_FALSE;
+  memset(pZip->m_pState, 0, sizeof(mz_zip_internal_state));
+  MZ_ZIP_ARRAY_SET_ELEMENT_SIZE(&pZip->m_pState->m_central_dir, sizeof(mz_uint8));
+  MZ_ZIP_ARRAY_SET_ELEMENT_SIZE(&pZip->m_pState->m_central_dir_offsets, sizeof(mz_uint32));
+  MZ_ZIP_ARRAY_SET_ELEMENT_SIZE(&pZip->m_pState->m_sorted_central_dir_offsets, sizeof(mz_uint32));
+  return MZ_TRUE;
+}
+
+static size_t mz_zip_heap_write_func(void *pOpaque, mz_uint64 file_ofs, const void *pBuf, size_t n)
+{
+  mz_zip_archive *pZip = (mz_zip_archive *)pOpaque;
+  mz_zip_internal_state *pState = pZip->m_pState;
+  mz_uint64 new_size = MZ_MAX(file_ofs + n, pState->m_mem_size);
+#ifdef _MSC_VER
+  if ((!n) || ((0, sizeof(size_t) == sizeof(mz_uint32)) && (new_size > 0x7FFFFFFF)))
+#else
+  if ((!n) || ((sizeof(size_t) == sizeof(mz_uint32)) && (new_size > 0x7FFFFFFF)))
+#endif
+    return 0;
+  if (new_size > pState->m_mem_capacity)
+  {
+    void *pNew_block;
+    size_t new_capacity = MZ_MAX(64, pState->m_mem_capacity); while (new_capacity < new_size) new_capacity *= 2;
+    if (NULL == (pNew_block = pZip->m_pRealloc(pZip->m_pAlloc_opaque, pState->m_pMem, 1, new_capacity)))
+      return 0;
+    pState->m_pMem = pNew_block; pState->m_mem_capacity = new_capacity;
+  }
+  memcpy((mz_uint8 *)pState->m_pMem + file_ofs, pBuf, n);
+  pState->m_mem_size = (size_t)new_size;
+  return n;
+}
+
+mz_bool mz_zip_writer_init_heap(mz_zip_archive *pZip, size_t size_to_reserve_at_beginning, size_t initial_allocation_size)
+{
+  pZip->m_pWrite = mz_zip_heap_write_func;
+  pZip->m_pIO_opaque = pZip;
+  if (!mz_zip_writer_init(pZip, size_to_reserve_at_beginning))
+    return MZ_FALSE;
+  if (0 != (initial_allocation_size = MZ_MAX(initial_allocation_size, size_to_reserve_at_beginning)))
+  {
+    if (NULL == (pZip->m_pState->m_pMem = pZip->m_pAlloc(pZip->m_pAlloc_opaque, 1, initial_allocation_size)))
+    {
+      mz_zip_writer_end(pZip);
+      return MZ_FALSE;
+    }
+    pZip->m_pState->m_mem_capacity = initial_allocation_size;
+  }
+  return MZ_TRUE;
+}
+
+#ifndef MINIZ_NO_STDIO
+static size_t mz_zip_file_write_func(void *pOpaque, mz_uint64 file_ofs, const void *pBuf, size_t n)
+{
+  mz_zip_archive *pZip = (mz_zip_archive *)pOpaque;
+  mz_int64 cur_ofs = MZ_FTELL64(pZip->m_pState->m_pFile);
+  if (((mz_int64)file_ofs < 0) || (((cur_ofs != (mz_int64)file_ofs)) && (MZ_FSEEK64(pZip->m_pState->m_pFile, (mz_int64)file_ofs, SEEK_SET))))
+    return 0;
+  return MZ_FWRITE(pBuf, 1, n, pZip->m_pState->m_pFile);
+}
+
+mz_bool mz_zip_writer_init_file(mz_zip_archive *pZip, const char *pFilename, mz_uint64 size_to_reserve_at_beginning)
+{
+  MZ_FILE *pFile;
+  pZip->m_pWrite = mz_zip_file_write_func;
+  pZip->m_pIO_opaque = pZip;
+  if (!mz_zip_writer_init(pZip, size_to_reserve_at_beginning))
+    return MZ_FALSE;
+  if (NULL == (pFile = MZ_FOPEN(pFilename, "wb")))
+  {
+    mz_zip_writer_end(pZip);
+    return MZ_FALSE;
+  }
+  pZip->m_pState->m_pFile = pFile;
+  if (size_to_reserve_at_beginning)
+  {
+    mz_uint64 cur_ofs = 0; char buf[4096]; MZ_CLEAR_OBJ(buf);
+    do
+    {
+      size_t n = (size_t)MZ_MIN(sizeof(buf), size_to_reserve_at_beginning);
+      if (pZip->m_pWrite(pZip->m_pIO_opaque, cur_ofs, buf, n) != n)
+      {
+        mz_zip_writer_end(pZip);
+        return MZ_FALSE;
+      }
+      cur_ofs += n; size_to_reserve_at_beginning -= n;
+    } while (size_to_reserve_at_beginning);
+  }
+  return MZ_TRUE;
+}
+#endif // #ifndef MINIZ_NO_STDIO
+
+mz_bool mz_zip_writer_init_from_reader(mz_zip_archive *pZip, const char *pFilename)
+{
+  mz_zip_internal_state *pState;
+  if ((!pZip) || (!pZip->m_pState) || (pZip->m_zip_mode != MZ_ZIP_MODE_READING))
+    return MZ_FALSE;
+  // No sense in trying to write to an archive that's already at the support max size
+  if ((pZip->m_total_files == 0xFFFF) || ((pZip->m_archive_size + MZ_ZIP_CENTRAL_DIR_HEADER_SIZE + MZ_ZIP_LOCAL_DIR_HEADER_SIZE) > 0xFFFFFFFF))
+    return MZ_FALSE;
+
+  pState = pZip->m_pState;
+
+  if (pState->m_pFile)
+  {
+#ifdef MINIZ_NO_STDIO
+    pFilename; return MZ_FALSE;
+#else
+    // Archive is being read from stdio - try to reopen as writable.
+    if (pZip->m_pIO_opaque != pZip)
+      return MZ_FALSE;
+    if (!pFilename)
+      return MZ_FALSE;
+    pZip->m_pWrite = mz_zip_file_write_func;
+    if (NULL == (pState->m_pFile = MZ_FREOPEN(pFilename, "r+b", pState->m_pFile)))
+    {
+      // The mz_zip_archive is now in a bogus state because pState->m_pFile is NULL, so just close it.
+      mz_zip_reader_end(pZip);
+      return MZ_FALSE;
+    }
+#endif // #ifdef MINIZ_NO_STDIO
+  }
+  else if (pState->m_pMem)
+  {
+    // Archive lives in a memory block. Assume it's from the heap that we can resize using the realloc callback.
+    if (pZip->m_pIO_opaque != pZip)
+      return MZ_FALSE;
+    pState->m_mem_capacity = pState->m_mem_size;
+    pZip->m_pWrite = mz_zip_heap_write_func;
+  }
+  // Archive is being read via a user provided read function - make sure the user has specified a write function too.
+  else if (!pZip->m_pWrite)
+    return MZ_FALSE;
+
+  // Start writing new files at the archive's current central directory location.
+  pZip->m_archive_size = pZip->m_central_directory_file_ofs;
+  pZip->m_zip_mode = MZ_ZIP_MODE_WRITING;
+  pZip->m_central_directory_file_ofs = 0;
+
+  return MZ_TRUE;
+}
+
+mz_bool mz_zip_writer_add_mem(mz_zip_archive *pZip, const char *pArchive_name, const void *pBuf, size_t buf_size, mz_uint level_and_flags)
+{
+  return mz_zip_writer_add_mem_ex(pZip, pArchive_name, pBuf, buf_size, NULL, 0, level_and_flags, 0, 0);
+}
+
+typedef struct
+{
+  mz_zip_archive *m_pZip;
+  mz_uint64 m_cur_archive_file_ofs;
+  mz_uint64 m_comp_size;
+} mz_zip_writer_add_state;
+
+static mz_bool mz_zip_writer_add_put_buf_callback(const void* pBuf, int len, void *pUser)
+{
+  mz_zip_writer_add_state *pState = (mz_zip_writer_add_state *)pUser;
+  if ((int)pState->m_pZip->m_pWrite(pState->m_pZip->m_pIO_opaque, pState->m_cur_archive_file_ofs, pBuf, len) != len)
+    return MZ_FALSE;
+  pState->m_cur_archive_file_ofs += len;
+  pState->m_comp_size += len;
+  return MZ_TRUE;
+}
+
+static mz_bool mz_zip_writer_create_local_dir_header(mz_zip_archive *pZip, mz_uint8 *pDst, mz_uint16 filename_size, mz_uint16 extra_size, mz_uint64 uncomp_size, mz_uint64 comp_size, mz_uint32 uncomp_crc32, mz_uint16 method, mz_uint16 bit_flags, mz_uint16 dos_time, mz_uint16 dos_date)
+{
+  (void)pZip;
+  memset(pDst, 0, MZ_ZIP_LOCAL_DIR_HEADER_SIZE);
+  MZ_WRITE_LE32(pDst + MZ_ZIP_LDH_SIG_OFS, MZ_ZIP_LOCAL_DIR_HEADER_SIG);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_LDH_VERSION_NEEDED_OFS, method ? 20 : 0);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_LDH_BIT_FLAG_OFS, bit_flags);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_LDH_METHOD_OFS, method);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_LDH_FILE_TIME_OFS, dos_time);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_LDH_FILE_DATE_OFS, dos_date);
+  MZ_WRITE_LE32(pDst + MZ_ZIP_LDH_CRC32_OFS, uncomp_crc32);
+  MZ_WRITE_LE32(pDst + MZ_ZIP_LDH_COMPRESSED_SIZE_OFS, comp_size);
+  MZ_WRITE_LE32(pDst + MZ_ZIP_LDH_DECOMPRESSED_SIZE_OFS, uncomp_size);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_LDH_FILENAME_LEN_OFS, filename_size);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_LDH_EXTRA_LEN_OFS, extra_size);
+  return MZ_TRUE;
+}
+
+static mz_bool mz_zip_writer_create_central_dir_header(mz_zip_archive *pZip, mz_uint8 *pDst, mz_uint16 filename_size, mz_uint16 extra_size, mz_uint16 comment_size, mz_uint64 uncomp_size, mz_uint64 comp_size, mz_uint32 uncomp_crc32, mz_uint16 method, mz_uint16 bit_flags, mz_uint16 dos_time, mz_uint16 dos_date, mz_uint64 local_header_ofs, mz_uint32 ext_attributes)
+{
+  (void)pZip;
+  memset(pDst, 0, MZ_ZIP_CENTRAL_DIR_HEADER_SIZE);
+  MZ_WRITE_LE32(pDst + MZ_ZIP_CDH_SIG_OFS, MZ_ZIP_CENTRAL_DIR_HEADER_SIG);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_CDH_VERSION_NEEDED_OFS, method ? 20 : 0);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_CDH_BIT_FLAG_OFS, bit_flags);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_CDH_METHOD_OFS, method);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_CDH_FILE_TIME_OFS, dos_time);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_CDH_FILE_DATE_OFS, dos_date);
+  MZ_WRITE_LE32(pDst + MZ_ZIP_CDH_CRC32_OFS, uncomp_crc32);
+  MZ_WRITE_LE32(pDst + MZ_ZIP_CDH_COMPRESSED_SIZE_OFS, comp_size);
+  MZ_WRITE_LE32(pDst + MZ_ZIP_CDH_DECOMPRESSED_SIZE_OFS, uncomp_size);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_CDH_FILENAME_LEN_OFS, filename_size);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_CDH_EXTRA_LEN_OFS, extra_size);
+  MZ_WRITE_LE16(pDst + MZ_ZIP_CDH_COMMENT_LEN_OFS, comment_size);
+  MZ_WRITE_LE32(pDst + MZ_ZIP_CDH_EXTERNAL_ATTR_OFS, ext_attributes);
+  MZ_WRITE_LE32(pDst + MZ_ZIP_CDH_LOCAL_HEADER_OFS, local_header_ofs);
+  return MZ_TRUE;
+}
+
+static mz_bool mz_zip_writer_add_to_central_dir(mz_zip_archive *pZip, const char *pFilename, mz_uint16 filename_size, const void *pExtra, mz_uint16 extra_size, const void *pComment, mz_uint16 comment_size, mz_uint64 uncomp_size, mz_uint64 comp_size, mz_uint32 uncomp_crc32, mz_uint16 method, mz_uint16 bit_flags, mz_uint16 dos_time, mz_uint16 dos_date, mz_uint64 local_header_ofs, mz_uint32 ext_attributes)
+{
+  mz_zip_internal_state *pState = pZip->m_pState;
+  mz_uint32 central_dir_ofs = (mz_uint32)pState->m_central_dir.m_size;
+  size_t orig_central_dir_size = pState->m_central_dir.m_size;
+  mz_uint8 central_dir_header[MZ_ZIP_CENTRAL_DIR_HEADER_SIZE];
+
+  // No zip64 support yet
+  if ((local_header_ofs > 0xFFFFFFFF) || (((mz_uint64)pState->m_central_dir.m_size + MZ_ZIP_CENTRAL_DIR_HEADER_SIZE + filename_size + extra_size + comment_size) > 0xFFFFFFFF))
+    return MZ_FALSE;
+
+  if (!mz_zip_writer_create_central_dir_header(pZip, central_dir_header, filename_size, extra_size, comment_size, uncomp_size, comp_size, uncomp_crc32, method, bit_flags, dos_time, dos_date, local_header_ofs, ext_attributes))
+    return MZ_FALSE;
+
+  if ((!mz_zip_array_push_back(pZip, &pState->m_central_dir, central_dir_header, MZ_ZIP_CENTRAL_DIR_HEADER_SIZE)) ||
+      (!mz_zip_array_push_back(pZip, &pState->m_central_dir, pFilename, filename_size)) ||
+      (!mz_zip_array_push_back(pZip, &pState->m_central_dir, pExtra, extra_size)) ||
+      (!mz_zip_array_push_back(pZip, &pState->m_central_dir, pComment, comment_size)) ||
+      (!mz_zip_array_push_back(pZip, &pState->m_central_dir_offsets, &central_dir_ofs, 1)))
+  {
+    // Try to push the central directory array back into its original state.
+    mz_zip_array_resize(pZip, &pState->m_central_dir, orig_central_dir_size, MZ_FALSE);
+    return MZ_FALSE;
+  }
+
+  return MZ_TRUE;
+}
+
+static mz_bool mz_zip_writer_validate_archive_name(const char *pArchive_name)
+{
+  // Basic ZIP archive filename validity checks: Valid filenames cannot start with a forward slash, cannot contain a drive letter, and cannot use DOS-style backward slashes.
+  if (*pArchive_name == '/')
+    return MZ_FALSE;
+  while (*pArchive_name)
+  {
+    if ((*pArchive_name == '\\') || (*pArchive_name == ':'))
+      return MZ_FALSE;
+    pArchive_name++;
+  }
+  return MZ_TRUE;
+}
+
+static mz_uint mz_zip_writer_compute_padding_needed_for_file_alignment(mz_zip_archive *pZip)
+{
+  mz_uint32 n;
+  if (!pZip->m_file_offset_alignment)
+    return 0;
+  n = (mz_uint32)(pZip->m_archive_size & (pZip->m_file_offset_alignment - 1));
+  return (pZip->m_file_offset_alignment - n) & (pZip->m_file_offset_alignment - 1);
+}
+
+static mz_bool mz_zip_writer_write_zeros(mz_zip_archive *pZip, mz_uint64 cur_file_ofs, mz_uint32 n)
+{
+  char buf[4096];
+  memset(buf, 0, MZ_MIN(sizeof(buf), n));
+  while (n)
+  {
+    mz_uint32 s = MZ_MIN(sizeof(buf), n);
+    if (pZip->m_pWrite(pZip->m_pIO_opaque, cur_file_ofs, buf, s) != s)
+      return MZ_FALSE;
+    cur_file_ofs += s; n -= s;
+  }
+  return MZ_TRUE;
+}
+
+mz_bool mz_zip_writer_add_mem_ex(mz_zip_archive *pZip, const char *pArchive_name, const void *pBuf, size_t buf_size, const void *pComment, mz_uint16 comment_size, mz_uint level_and_flags, mz_uint64 uncomp_size, mz_uint32 uncomp_crc32)
+{
+  mz_uint16 method = 0, dos_time = 0, dos_date = 0;
+  mz_uint level, ext_attributes = 0, num_alignment_padding_bytes;
+  mz_uint64 local_dir_header_ofs = pZip->m_archive_size, cur_archive_file_ofs = pZip->m_archive_size, comp_size = 0;
+  size_t archive_name_size;
+  mz_uint8 local_dir_header[MZ_ZIP_LOCAL_DIR_HEADER_SIZE];
+  tdefl_compressor *pComp = NULL;
+  mz_bool store_data_uncompressed;
+  mz_zip_internal_state *pState;
+
+  if ((int)level_and_flags < 0)
+    level_and_flags = MZ_DEFAULT_LEVEL;
+  level = level_and_flags & 0xF;
+  store_data_uncompressed = ((!level) || (level_and_flags & MZ_ZIP_FLAG_COMPRESSED_DATA));
+
+  if ((!pZip) || (!pZip->m_pState) || (pZip->m_zip_mode != MZ_ZIP_MODE_WRITING) || ((buf_size) && (!pBuf)) || (!pArchive_name) || ((comment_size) && (!pComment)) || (pZip->m_total_files == 0xFFFF) || (level > MZ_UBER_COMPRESSION))
+    return MZ_FALSE;
+
+  pState = pZip->m_pState;
+
+  if ((!(level_and_flags & MZ_ZIP_FLAG_COMPRESSED_DATA)) && (uncomp_size))
+    return MZ_FALSE;
+  // No zip64 support yet
+  if ((buf_size > 0xFFFFFFFF) || (uncomp_size > 0xFFFFFFFF))
+    return MZ_FALSE;
+  if (!mz_zip_writer_validate_archive_name(pArchive_name))
+    return MZ_FALSE;
+
+#ifndef MINIZ_NO_TIME
+  {
+    time_t cur_time; time(&cur_time);
+    mz_zip_time_to_dos_time(cur_time, &dos_time, &dos_date);
+  }
+#endif // #ifndef MINIZ_NO_TIME
+
+  archive_name_size = strlen(pArchive_name);
+  if (archive_name_size > 0xFFFF)
+    return MZ_FALSE;
+
+  num_alignment_padding_bytes = mz_zip_writer_compute_padding_needed_for_file_alignment(pZip);
+
+  // no zip64 support yet
+  if ((pZip->m_total_files == 0xFFFF) || ((pZip->m_archive_size + num_alignment_padding_bytes + MZ_ZIP_LOCAL_DIR_HEADER_SIZE + MZ_ZIP_CENTRAL_DIR_HEADER_SIZE + comment_size + archive_name_size) > 0xFFFFFFFF))
+    return MZ_FALSE;
+
+  if ((archive_name_size) && (pArchive_name[archive_name_size - 1] == '/'))
+  {
+    // Set DOS Subdirectory attribute bit.
+    ext_attributes |= 0x10;
+    // Subdirectories cannot contain data.
+    if ((buf_size) || (uncomp_size))
+      return MZ_FALSE;
+  }
+
+  // Try to do any allocations before writing to the archive, so if an allocation fails the file remains unmodified. (A good idea if we're doing an in-place modification.)
+  if ((!mz_zip_array_ensure_room(pZip, &pState->m_central_dir, MZ_ZIP_CENTRAL_DIR_HEADER_SIZE + archive_name_size + comment_size)) || (!mz_zip_array_ensure_room(pZip, &pState->m_central_dir_offsets, 1)))
+    return MZ_FALSE;
+
+  if ((!store_data_uncompressed) && (buf_size))
+  {
+    if (NULL == (pComp = (tdefl_compressor *)pZip->m_pAlloc(pZip->m_pAlloc_opaque, 1, sizeof(tdefl_compressor))))
+      return MZ_FALSE;
+  }
+
+  if (!mz_zip_writer_write_zeros(pZip, cur_archive_file_ofs, num_alignment_padding_bytes + sizeof(local_dir_header)))
+  {
+    pZip->m_pFree(pZip->m_pAlloc_opaque, pComp);
+    return MZ_FALSE;
+  }
+  local_dir_header_ofs += num_alignment_padding_bytes;
+  if (pZip->m_file_offset_alignment) { MZ_ASSERT((local_dir_header_ofs & (pZip->m_file_offset_alignment - 1)) == 0); }
+  cur_archive_file_ofs += num_alignment_padding_bytes + sizeof(local_dir_header);
+
+  MZ_CLEAR_OBJ(local_dir_header);
+  if (pZip->m_pWrite(pZip->m_pIO_opaque, cur_archive_file_ofs, pArchive_name, archive_name_size) != archive_name_size)
+  {
+    pZip->m_pFree(pZip->m_pAlloc_opaque, pComp);
+    return MZ_FALSE;
+  }
+  cur_archive_file_ofs += archive_name_size;
+
+  if (!(level_and_flags & MZ_ZIP_FLAG_COMPRESSED_DATA))
+  {
+    uncomp_crc32 = (mz_uint32)mz_crc32(MZ_CRC32_INIT, (const mz_uint8*)pBuf, buf_size);
+    uncomp_size = buf_size;
+    if (uncomp_size <= 3)
+    {
+      level = 0;
+      store_data_uncompressed = MZ_TRUE;
+    }
+  }
+
+  if (store_data_uncompressed)
+  {
+    if (pZip->m_pWrite(pZip->m_pIO_opaque, cur_archive_file_ofs, pBuf, buf_size) != buf_size)
+    {
+      pZip->m_pFree(pZip->m_pAlloc_opaque, pComp);
+      return MZ_FALSE;
+    }
+
+    cur_archive_file_ofs += buf_size;
+    comp_size = buf_size;
+
+    if (level_and_flags & MZ_ZIP_FLAG_COMPRESSED_DATA)
+      method = MZ_DEFLATED;
+  }
+  else if (buf_size)
+  {
+    mz_zip_writer_add_state state;
+
+    state.m_pZip = pZip;
+    state.m_cur_archive_file_ofs = cur_archive_file_ofs;
+    state.m_comp_size = 0;
+
+    if ((tdefl_init(pComp, mz_zip_writer_add_put_buf_callback, &state, tdefl_create_comp_flags_from_zip_params(level, -15, MZ_DEFAULT_STRATEGY)) != TDEFL_STATUS_OKAY) ||
+        (tdefl_compress_buffer(pComp, pBuf, buf_size, TDEFL_FINISH) != TDEFL_STATUS_DONE))
+    {
+      pZip->m_pFree(pZip->m_pAlloc_opaque, pComp);
+      return MZ_FALSE;
+    }
+
+    comp_size = state.m_comp_size;
+    cur_archive_file_ofs = state.m_cur_archive_file_ofs;
+
+    method = MZ_DEFLATED;
+  }
+
+  pZip->m_pFree(pZip->m_pAlloc_opaque, pComp);
+  pComp = NULL;
+
+  // no zip64 support yet
+  if ((comp_size > 0xFFFFFFFF) || (cur_archive_file_ofs > 0xFFFFFFFF))
+    return MZ_FALSE;
+
+  if (!mz_zip_writer_create_local_dir_header(pZip, local_dir_header, (mz_uint16)archive_name_size, 0, uncomp_size, comp_size, uncomp_crc32, method, 0, dos_time, dos_date))
+    return MZ_FALSE;
+
+  if (pZip->m_pWrite(pZip->m_pIO_opaque, local_dir_header_ofs, local_dir_header, sizeof(local_dir_header)) != sizeof(local_dir_header))
+    return MZ_FALSE;
+
+  if (!mz_zip_writer_add_to_central_dir(pZip, pArchive_name, (mz_uint16)archive_name_size, NULL, 0, pComment, comment_size, uncomp_size, comp_size, uncomp_crc32, method, 0, dos_time, dos_date, local_dir_header_ofs, ext_attributes))
+    return MZ_FALSE;
+
+  pZip->m_total_files++;
+  pZip->m_archive_size = cur_archive_file_ofs;
+
+  return MZ_TRUE;
+}
+
+#ifndef MINIZ_NO_STDIO
+mz_bool mz_zip_writer_add_file(mz_zip_archive *pZip, const char *pArchive_name, const char *pSrc_filename, const void *pComment, mz_uint16 comment_size, mz_uint level_and_flags)
+{
+  mz_uint uncomp_crc32 = MZ_CRC32_INIT, level, num_alignment_padding_bytes;
+  mz_uint16 method = 0, dos_time = 0, dos_date = 0, ext_attributes = 0;
+  mz_uint64 local_dir_header_ofs = pZip->m_archive_size, cur_archive_file_ofs = pZip->m_archive_size, uncomp_size = 0, comp_size = 0;
+  size_t archive_name_size;
+  mz_uint8 local_dir_header[MZ_ZIP_LOCAL_DIR_HEADER_SIZE];
+  MZ_FILE *pSrc_file = NULL;
+
+  if ((int)level_and_flags < 0)
+    level_and_flags = MZ_DEFAULT_LEVEL;
+  level = level_and_flags & 0xF;
+
+  if ((!pZip) || (!pZip->m_pState) || (pZip->m_zip_mode != MZ_ZIP_MODE_WRITING) || (!pArchive_name) || ((comment_size) && (!pComment)) || (level > MZ_UBER_COMPRESSION))
+    return MZ_FALSE;
+  if (level_and_flags & MZ_ZIP_FLAG_COMPRESSED_DATA)
+    return MZ_FALSE;
+  if (!mz_zip_writer_validate_archive_name(pArchive_name))
+    return MZ_FALSE;
+
+  archive_name_size = strlen(pArchive_name);
+  if (archive_name_size > 0xFFFF)
+    return MZ_FALSE;
+
+  num_alignment_padding_bytes = mz_zip_writer_compute_padding_needed_for_file_alignment(pZip);
+
+  // no zip64 support yet
+  if ((pZip->m_total_files == 0xFFFF) || ((pZip->m_archive_size + num_alignment_padding_bytes + MZ_ZIP_LOCAL_DIR_HEADER_SIZE + MZ_ZIP_CENTRAL_DIR_HEADER_SIZE + comment_size + archive_name_size) > 0xFFFFFFFF))
+    return MZ_FALSE;
+
+  if (!mz_zip_get_file_modified_time(pSrc_filename, &dos_time, &dos_date))
+    return MZ_FALSE;
+    
+  pSrc_file = MZ_FOPEN(pSrc_filename, "rb");
+  if (!pSrc_file)
+    return MZ_FALSE;
+  MZ_FSEEK64(pSrc_file, 0, SEEK_END);
+  uncomp_size = MZ_FTELL64(pSrc_file);
+  MZ_FSEEK64(pSrc_file, 0, SEEK_SET);
+
+  if (uncomp_size > 0xFFFFFFFF)
+  {
+    // No zip64 support yet
+    MZ_FCLOSE(pSrc_file);
+    return MZ_FALSE;
+  }
+  if (uncomp_size <= 3)
+    level = 0;
+
+  if (!mz_zip_writer_write_zeros(pZip, cur_archive_file_ofs, num_alignment_padding_bytes + sizeof(local_dir_header)))
+  {
+    MZ_FCLOSE(pSrc_file);
+    return MZ_FALSE;
+  }
+  local_dir_header_ofs += num_alignment_padding_bytes;
+  if (pZip->m_file_offset_alignment) { MZ_ASSERT((local_dir_header_ofs & (pZip->m_file_offset_alignment - 1)) == 0); }
+  cur_archive_file_ofs += num_alignment_padding_bytes + sizeof(local_dir_header);
+
+  MZ_CLEAR_OBJ(local_dir_header);
+  if (pZip->m_pWrite(pZip->m_pIO_opaque, cur_archive_file_ofs, pArchive_name, archive_name_size) != archive_name_size)
+  {
+    MZ_FCLOSE(pSrc_file);
+    return MZ_FALSE;
+  }
+  cur_archive_file_ofs += archive_name_size;
+
+  if (uncomp_size)
+  {
+    mz_uint64 uncomp_remaining = uncomp_size;
+    void *pRead_buf = pZip->m_pAlloc(pZip->m_pAlloc_opaque, 1, MZ_ZIP_MAX_IO_BUF_SIZE);
+    if (!pRead_buf)
+    {
+      MZ_FCLOSE(pSrc_file);
+      return MZ_FALSE;
+    }
+
+    if (!level)
+    {
+      while (uncomp_remaining)
+      {
+        mz_uint n = (mz_uint)MZ_MIN(MZ_ZIP_MAX_IO_BUF_SIZE, uncomp_remaining);
+        if ((MZ_FREAD(pRead_buf, 1, n, pSrc_file) != n) || (pZip->m_pWrite(pZip->m_pIO_opaque, cur_archive_file_ofs, pRead_buf, n) != n))
+        {
+          pZip->m_pFree(pZip->m_pAlloc_opaque, pRead_buf);
+          MZ_FCLOSE(pSrc_file);
+          return MZ_FALSE;
+        }
+        uncomp_crc32 = (mz_uint32)mz_crc32(uncomp_crc32, (const mz_uint8 *)pRead_buf, n);
+        uncomp_remaining -= n;
+        cur_archive_file_ofs += n;
+      }
+      comp_size = uncomp_size;
+    }
+    else
+    {
+      mz_bool result = MZ_FALSE;
+      mz_zip_writer_add_state state;
+      tdefl_compressor *pComp = (tdefl_compressor *)pZip->m_pAlloc(pZip->m_pAlloc_opaque, 1, sizeof(tdefl_compressor));
+      if (!pComp)
+      {
+        pZip->m_pFree(pZip->m_pAlloc_opaque, pRead_buf);
+        MZ_FCLOSE(pSrc_file);
+        return MZ_FALSE;
+      }
+
+      state.m_pZip = pZip;
+      state.m_cur_archive_file_ofs = cur_archive_file_ofs;
+      state.m_comp_size = 0;
+
+      if (tdefl_init(pComp, mz_zip_writer_add_put_buf_callback, &state, tdefl_create_comp_flags_from_zip_params(level, -15, MZ_DEFAULT_STRATEGY)) != TDEFL_STATUS_OKAY)
+      {
+        pZip->m_pFree(pZip->m_pAlloc_opaque, pComp);
+        pZip->m_pFree(pZip->m_pAlloc_opaque, pRead_buf);
+        MZ_FCLOSE(pSrc_file);
+        return MZ_FALSE;
+      }
+
+      for ( ; ; )
+      {
+        size_t in_buf_size = (mz_uint32)MZ_MIN(uncomp_remaining, MZ_ZIP_MAX_IO_BUF_SIZE);
+        tdefl_status status;
+
+        if (MZ_FREAD(pRead_buf, 1, in_buf_size, pSrc_file) != in_buf_size)
+          break;
+
+        uncomp_crc32 = (mz_uint32)mz_crc32(uncomp_crc32, (const mz_uint8 *)pRead_buf, in_buf_size);
+        uncomp_remaining -= in_buf_size;
+
+        status = tdefl_compress_buffer(pComp, pRead_buf, in_buf_size, uncomp_remaining ? TDEFL_NO_FLUSH : TDEFL_FINISH);
+        if (status == TDEFL_STATUS_DONE)
+        {
+          result = MZ_TRUE;
+          break;
+        }
+        else if (status != TDEFL_STATUS_OKAY)
+          break;
+      }
+
+      pZip->m_pFree(pZip->m_pAlloc_opaque, pComp);
+
+      if (!result)
+      {
+        pZip->m_pFree(pZip->m_pAlloc_opaque, pRead_buf);
+        MZ_FCLOSE(pSrc_file);
+        return MZ_FALSE;
+      }
+
+      comp_size = state.m_comp_size;
+      cur_archive_file_ofs = state.m_cur_archive_file_ofs;
+
+      method = MZ_DEFLATED;
+    }
+
+    pZip->m_pFree(pZip->m_pAlloc_opaque, pRead_buf);
+  }
+
+  MZ_FCLOSE(pSrc_file); pSrc_file = NULL;
+
+  // no zip64 support yet
+  if ((comp_size > 0xFFFFFFFF) || (cur_archive_file_ofs > 0xFFFFFFFF))
+    return MZ_FALSE;
+
+  if (!mz_zip_writer_create_local_dir_header(pZip, local_dir_header, (mz_uint16)archive_name_size, 0, uncomp_size, comp_size, uncomp_crc32, method, 0, dos_time, dos_date))
+    return MZ_FALSE;
+
+  if (pZip->m_pWrite(pZip->m_pIO_opaque, local_dir_header_ofs, local_dir_header, sizeof(local_dir_header)) != sizeof(local_dir_header))
+    return MZ_FALSE;
+
+  if (!mz_zip_writer_add_to_central_dir(pZip, pArchive_name, (mz_uint16)archive_name_size, NULL, 0, pComment, comment_size, uncomp_size, comp_size, uncomp_crc32, method, 0, dos_time, dos_date, local_dir_header_ofs, ext_attributes))
+    return MZ_FALSE;
+
+  pZip->m_total_files++;
+  pZip->m_archive_size = cur_archive_file_ofs;
+
+  return MZ_TRUE;
+}
+#endif // #ifndef MINIZ_NO_STDIO
+
+mz_bool mz_zip_writer_add_from_zip_reader(mz_zip_archive *pZip, mz_zip_archive *pSource_zip, mz_uint file_index)
+{
+  mz_uint n, bit_flags, num_alignment_padding_bytes;
+  mz_uint64 comp_bytes_remaining, local_dir_header_ofs;
+  mz_uint64 cur_src_file_ofs, cur_dst_file_ofs;
+  mz_uint32 local_header_u32[(MZ_ZIP_LOCAL_DIR_HEADER_SIZE + sizeof(mz_uint32) - 1) / sizeof(mz_uint32)]; mz_uint8 *pLocal_header = (mz_uint8 *)local_header_u32;
+  mz_uint8 central_header[MZ_ZIP_CENTRAL_DIR_HEADER_SIZE];
+  size_t orig_central_dir_size;
+  mz_zip_internal_state *pState;
+  void *pBuf; const mz_uint8 *pSrc_central_header;
+
+  if ((!pZip) || (!pZip->m_pState) || (pZip->m_zip_mode != MZ_ZIP_MODE_WRITING))
+    return MZ_FALSE;
+  if (NULL == (pSrc_central_header = mz_zip_reader_get_cdh(pSource_zip, file_index)))
+    return MZ_FALSE;
+  pState = pZip->m_pState;
+
+  num_alignment_padding_bytes = mz_zip_writer_compute_padding_needed_for_file_alignment(pZip);
+
+  // no zip64 support yet
+  if ((pZip->m_total_files == 0xFFFF) || ((pZip->m_archive_size + num_alignment_padding_bytes + MZ_ZIP_LOCAL_DIR_HEADER_SIZE + MZ_ZIP_CENTRAL_DIR_HEADER_SIZE) > 0xFFFFFFFF))
+    return MZ_FALSE;
+
+  cur_src_file_ofs = MZ_READ_LE32(pSrc_central_header + MZ_ZIP_CDH_LOCAL_HEADER_OFS);
+  cur_dst_file_ofs = pZip->m_archive_size;
+
+  if (pSource_zip->m_pRead(pSource_zip->m_pIO_opaque, cur_src_file_ofs, pLocal_header, MZ_ZIP_LOCAL_DIR_HEADER_SIZE) != MZ_ZIP_LOCAL_DIR_HEADER_SIZE)
+    return MZ_FALSE;
+  if (MZ_READ_LE32(pLocal_header) != MZ_ZIP_LOCAL_DIR_HEADER_SIG)
+    return MZ_FALSE;
+  cur_src_file_ofs += MZ_ZIP_LOCAL_DIR_HEADER_SIZE;
+
+  if (!mz_zip_writer_write_zeros(pZip, cur_dst_file_ofs, num_alignment_padding_bytes))
+    return MZ_FALSE;
+  cur_dst_file_ofs += num_alignment_padding_bytes;
+  local_dir_header_ofs = cur_dst_file_ofs;
+  if (pZip->m_file_offset_alignment) { MZ_ASSERT((local_dir_header_ofs & (pZip->m_file_offset_alignment - 1)) == 0); }
+
+  if (pZip->m_pWrite(pZip->m_pIO_opaque, cur_dst_file_ofs, pLocal_header, MZ_ZIP_LOCAL_DIR_HEADER_SIZE) != MZ_ZIP_LOCAL_DIR_HEADER_SIZE)
+    return MZ_FALSE;
+  cur_dst_file_ofs += MZ_ZIP_LOCAL_DIR_HEADER_SIZE;
+
+  n = MZ_READ_LE16(pLocal_header + MZ_ZIP_LDH_FILENAME_LEN_OFS) + MZ_READ_LE16(pLocal_header + MZ_ZIP_LDH_EXTRA_LEN_OFS);
+  comp_bytes_remaining = n + MZ_READ_LE32(pSrc_central_header + MZ_ZIP_CDH_COMPRESSED_SIZE_OFS);
+
+  if (NULL == (pBuf = pZip->m_pAlloc(pZip->m_pAlloc_opaque, 1, (size_t)MZ_MAX(sizeof(mz_uint32) * 4, MZ_MIN(MZ_ZIP_MAX_IO_BUF_SIZE, comp_bytes_remaining)))))
+    return MZ_FALSE;
+
+  while (comp_bytes_remaining)
+  {
+    n = (mz_uint)MZ_MIN(MZ_ZIP_MAX_IO_BUF_SIZE, comp_bytes_remaining);
+    if (pSource_zip->m_pRead(pSource_zip->m_pIO_opaque, cur_src_file_ofs, pBuf, n) != n)
+    {
+      pZip->m_pFree(pZip->m_pAlloc_opaque, pBuf);
+      return MZ_FALSE;
+    }
+    cur_src_file_ofs += n;
+
+    if (pZip->m_pWrite(pZip->m_pIO_opaque, cur_dst_file_ofs, pBuf, n) != n)
+    {
+      pZip->m_pFree(pZip->m_pAlloc_opaque, pBuf);
+      return MZ_FALSE;
+    }
+    cur_dst_file_ofs += n;
+
+    comp_bytes_remaining -= n;
+  }
+
+  bit_flags = MZ_READ_LE16(pLocal_header + MZ_ZIP_LDH_BIT_FLAG_OFS);
+  if (bit_flags & 8)
+  {
+    // Copy data descriptor
+    if (pSource_zip->m_pRead(pSource_zip->m_pIO_opaque, cur_src_file_ofs, pBuf, sizeof(mz_uint32) * 4) != sizeof(mz_uint32) * 4)
+    {
+      pZip->m_pFree(pZip->m_pAlloc_opaque, pBuf);
+      return MZ_FALSE;
+    }
+
+    n = sizeof(mz_uint32) * ((MZ_READ_LE32(pBuf) == 0x08074b50) ? 4 : 3);
+    if (pZip->m_pWrite(pZip->m_pIO_opaque, cur_dst_file_ofs, pBuf, n) != n)
+    {
+      pZip->m_pFree(pZip->m_pAlloc_opaque, pBuf);
+      return MZ_FALSE;
+    }
+
+    cur_src_file_ofs += n;
+    cur_dst_file_ofs += n;
+  }
+  pZip->m_pFree(pZip->m_pAlloc_opaque, pBuf);
+
+  // no zip64 support yet
+  if (cur_dst_file_ofs > 0xFFFFFFFF)
+    return MZ_FALSE;
+
+  orig_central_dir_size = pState->m_central_dir.m_size;
+
+  memcpy(central_header, pSrc_central_header, MZ_ZIP_CENTRAL_DIR_HEADER_SIZE);
+  MZ_WRITE_LE32(central_header + MZ_ZIP_CDH_LOCAL_HEADER_OFS, local_dir_header_ofs);
+  if (!mz_zip_array_push_back(pZip, &pState->m_central_dir, central_header, MZ_ZIP_CENTRAL_DIR_HEADER_SIZE))
+    return MZ_FALSE;
+
+  n = MZ_READ_LE16(pSrc_central_header + MZ_ZIP_CDH_FILENAME_LEN_OFS) + MZ_READ_LE16(pSrc_central_header + MZ_ZIP_CDH_EXTRA_LEN_OFS) + MZ_READ_LE16(pSrc_central_header + MZ_ZIP_CDH_COMMENT_LEN_OFS);
+  if (!mz_zip_array_push_back(pZip, &pState->m_central_dir, pSrc_central_header + MZ_ZIP_CENTRAL_DIR_HEADER_SIZE, n))
+  {
+    mz_zip_array_resize(pZip, &pState->m_central_dir, orig_central_dir_size, MZ_FALSE);
+    return MZ_FALSE;
+  }
+
+  if (pState->m_central_dir.m_size > 0xFFFFFFFF)
+    return MZ_FALSE;
+  n = (mz_uint32)orig_central_dir_size;
+  if (!mz_zip_array_push_back(pZip, &pState->m_central_dir_offsets, &n, 1))
+  {
+    mz_zip_array_resize(pZip, &pState->m_central_dir, orig_central_dir_size, MZ_FALSE);
+    return MZ_FALSE;
+  }
+
+  pZip->m_total_files++;
+  pZip->m_archive_size = cur_dst_file_ofs;
+
+  return MZ_TRUE;
+}
+
+mz_bool mz_zip_writer_finalize_archive(mz_zip_archive *pZip)
+{
+  mz_zip_internal_state *pState;
+  mz_uint64 central_dir_ofs, central_dir_size;
+  mz_uint8 hdr[MZ_ZIP_END_OF_CENTRAL_DIR_HEADER_SIZE];
+
+  if ((!pZip) || (!pZip->m_pState) || (pZip->m_zip_mode != MZ_ZIP_MODE_WRITING))
+    return MZ_FALSE;
+
+  pState = pZip->m_pState;
+
+  // no zip64 support yet
+  if ((pZip->m_total_files > 0xFFFF) || ((pZip->m_archive_size + pState->m_central_dir.m_size + MZ_ZIP_END_OF_CENTRAL_DIR_HEADER_SIZE) > 0xFFFFFFFF))
+    return MZ_FALSE;
+
+  central_dir_ofs = 0;
+  central_dir_size = 0;
+  if (pZip->m_total_files)
+  {
+    // Write central directory
+    central_dir_ofs = pZip->m_archive_size;
+    central_dir_size = pState->m_central_dir.m_size;
+    pZip->m_central_directory_file_ofs = central_dir_ofs;
+    if (pZip->m_pWrite(pZip->m_pIO_opaque, central_dir_ofs, pState->m_central_dir.m_p, (size_t)central_dir_size) != central_dir_size)
+      return MZ_FALSE;
+    pZip->m_archive_size += central_dir_size;
+  }
+
+  // Write end of central directory record
+  MZ_CLEAR_OBJ(hdr);
+  MZ_WRITE_LE32(hdr + MZ_ZIP_ECDH_SIG_OFS, MZ_ZIP_END_OF_CENTRAL_DIR_HEADER_SIG);
+  MZ_WRITE_LE16(hdr + MZ_ZIP_ECDH_CDIR_NUM_ENTRIES_ON_DISK_OFS, pZip->m_total_files);
+  MZ_WRITE_LE16(hdr + MZ_ZIP_ECDH_CDIR_TOTAL_ENTRIES_OFS, pZip->m_total_files);
+  MZ_WRITE_LE32(hdr + MZ_ZIP_ECDH_CDIR_SIZE_OFS, central_dir_size);
+  MZ_WRITE_LE32(hdr + MZ_ZIP_ECDH_CDIR_OFS_OFS, central_dir_ofs);
+
+  if (pZip->m_pWrite(pZip->m_pIO_opaque, pZip->m_archive_size, hdr, sizeof(hdr)) != sizeof(hdr))
+    return MZ_FALSE;
+#ifndef MINIZ_NO_STDIO
+  if ((pState->m_pFile) && (MZ_FFLUSH(pState->m_pFile) == EOF))
+    return MZ_FALSE;
+#endif // #ifndef MINIZ_NO_STDIO
+
+  pZip->m_archive_size += sizeof(hdr);
+
+  pZip->m_zip_mode = MZ_ZIP_MODE_WRITING_HAS_BEEN_FINALIZED;
+  return MZ_TRUE;
+}
+
+mz_bool mz_zip_writer_finalize_heap_archive(mz_zip_archive *pZip, void **pBuf, size_t *pSize)
+{
+  if ((!pZip) || (!pZip->m_pState) || (!pBuf) || (!pSize))
+    return MZ_FALSE;
+  if (pZip->m_pWrite != mz_zip_heap_write_func)
+    return MZ_FALSE;
+  if (!mz_zip_writer_finalize_archive(pZip))
+    return MZ_FALSE;
+
+  *pBuf = pZip->m_pState->m_pMem;
+  *pSize = pZip->m_pState->m_mem_size;
+  pZip->m_pState->m_pMem = NULL;
+  pZip->m_pState->m_mem_size = pZip->m_pState->m_mem_capacity = 0;
+  return MZ_TRUE;
+}
+
+mz_bool mz_zip_writer_end(mz_zip_archive *pZip)
+{
+  mz_zip_internal_state *pState;
+  mz_bool status = MZ_TRUE;
+  if ((!pZip) || (!pZip->m_pState) || (!pZip->m_pAlloc) || (!pZip->m_pFree) || ((pZip->m_zip_mode != MZ_ZIP_MODE_WRITING) && (pZip->m_zip_mode != MZ_ZIP_MODE_WRITING_HAS_BEEN_FINALIZED)))
+    return MZ_FALSE;
+
+  pState = pZip->m_pState;
+  pZip->m_pState = NULL;
+  mz_zip_array_clear(pZip, &pState->m_central_dir);
+  mz_zip_array_clear(pZip, &pState->m_central_dir_offsets);
+  mz_zip_array_clear(pZip, &pState->m_sorted_central_dir_offsets);
+
+#ifndef MINIZ_NO_STDIO
+  if (pState->m_pFile)
+  {
+    MZ_FCLOSE(pState->m_pFile);
+    pState->m_pFile = NULL;
+  }
+#endif // #ifndef MINIZ_NO_STDIO
+
+  if ((pZip->m_pWrite == mz_zip_heap_write_func) && (pState->m_pMem))
+  {
+    pZip->m_pFree(pZip->m_pAlloc_opaque, pState->m_pMem);
+    pState->m_pMem = NULL;
+  }
+
+  pZip->m_pFree(pZip->m_pAlloc_opaque, pState);
+  pZip->m_zip_mode = MZ_ZIP_MODE_INVALID;
+  return status;
+}
+
+#ifndef MINIZ_NO_STDIO
+mz_bool mz_zip_add_mem_to_archive_file_in_place(const char *pZip_filename, const char *pArchive_name, const void *pBuf, size_t buf_size, const void *pComment, mz_uint16 comment_size, mz_uint level_and_flags)
+{
+  mz_bool status, created_new_archive = MZ_FALSE;
+  mz_zip_archive zip_archive;
+  struct MZ_FILE_STAT_STRUCT file_stat;
+  MZ_CLEAR_OBJ(zip_archive);
+  if ((int)level_and_flags < 0)
+     level_and_flags = MZ_DEFAULT_LEVEL;
+  if ((!pZip_filename) || (!pArchive_name) || ((buf_size) && (!pBuf)) || ((comment_size) && (!pComment)) || ((level_and_flags & 0xF) > MZ_UBER_COMPRESSION))
+    return MZ_FALSE;
+  if (!mz_zip_writer_validate_archive_name(pArchive_name))
+    return MZ_FALSE;
+  if (MZ_FILE_STAT(pZip_filename, &file_stat) != 0)
+  {
+    // Create a new archive.
+    if (!mz_zip_writer_init_file(&zip_archive, pZip_filename, 0))
+      return MZ_FALSE;
+    created_new_archive = MZ_TRUE;
+  }
+  else
+  {
+    // Append to an existing archive.
+    if (!mz_zip_reader_init_file(&zip_archive, pZip_filename, level_and_flags | MZ_ZIP_FLAG_DO_NOT_SORT_CENTRAL_DIRECTORY))
+      return MZ_FALSE;
+    if (!mz_zip_writer_init_from_reader(&zip_archive, pZip_filename))
+    {
+      mz_zip_reader_end(&zip_archive);
+      return MZ_FALSE;
+    }
+  }
+  status = mz_zip_writer_add_mem_ex(&zip_archive, pArchive_name, pBuf, buf_size, pComment, comment_size, level_and_flags, 0, 0);
+  // Always finalize, even if adding failed for some reason, so we have a valid central directory. (This may not always succeed, but we can try.)
+  if (!mz_zip_writer_finalize_archive(&zip_archive))
+    status = MZ_FALSE;
+  if (!mz_zip_writer_end(&zip_archive))
+    status = MZ_FALSE;
+  if ((!status) && (created_new_archive))
+  {
+    // It's a new archive and something went wrong, so just delete it.
+    int ignoredStatus = MZ_DELETE_FILE(pZip_filename);
+    (void)ignoredStatus;
+  }
+  return status;
+}
+
+void *mz_zip_extract_archive_file_to_heap(const char *pZip_filename, const char *pArchive_name, size_t *pSize, mz_uint flags)
+{
+  int file_index;
+  mz_zip_archive zip_archive;
+  void *p = NULL;
+
+  if (pSize)
+    *pSize = 0;
+
+  if ((!pZip_filename) || (!pArchive_name))
+    return NULL;
+
+  MZ_CLEAR_OBJ(zip_archive);
+  if (!mz_zip_reader_init_file(&zip_archive, pZip_filename, flags | MZ_ZIP_FLAG_DO_NOT_SORT_CENTRAL_DIRECTORY))
+    return NULL;
+
+  if ((file_index = mz_zip_reader_locate_file(&zip_archive, pArchive_name, NULL, flags)) >= 0)
+    p = mz_zip_reader_extract_to_heap(&zip_archive, file_index, pSize, flags);
+
+  mz_zip_reader_end(&zip_archive);
+  return p;
+}
+
+#endif // #ifndef MINIZ_NO_STDIO
+
+#endif // #ifndef MINIZ_NO_ARCHIVE_WRITING_APIS
+
+#endif // #ifndef MINIZ_NO_ARCHIVE_APIS
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // MINIZ_HEADER_FILE_ONLY
+
+/*
+  This is free and unencumbered software released into the public domain.
+
+  Anyone is free to copy, modify, publish, use, compile, sell, or
+  distribute this software, either in source code form or as a compiled
+  binary, for any purpose, commercial or non-commercial, and by any
+  means.
+
+  In jurisdictions that recognize copyright laws, the author or authors
+  of this software dedicate any and all copyright interest in the
+  software to the public domain. We make this dedication for the benefit
+  of the public at large and to the detriment of our heirs and
+  successors. We intend this dedication to be an overt act of
+  relinquishment in perpetuity of all present and future rights to this
+  software under copyright law.
+
+  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+  IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+  OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+  ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+  OTHER DEALINGS IN THE SOFTWARE.
+
+  For more information, please refer to <http://unlicense.org/>
+*/
@@ -0,0 +1,7 @@
+#ifndef MINIZ_H_
+#define MINIZ_H_
+
+#define MINIZ_HEADER_FILE_ONLY
+#include "miniz.c"
+
+#endif
@@ -34,6 +34,8 @@ Zeev Tarantov <zeev.tarantov@gmail.com>
 
 File modified for Sereal by
 Steffen Mueller <smueller@cpan.org>
+Yves Orton <demerphq@gmail.com>
+
 */
 
 #ifndef CSNAPPY_INTERNAL_USERSPACE_H_
@@ -171,6 +173,64 @@ Albert Lee
 #define __LITTLE_ENDIAN LITTLE_ENDIAN
 #define __BIG_ENDIAN BIG_ENDIAN
 
+#elif defined(__hpux)
+
+#ifdef __LP64__
+#define __LITTLE_ENDIAN 12345678
+#define __BIG_ENDIAN 87654321
+#define int64_t long
+#else
+#define __LITTLE_ENDIAN 1234
+#define __BIG_ENDIAN 4321
+#define int64_t long long
+#endif
+
+#define __BYTE_ORDER __BIG_ENDIAN /* HP-UX always */
+#define int32_t int
+#define int16_t short
+
+#define __SNAPPY_STRICT_ALIGN
+
+#elif defined(__s390x__) || defined(__zarch__) || defined(__SYSC_ZARCH__)
+
+#ifndef __BIG_ENDIAN
+#define __BIG_ENDIAN    87654321
+#endif
+#ifndef __LITTLE_ENDIAN
+#define __LITTLE_ENDIAN 12345678
+#endif
+#ifndef __BYTE_ORDER
+#define __BYTE_ORDER __BIG_ENDIAN
+#endif
+
+#define __SNAPPY_STRICT_ALIGN
+
+#endif
+
+#ifndef bswap_16
+#define bswap_16(x) \
+  (((uint16_t)(x) & 0xFF00) >> 8 | \
+   ((uint16_t)(x) & 0x00FF) << 8)
+#endif
+
+#ifndef bswap_32
+#define bswap_32(x) \
+  (((uint32_t)(x) & 0xFF000000) >> 24 | \
+   ((uint32_t)(x) & 0x00FF0000) >>  8 | \
+   ((uint32_t)(x) & 0x0000FF00) <<  8 | \
+   ((uint32_t)(x) & 0x000000FF) << 24)
+#endif
+
+#ifndef bswap_64
+#define bswap_64(x) \
+  (((uint64_t)(x) & 0xFF00000000000000) >> 56 | \
+   ((uint64_t)(x) & 0x00FF000000000000) >> 40 | \
+   ((uint64_t)(x) & 0x0000FF0000000000) >> 24 | \
+   ((uint64_t)(x) & 0x000000FF00000000) >>  8 | \
+   ((uint64_t)(x) & 0x00000000FF000000) <<  8 | \
+   ((uint64_t)(x) & 0x0000000000FF0000) << 24 | \
+   ((uint64_t)(x) & 0x000000000000FF00) << 40 | \
+   ((uint64_t)(x) & 0x00000000000000FF) << 56)
 #endif
 
 
@@ -219,8 +279,153 @@ static INLINE void UNALIGNED_STORE64(void *p, uint64_t v)
 	ptr->x = v;
 }
 
+#elif defined(__SNAPPY_STRICT_ALIGN) || defined(__sparc) || defined(__sparc__) /* strict architectures */
+
+/* For these platforms, there really are no unaligned loads/stores.
+ * Read/write everything as uint8_t. Smart compilers might recognize
+ * these patterns and generate something smart. */
+
+/* Possible future enhancement: see if the ptr is evenly divisible
+ * (as uintNN_t) by 2/4/8, and if so, do the cast-as-uintNN_t-ptr-
+ * and-deref-as-uintNN_t.  Balancing act: adding the branch
+ * will slow things down, while reading/writing aligned might speed
+ * things up. */
+
+#if __BYTE_ORDER == __BIG_ENDIAN
+
+static INLINE uint16_t UNALIGNED_LOAD16(const void *p)
+{
+	return
+          (uint16_t)(((uint8_t*)p)[0]) << 8 |
+          (uint16_t)(((uint8_t*)p)[1]);
+}
+
+static INLINE uint32_t UNALIGNED_LOAD32(const void *p)
+{
+	return
+          (uint32_t)(((uint8_t*)p)[0]) << 24 |
+          (uint32_t)(((uint8_t*)p)[1]) << 16 |
+          (uint32_t)(((uint8_t*)p)[2]) <<  8 |
+          (uint32_t)(((uint8_t*)p)[3]);
+}
+
+static INLINE uint64_t UNALIGNED_LOAD64(const void *p)
+{
+	return
+          (uint64_t)((uint8_t*)p)[0] << 56 |
+          (uint64_t)((uint8_t*)p)[1] << 48 |
+          (uint64_t)((uint8_t*)p)[2] << 40 |
+          (uint64_t)((uint8_t*)p)[3] << 32 |
+          (uint64_t)((uint8_t*)p)[4] << 24 |
+          (uint64_t)((uint8_t*)p)[5] << 16 |
+          (uint64_t)((uint8_t*)p)[5] <<  8 |
+          (uint64_t)((uint8_t*)p)[7];
+}
+
+static INLINE void UNALIGNED_STORE16(void *p, uint16_t v)
+{
+	uint8_t* s = (uint8_t*)p;
+	s[0] = (v & 0xFF00) >> 8;
+	s[1] = (v & 0x00FF);
+}
+
+static INLINE void UNALIGNED_STORE32(void *p, uint32_t v)
+{
+	uint8_t* s = (uint8_t*)p;
+	s[0] = (v & 0xFF000000) >> 24;
+	s[1] = (v & 0x00FF0000) >> 16;
+	s[2] = (v & 0x0000FF00) >>  8;
+	s[3] = (v & 0x000000FF);
+}
+
+static INLINE void UNALIGNED_STORE64(void *p, uint64_t v)
+{
+	uint8_t* s = (uint8_t*)p;
+	s[0] = (v & 0xFF00000000000000) >> 56;
+	s[1] = (v & 0x00FF000000000000) >> 48;
+	s[2] = (v & 0x0000FF0000000000) >> 40;
+	s[3] = (v & 0x000000FF00000000) >> 32;
+	s[4] = (v & 0x00000000FF000000) >> 24;
+	s[5] = (v & 0x0000000000FF0000) >> 16;
+	s[6] = (v & 0x000000000000FF00) >>  8;
+	s[7] = (v & 0x00000000000000FF);
+}
+
+#endif /* #if __BYTE_ORDER == __BIG_ENDIAN */
+
+#if __BYTE_ORDER == __LITTLE_ENDIAN
+
+static INLINE uint16_t UNALIGNED_LOAD16(const void *p)
+{
+	return
+          (uint16_t)(((uint8_t*)p)[1]) << 8) |
+          (uint16_t)(((uint8_t*)p)[0]);
+}
+
+static INLINE uint32_t UNALIGNED_LOAD32(const void *p)
+{
+	return
+          (uint32_t)(((uint8_t*)p)[3]) << 24 |
+          (uint32_t)(((uint8_t*)p)[2]) << 16 |
+          (uint32_t)(((uint8_t*)p)[1]) <<  8 |
+          (uint32_t)(((uint8_t*)p)[0]);
+}
+
+static INLINE uint64_t UNALIGNED_LOAD64(const void *p)
+{
+	return
+          (uint64_t)(((uint8_t*)p)[7]) << 56 |
+          (uint64_t)(((uint8_t*)p)[6]) << 48 |
+          (uint64_t)(((uint8_t*)p)[5]) << 40 |
+          (uint64_t)(((uint8_t*)p)[4]) << 32 |
+          (uint64_t)(((uint8_t*)p)[3]) << 24 |
+          (uint64_t)(((uint8_t*)p)[2]) << 16 |
+          (uint64_t)(((uint8_t*)p)[1]) <<  8 |
+          (uint64_t)(((uint8_t*)p)[0]);
+}
+
+static INLINE void UNALIGNED_STORE16(void *p, uint16_t v)
+{
+	uint8_t* s = (uint8_t*)p;
+	s[1] = (v & 0xFF00) >> 8;
+	s[0] = (v & 0x00FF);
+}
+
+static INLINE void UNALIGNED_STORE32(void *p, uint32_t v)
+{
+	uint8_t* s = (uint8_t*)p;
+	s[3] = (v & 0xFF000000) >> 24;
+	s[2] = (v & 0x00FF0000) >> 16;
+	s[1] = (v & 0x0000FF00) >>  8;
+	s[0] = (v & 0x000000FF);
+}
+
+static INLINE void UNALIGNED_STORE64(void *p, uint64_t v)
+{
+	uint8_t* s = (uint8_t*)p;
+	s[7] = (v & 0xFF00000000000000) >> 56;
+	s[6] = (v & 0x00FF000000000000) >> 48;
+	s[5] = (v & 0x0000FF0000000000) >> 40;
+	s[4] = (v & 0x000000FF00000000) >> 32;
+	s[3] = (v & 0x00000000FF000000) >> 24;
+	s[2] = (v & 0x0000000000FF0000) >> 16;
+	s[1] = (v & 0x000000000000FF00) >>  8;
+	s[0] = (v & 0x00000000000000FF);
+}
+
+#endif /* #if __BYTE_ORDER == __LITTLE_ENDIAN */
+
 #else /* !(x86 || powerpc) && !(arm && !(old arm architectures)) */
 
+/* pragma pack is available in gcc (though originally apparently by
+ * Microsoft) and in some other compilers (probably inspired by either
+ * the two big ones), but there is no good portable way to detect
+ * whether it's supported.  The bad news: on platforms where it's not
+ * supported (unsupported pragmas are ignored) but which do require
+ * strict alignment, the below pragma pack trickery will fail.
+ * Therefore this option is the last and the default, and the platforms
+ * requiring strict alignment are detected earlier. */
+
 #pragma pack(1)
 struct una_u16 { uint16_t x; };
 struct una_u32 { uint32_t x; };
@@ -263,7 +468,7 @@ static INLINE void UNALIGNED_STORE64(void *p, uint64_t v)
 	ptr->x = v;
 }
 
-#endif /* !(x86 || powerpc) && !(arm && !armv5 && !armv6) */
+#endif /* defining UNALIGNED_LOADNN and UNALIGNED_STORENN */
 
 
 #if __BYTE_ORDER == __LITTLE_ENDIAN
@@ -13,4 +13,59 @@
 #define expect_false(expr) expect((expr) != 0, 0)
 #define expect_true(expr)  expect((expr) != 0, 1)
 
+/* these defines are somewhat borrowed from miniz.c */
+
+#if defined(_M_IX86) || defined(_M_X64) || defined(__i386__) || defined(__i386) || defined(__i486__) || defined(__i486) || defined(i386) || defined(__ia64__) || defined(__x86_64__) || defined(__x86_64)
+/* SRL_X86_OR_X64_CPU is only used to help set the below macros. */
+#define SRL_X86_OR_X64_CPU
+#ifndef SRL_EXTENDED_PRECISION_LONG_DOUBLE
+#define SRL_EXTENDED_PRECISION_LONG_DOUBLE 1
+#endif
+#endif
+
+#ifndef SRL_EXTENDED_PRECISION_LONG_DOUBLE
+#define SRL_EXTENDED_PRECISION_LONG_DOUBLE 0
+#endif
+
+#ifndef SRL_USE_ALIGNED_LOADS_AND_STORES
+
+#ifdef __hpux
+/* HP-UX runs on Itanium but has strict alignment so we check it first. */
+#define SRL_USE_ALIGNED_LOADS_AND_STORES 1
+#elif defined(SRL_X86_OR_X64_CPU)
+/* Set SRL_USE_ALIGNED_LOADS_AND_STORES to 0 on CPU's that permit efficient integer loads and stores from unaligned addresses. */
+#define SRL_USE_ALIGNED_LOADS_AND_STORES 0
+#else
+/* When in doubt use aligned loads and stores */
+#define SRL_USE_ALIGNED_LOADS_AND_STORES 1
+#endif
+
+#endif
+
+
+/* In x86 one can try to enforce strict alignment in runtime.
+ *
+ * Setting the CPU flag bit 18 (called "AC", aligment check) in
+ * the "EFLAGS" (user-settable) causes unaligned access traps but
+ * only iff the system register CR0 (only system-settable, usually done
+ * (or not) during kernel boot) has the same bit set (there called "AM",
+ * alignment mask).  If both flags are not set, the strict alignment
+ * traps (silently) do not happen.
+ *
+ * The Linux kernel and the Solarix x86 set the "AM".  The Windows and
+ * OX X do not.  The *BSD behavior is unknown, though suspecting they do.
+ *
+ * http://en.wikipedia.org/wiki/Control_register
+ * http://en.wikipedia.org/wiki/FLAGS_register_(computing)
+ */
+#ifdef SRL_X86_OR_X64_CPU
+#  if __x86_64__ || __x86_64
+#    define SRL_TRY_ENABLE_STRICT_ALIGN() asm("pushf\norl $0x40000, (%rsp)\npopf")
+#  elif __i386__ || __i386
+#    define SRL_TRY_ENABLE_STRICT_ALIGN() asm("pushf\norl $0x40000, (%esp)\npopf")
+#  endif
+#else
+#  define SRL_TRY_ENABLE_STRICT_ALIGN() (void)0
+#endif
+
 #endif
@@ -40,7 +40,9 @@ extern "C" {
 #define MY_CAN_FIND_PLACEHOLDERS
 #define HAS_SV2OBJ
 #endif
-
+#if (PERL_VERSION < 10)
+#   define FIXUP_RITER 1
+#endif
 #define DEFAULT_MAX_RECUR_DEPTH 10000
 
 #include "srl_decoder.h"
@@ -50,6 +52,7 @@ extern "C" {
 #include "srl_protocol.h"
 
 #include "snappy/csnappy_decompress.c"
+#include "miniz.h"
 
 /* 5.8.8 and earlier have a nasty bug in their handling of overloading:
  * The overload-flag is set on the referer of the blessed object instead of
@@ -92,15 +95,21 @@ void srl_destroy_decoder(pTHX_ srl_decoder_t *dec);                 /* destructo
 void srl_decoder_destructor_hook(pTHX_ void *p);                    /* destructor hook - called automagically */
 
 /* the top level components of the decode process - called by srl_decode_into() */
-SRL_STATIC_INLINE void srl_begin_decoding(pTHX_ srl_decoder_t *dec, SV *src, UV start_offset);       /* set up the decoder to handle a given var */
+/* srl_begin_decoding: set up the decoder to handle a given var */
+SRL_STATIC_INLINE srl_decoder_t *srl_begin_decoding(pTHX_ srl_decoder_t *dec, SV *src, UV start_offset);
 SRL_STATIC_INLINE void srl_read_header(pTHX_ srl_decoder_t *dec, SV *header_user_data); /* read/validate header */
 SRL_STATIC_INLINE void srl_read_single_value(pTHX_ srl_decoder_t *dec, SV* into);   /* main recursive dump routine */
+SRL_STATIC_INLINE void srl_read_single_value_into_container(pTHX_ srl_decoder_t *dec,
+        SV** container);   /* wrapper for main recursive dump routine for handling aliasing  */
 SRL_STATIC_INLINE void srl_finalize_structure(pTHX_ srl_decoder_t *dec);             /* optional finalize structure logic */
 SRL_STATIC_INLINE void srl_clear_decoder(pTHX_ srl_decoder_t *dec);                 /* clean up decoder after a dump */
 SRL_STATIC_INLINE void srl_clear_decoder_body_state(pTHX_ srl_decoder_t *dec);      /* clean up after each document body */
+SRL_STATIC_INLINE void srl_realloc_empty_buffer(pTHX_ srl_decoder_t *dec, const STRLEN header_len, const STRLEN body_len);
+SRL_STATIC_INLINE void srl_decompress_body_snappy(pTHX_ srl_decoder_t *dec);
+SRL_STATIC_INLINE void srl_decompress_body_zlib(pTHX_ srl_decoder_t *dec);
+
 
 /* the internal routines to handle each kind of object we have to deserialize */
-SRL_STATIC_INLINE SV *srl_read_alias(pTHX_ srl_decoder_t *dec);
 SRL_STATIC_INLINE void srl_read_copy(pTHX_ srl_decoder_t *dec, SV* into);
 
 SRL_STATIC_INLINE void srl_read_hash(pTHX_ srl_decoder_t *dec, SV* into, U8 tag);
@@ -127,10 +136,12 @@ SRL_STATIC_INLINE void srl_read_frozen_object(pTHX_ srl_decoder_t *dec, HV *clas
 SRL_STATIC_INLINE SV *srl_read_extend(pTHX_ srl_decoder_t *dec, SV* into);
 
 
-#define ASSERT_BUF_SPACE(dec,len,msg) STMT_START {              \
-    if (expect_false( (UV)BUF_SPACE((dec)) < (UV)(len) )) { \
-        SRL_ERRORf3("Unexpected termination of packet%s, want %lu bytes, only have %lu available", (msg), (UV)(len), (UV)BUF_SPACE((dec)));  \
-    }                                                       \
+#define ASSERT_BUF_SPACE(dec,len,msg) STMT_START {                  \
+    if (expect_false( (UV)BUF_SPACE((dec)) < (UV)(len) )) {         \
+        SRL_ERRORf3("Unexpected termination of packet%s, "          \
+                    "want %lu bytes, only have %lu available",      \
+                    (msg), (UV)(len), (UV)BUF_SPACE((dec)));        \
+    }                                                               \
 } STMT_END
 
 #define IS_SRL_HDR_ARRAYREF(tag) (((tag) & SRL_HDR_ARRAYREF) == SRL_HDR_ARRAYREF)
@@ -138,20 +149,6 @@ SRL_STATIC_INLINE SV *srl_read_extend(pTHX_ srl_decoder_t *dec, SV* into);
 #define IS_SRL_HDR_SHORT_BINARY(tag) (((tag) & SRL_HDR_SHORT_BINARY_LOW) == SRL_HDR_SHORT_BINARY_LOW)
 #define SRL_HDR_SHORT_BINARY_LEN_FROM_TAG(tag) ((tag) & SRL_MASK_SHORT_BINARY_LEN)
 
-/* Macro to assert that the type of an SV is complex enough to
- * be an RV. Differs on old perls since there used to be an RV type.
- */
-#if PERL_VERSION < 12
-#   define SRL_ASSERT_TYPE_FOR_RV(sv) STMT_START {  \
-            if (SvTYPE(sv) < SVt_PV)                \
-                sv_upgrade(into, SVt_RV);           \
-        } STMT_END
-#else
-#   define SRL_ASSERT_TYPE_FOR_RV(sv) STMT_START {  \
-            if (SvTYPE(sv) < SVt_PV)                \
-                sv_upgrade(into, SVt_IV);           \
-        } STMT_END
-#endif
 
 #define SRL_ASSERT_REF_PTR_TABLES(dec) STMT_START {     \
             if (expect_false( !(dec)->ref_stashes )) {  \
@@ -198,6 +195,9 @@ srl_build_decoder_struct(pTHX_ HV *opt)
         if ( (svp = hv_fetchs(opt, "refuse_snappy", 0)) && SvTRUE(*svp))
             SRL_DEC_SET_OPTION(dec, SRL_F_DECODER_REFUSE_SNAPPY);
 
+        if ( (svp = hv_fetchs(opt, "refuse_zlib", 0)) && SvTRUE(*svp))
+            SRL_DEC_SET_OPTION(dec, SRL_F_DECODER_REFUSE_ZLIB);
+
         if ( (svp = hv_fetchs(opt, "refuse_objects", 0)) && SvTRUE(*svp))
             SRL_DEC_SET_OPTION(dec, SRL_F_DECODER_REFUSE_OBJECTS);
 
@@ -215,11 +215,86 @@ srl_build_decoder_struct(pTHX_ HV *opt)
 
         if ( (svp = hv_fetchs(opt, "incremental", 0)) && SvTRUE(*svp))
             SRL_DEC_SET_OPTION(dec,SRL_F_DECODER_DESTRUCTIVE_INCREMENTAL);
+
+        /* see if they want us to alias varints, value is an unsigned integer.
+         * setting it to a true value smaller than 16 is the same as
+         * using the "alias_smallint" option. Setting it to a true value larger
+         * than 15 enables aliasing of smallints, and implies "alias_smallint" as
+         * well. */
+        if ( (svp = hv_fetchs(opt, "alias_varint_under", 0)) && SvTRUE(*svp)) {
+            /* if they use this then they automatically imply doing it for
+             * smallint as well */
+            SRL_DEC_SET_OPTION(dec,SRL_F_DECODER_ALIAS_SMALLINT);
+            SRL_DEC_SET_OPTION(dec,SRL_F_DECODER_ALIAS_VARINT);
+            if (SvUV(*svp) < 16) {
+                /* too small, just enable for SMALLINT (POS/NEG)*/
+                dec->alias_varint_under= 16;
+            } else {
+                /* larger than POS/NEG range, also alias some VARINTs */
+                /* anything smaller than this number will be aliased */
+                dec->alias_varint_under= SvUV(*svp);
+            }
+            /* create the alias cache */
+            dec->alias_cache= newAV();
+            /* extend it to the right size 16 for NEG,
+             * dec->alias_varint_under is at least 15, and 1 more for zero,
+             * so we allocate enough for POS/NEG as well as for the additional varints*/
+            av_extend(dec->alias_cache, 16 + dec->alias_varint_under);
+            AvFILLp(dec->alias_cache)= 16 + dec->alias_varint_under - 1; /* remove 1 as this is $#ary */
+        }
+
+        /* they can enable aliasing of SMALLINT's alone */
+        if ( !SRL_DEC_HAVE_OPTION(dec,SRL_F_DECODER_ALIAS_SMALLINT) &&
+             (svp = hv_fetchs(opt, "alias_smallint", 0)) && SvTRUE(*svp)
+        ) {
+            /* set the flag */
+            SRL_DEC_SET_OPTION(dec,SRL_F_DECODER_ALIAS_SMALLINT);
+            /* create the alias cache */
+            dec->alias_cache= newAV();
+            /* extend it to the right size of 32 items */
+            av_extend(dec->alias_cache,32);
+            AvFILLp(dec->alias_cache)= 31; /* $#ary == 32 */
+        }
+        /* check if they want us to use &PL_sv_undef for SRL_HEADER_UNDEF
+         * even if this might break referential integrity. */
+        if ( (svp = hv_fetchs(opt, "use_undef", 0)) && SvTRUE(*svp))
+            SRL_DEC_SET_OPTION(dec,SRL_F_DECODER_USE_UNDEF);
+
+        /* check if they want us to set all SVs readonly. */
+        if ( (svp = hv_fetchs(opt, "set_readonly", 0)) && SvTRUE(*svp))
+            SRL_DEC_SET_OPTION(dec, SRL_F_DECODER_SET_READONLY);
+
+        /* check if they want us to set normal scalars readonly. */
+        if ( (svp = hv_fetchs(opt, "set_readonly_scalars", 0)) && SvTRUE(*svp))
+            SRL_DEC_SET_OPTION(dec, SRL_F_DECODER_SET_READONLY_SCALARS);
+
     }
 
     return dec;
 }
 
+/* Clone a decoder whilst resetting ephemeral state on the clone. */
+SRL_STATIC_INLINE srl_decoder_t *
+srl_build_decoder_struct_alike(pTHX_ srl_decoder_t *proto)
+{
+    srl_decoder_t *dec;
+
+    Newxz(dec, 1, srl_decoder_t);
+
+    dec->ref_seenhash = PTABLE_new();
+    dec->max_recursion_depth = proto->max_recursion_depth;
+    dec->max_num_hash_entries = proto->max_num_hash_entries;
+
+    if (dec->alias_cache) {
+        dec->alias_cache = proto->alias_cache;
+        SvREFCNT_inc(dec->alias_cache);
+    }
+    dec->flags = proto->flags;
+    SRL_DEC_RESET_VOLATILE_FLAGS(dec);
+
+    return dec;
+}
+
 /* Explicit destructor */
 void
 srl_destroy_decoder(pTHX_ srl_decoder_t *dec)
@@ -235,6 +310,8 @@ srl_destroy_decoder(pTHX_ srl_decoder_t *dec)
     }
     if (dec->ref_thawhash)
         PTABLE_free(dec->ref_thawhash);
+    if (dec->alias_cache)
+        SvREFCNT_dec(dec->alias_cache);
     Safefree(dec);
 }
 
@@ -248,9 +325,6 @@ srl_decoder_destructor_hook(pTHX_ void *p)
 {
     srl_decoder_t *dec = (srl_decoder_t *)p;
 
-    assert(SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_DESTRUCTOR_OK));
-    SRL_DEC_UNSET_OPTION(dec, SRL_F_DECODER_DESTRUCTOR_OK);
-
     /* Only free decoder if not for reuse */
     if (!SRL_DEC_HAVE_OPTION(dec, SRL_F_REUSE_DECODER)) {
         srl_destroy_decoder(aTHX_ dec);
@@ -261,66 +335,128 @@ srl_decoder_destructor_hook(pTHX_ void *p)
     }
 }
 
+/* Creates a new buffer of size header_len+body_len+1 and swaps it
+ * into place of the current decoder's buffer. Sets decoder
+ * position to right after the header and makes the decoder state
+ * internally consistent.
+ * The buffer is owned by a mortal SV. */
+SRL_STATIC_INLINE void
+srl_realloc_empty_buffer(pTHX_ srl_decoder_t *dec,
+                         const STRLEN header_len,
+                         const STRLEN body_len)
+{
+    SV *buf_sv;
+    unsigned char *buf;
+
+    /* Let perl clean this up. Yes, it's not the most efficient thing
+     * ever, but it's just one mortal per full decompression, so not
+     * a bottle-neck. */
+    buf_sv = sv_2mortal( newSV(header_len + body_len + 1 ));
+    buf = (unsigned char *)SvPVX(buf_sv);
+
+    dec->buf_start = buf;
+    dec->pos = buf + header_len;
+    SRL_UPDATE_BODY_POS(dec);
+    dec->buf_end = dec->pos + body_len;
+    dec->buf_len = body_len + header_len;
+}
+
+/* Decompress a Snappy-compressed document body and put the resulting
+ * document body back in the place of the old compressed blob. */
+SRL_STATIC_INLINE void
+srl_decompress_body_snappy(pTHX_ srl_decoder_t *dec)
+{
+    uint32_t dest_len;
+    unsigned char *old_pos;
+    const ptrdiff_t sereal_header_len = dec->pos - dec->buf_start;
+    const STRLEN compressed_packet_len =
+        dec->encoding_flags == SRL_PROTOCOL_ENCODING_SNAPPY_INCREMENTAL
+        ? (STRLEN)srl_read_varint_uv_length(aTHX_ dec, " while reading compressed packet size")
+        : (STRLEN)(dec->buf_end - dec->pos);
+    int decompress_ok;
+    int header_len;
+
+    /* All decl's above here, or we break C89 compilers */
+
+    dec->bytes_consumed= compressed_packet_len + (dec->pos - dec->buf_start);
+
+    header_len = csnappy_get_uncompressed_length(
+            (char *)dec->pos,
+            compressed_packet_len,
+            &dest_len
+            );
+    if (header_len == CSNAPPY_E_HEADER_BAD)
+        SRL_ERROR("Invalid Snappy header in Snappy-compressed Sereal packet");
+
+    old_pos = dec->pos;
+
+    /* Allocate output buffer and swap it into place within the decoder. */
+    srl_realloc_empty_buffer(aTHX_ dec, sereal_header_len, dest_len);
+
+    decompress_ok = csnappy_decompress_noheader((char *)(old_pos + header_len),
+            compressed_packet_len - header_len,
+            (char *)dec->pos,
+            &dest_len);
+    if (expect_false( decompress_ok != 0 ))
+    {
+        SRL_ERRORf1("Snappy decompression of Sereal packet payload failed with error %i!", decompress_ok);
+    }
+}
+
+/* Decompress a zlib-compressed document body and put the resulting
+ * document body back in the place of the old compressed blob. */
+SRL_STATIC_INLINE void
+srl_decompress_body_zlib(pTHX_ srl_decoder_t *dec)
+{
+    unsigned char *old_pos;
+    const ptrdiff_t sereal_header_len = dec->pos - dec->buf_start;
+    const STRLEN uncompressed_packet_len = (STRLEN)srl_read_varint_uv(aTHX_ dec);
+    const STRLEN compressed_packet_len =
+        (STRLEN)srl_read_varint_uv_length(aTHX_ dec, " while reading compressed packet size");
+    int decompress_ok;
+    mz_ulong tmp;
+
+    /* All decl's above here, or we break C89 compilers */
+
+    dec->bytes_consumed= compressed_packet_len + (dec->pos - dec->buf_start);
+
+    old_pos = dec->pos;
+
+
+    /* Allocate output buffer and swap it into place within the decoder. */
+    srl_realloc_empty_buffer(aTHX_ dec, sereal_header_len, uncompressed_packet_len);
+    tmp = uncompressed_packet_len;
+    decompress_ok = mz_uncompress(
+        (unsigned char *)dec->pos,
+        &tmp,
+        (const unsigned char *)old_pos,
+        compressed_packet_len
+    );
+
+    if (expect_false( decompress_ok != Z_OK ))
+    {
+        SRL_ERRORf1("ZLIB decompression of Sereal packet payload failed with error %i!", decompress_ok);
+    }
+}
+
+
+
 /* Logic shared by the various decoder entry points. */
 SRL_STATIC_INLINE void
-srl_decode_into_internal(pTHX_ srl_decoder_t *dec, SV *src, SV *header_into, SV *body_into, UV start_offset)
+srl_decode_into_internal(pTHX_ srl_decoder_t *origdec, SV *src, SV *header_into, SV *body_into, UV start_offset)
 {
-    assert(dec != NULL);
-    if (SvUTF8(src))
-        sv_utf8_downgrade(src, 0);
-    srl_begin_decoding(aTHX_ dec, src, start_offset);
+    srl_decoder_t *dec;
+
+    assert(origdec != NULL);
+    dec = srl_begin_decoding(aTHX_ origdec, src, start_offset);
     srl_read_header(aTHX_ dec, header_into);
     SRL_UPDATE_BODY_POS(dec);
-    if (SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_DECOMPRESS_SNAPPY)) {
-        /* uncompress */
-        uint32_t dest_len;
-        SV *buf_sv;
-        unsigned char *buf;
-        unsigned char *old_pos;
-        const ptrdiff_t sereal_header_len = dec->pos - dec->buf_start;
-        const STRLEN compressed_packet_len =
-                ( dec->proto_version_and_flags & SRL_PROTOCOL_ENCODING_MASK ) == SRL_PROTOCOL_ENCODING_SNAPPY_INCREMENTAL
-                ? (STRLEN)srl_read_varint_uv_length(aTHX_ dec, " while reading compressed packet size")
-                : (STRLEN)(dec->buf_end - dec->pos);
-        int decompress_ok;
-        int header_len;
-
-        /* all decl's above here, or we break C89 compilers */
-
-        dec->bytes_consumed= compressed_packet_len + (dec->pos - dec->buf_start);
-
-        header_len = csnappy_get_uncompressed_length(
-                            (char *)dec->pos,
-                            compressed_packet_len,
-                            &dest_len
-                         );
-        if (header_len == CSNAPPY_E_HEADER_BAD)
-            SRL_ERROR("Invalid Snappy header in Snappy-compressed Sereal packet");
-
-        /* Let perl clean this up. Yes, it's not the most efficient thing
-         * ever, but it's just one mortal per full decompression, so not
-         * a bottle-neck. */
-        buf_sv = sv_2mortal( newSV(sereal_header_len + dest_len + 1 ));
-        buf = (unsigned char *)SvPVX(buf_sv);
-
-        /* not necessary to copy the Sereal header! */
-        /* Copy(dec->buf_start, buf, sereal_header_len, unsigned char); */
-
-        old_pos = dec->pos;
-        dec->buf_start = buf;
-        dec->pos = buf + sereal_header_len;
-        SRL_UPDATE_BODY_POS(dec);
-        dec->buf_end = dec->pos + dest_len;
-        dec->buf_len = dest_len + sereal_header_len;
-
-        decompress_ok = csnappy_decompress_noheader((char *)(old_pos + header_len),
-                                                    compressed_packet_len - header_len,
-                                                    (char *)dec->pos,
-                                                    &dest_len);
-        if (expect_false( decompress_ok != 0 ))
-        {
-            SRL_ERRORf1("Snappy decompression of Sereal packet payload failed with error %i!", decompress_ok);
-        }
+    if (expect_false( SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_DECOMPRESS_SNAPPY) )) {
+        srl_decompress_body_snappy(aTHX_ dec);
+        origdec->bytes_consumed = dec->bytes_consumed;
+    } else if (expect_false( SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_DECOMPRESS_ZLIB) )) {
+        srl_decompress_body_zlib(aTHX_ dec);
+        origdec->bytes_consumed = dec->bytes_consumed;
     }
 
     /* The actual document body deserialization: */
@@ -331,8 +467,10 @@ srl_decode_into_internal(pTHX_ srl_decoder_t *dec, SV *src, SV *header_into, SV
 
     /* If we aren't reading from a decompressed buffer we have to remember the number
      * of bytes used for the user to query. */
-    if (dec->bytes_consumed == 0)
+    if (dec->bytes_consumed == 0) {
         dec->bytes_consumed = dec->pos - dec->buf_start;
+        origdec->bytes_consumed = dec->bytes_consumed;
+    }
 
     if (SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_DESTRUCTIVE_INCREMENTAL)) {
         STRLEN len;
@@ -346,12 +484,11 @@ srl_decode_into_internal(pTHX_ srl_decoder_t *dec, SV *src, SV *header_into, SV
 
 /* This is the main routine to deserialize just the header of a document. */
 SV *
-srl_decode_header_into(pTHX_ srl_decoder_t *dec, SV *src, SV* header_into, UV start_offset)
+srl_decode_header_into(pTHX_ srl_decoder_t *origdec, SV *src, SV* header_into, UV start_offset)
 {
-    assert(dec != NULL);
-    if (SvUTF8(src))
-        sv_utf8_downgrade(src, 0);
-    srl_begin_decoding(aTHX_ dec, src, start_offset);
+    srl_decoder_t *dec;
+    assert(origdec != NULL);
+    dec = srl_begin_decoding(aTHX_ origdec, src, start_offset);
     if (header_into == NULL)
         header_into = sv_newmortal();
     srl_read_header(aTHX_ dec, header_into);
@@ -410,20 +547,40 @@ srl_clear_decoder_body_state(pTHX_ srl_decoder_t *dec)
     dec->recursion_depth = 0;
 }
 
-SRL_STATIC_INLINE void
+SRL_STATIC_INLINE srl_decoder_t *
 srl_begin_decoding(pTHX_ srl_decoder_t *dec, SV *src, UV start_offset)
 {
     STRLEN len;
     unsigned char *tmp;
 
-    /* Assert that we did not push a destructor before */
-    assert(!SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_DESTRUCTOR_OK));
-    /* Push destructor, set destructor-is-pushed flag */
-    SRL_DEC_SET_OPTION(dec, SRL_F_DECODER_DESTRUCTOR_OK);
+    /* Check whether decoder is in use and create a new one on the
+     * fly if necessary. Should only happen in edge cases such as
+     * a THAW hook calling back into the same decoder. */
+    if (SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_DIRTY)) {
+        srl_decoder_t * const proto = dec;
+        dec = srl_build_decoder_struct_alike(aTHX_ proto);
+        SRL_DEC_UNSET_OPTION(dec, SRL_F_REUSE_DECODER);
+    }
+
+    /* Needs to be before setting DIRTY because DIRTY is volatile. */
+    SRL_DEC_RESET_VOLATILE_FLAGS(dec);
+
+    /* Set to being in use. */;
+    SRL_DEC_SET_OPTION(dec, SRL_F_DECODER_DIRTY);
+
     /* Register our structure for destruction on scope exit */
     SAVEDESTRUCTOR_X(&srl_decoder_destructor_hook, (void *)dec);
 
-    SRL_DEC_RESET_VOLATILE_FLAGS(dec);
+    if (SvUTF8(src)) {
+        /* If we are being asked to decode a utf8-on string then we
+         * make a mortal copy, and then try to downgrade the copy.
+         * The downgrade will croak if it cannot successfully downgrade
+         * the buffer. If it is sucessful then decode the downgraded
+         * copy. */
+        src= sv_mortalcopy(src);
+        sv_utf8_downgrade(src, 0);
+    }
+
     tmp = (unsigned char*)SvPV(src, len);
     if (expect_false( start_offset > len )) {
         SRL_ERROR("Start offset is beyond input string length");
@@ -433,37 +590,82 @@ srl_begin_decoding(pTHX_ srl_decoder_t *dec, SV *src, UV start_offset)
     dec->buf_len= len - start_offset;
     SRL_SET_BODY_POS(dec, dec->buf_start);
     dec->bytes_consumed = 0;
+
+    return dec;
 }
 
+IV
+srl_validate_header_version_pv_len(pTHX_ char *strdata, STRLEN len)
+{
+    if ( len >= SRL_MAGIC_STRLEN + 3 ) {
+        /* + 3 above because:
+         * at least one version/flag byte,
+         * one byte for header len,
+         * one type byte (smallest payload)
+         */
+
+        /* Do NOT do *((U32*)strdata at least for these reasons:
+         * (1) Unaligned access can "Bus error" on you
+         *     (char* can be much less aligned than U32).
+         * (2) In ILP64 even if aligned the U32 would be 64 bits wide,
+         *     and the deref would read 8 bytes, more than the smallest
+         *     (valid) message.
+         * (3) Endianness.
+         */
+        U8 version_encoding= strdata[SRL_MAGIC_STRLEN];
+        U8 version= version_encoding & SRL_PROTOCOL_VERSION_MASK;
+
+        if ( memEQ(SRL_MAGIC_STRING, strdata, SRL_MAGIC_STRLEN) ) {
+            if ( 0 < version && version < 3 ) {
+                return version_encoding;
+            }
+        }
+        else
+        if ( memEQ(SRL_MAGIC_STRING_HIGHBIT, strdata, SRL_MAGIC_STRLEN) ) {
+            if ( 3 <= version ) {
+                return version_encoding;
+           }
+        }
+        else
+        if ( memEQ(SRL_MAGIC_STRING_HIGHBIT_UTF8, strdata, SRL_MAGIC_STRLEN) ) {
+            return 0;
+        }
+    }
+    return -1;
+}
+
+
+
 SRL_STATIC_INLINE void
 srl_read_header(pTHX_ srl_decoder_t *dec, SV *header_user_data)
 {
     UV header_len;
+    IV proto_version_and_encoding_flags_int= srl_validate_header_version_pv_len(aTHX_ (char *)BUF_POS(dec), BUF_SPACE(dec));
 
-    /* 4 byte magic string + version/flags + hdr len at least */
-    ASSERT_BUF_SPACE(dec, 4 + 1 + 1," while reading header");
-    if (strnEQ((char*)dec->pos, SRL_MAGIC_STRING, 4)) {
-        unsigned int proto_version;
+    if ( proto_version_and_encoding_flags_int < 1 ) {
+        if (proto_version_and_encoding_flags_int == 0)
+            SRL_ERROR("Bad Sereal header: It seems your document was accidentally UTF-8 encoded");
+        else
+            SRL_ERROR("Bad Sereal header: Not a valid Sereal document.");
+    }
+    else {
+        dec->pos += 5;
 
-        dec->pos += 4;
-        dec->proto_version_and_flags = *dec->pos++;
+        dec->proto_version = (U8)(proto_version_and_encoding_flags_int & SRL_PROTOCOL_VERSION_MASK);
+        dec->encoding_flags = (U8)(proto_version_and_encoding_flags_int & SRL_PROTOCOL_ENCODING_MASK);
 
-        proto_version = dec->proto_version_and_flags & SRL_PROTOCOL_VERSION_MASK;
-        if (expect_false( proto_version == 1 ))
+        if (expect_false( dec->proto_version == 1 ))
             SRL_DEC_SET_OPTION(dec, SRL_F_DECODER_PROTOCOL_V1); /* compat mode */
-        else if (expect_false( proto_version != 2 ))
-            SRL_ERRORf1("Unsupported Sereal protocol version %u",
-                    dec->proto_version_and_flags & SRL_PROTOCOL_VERSION_MASK);
+        else if (expect_false( dec->proto_version > 3 || dec->proto_version < 1 ))
+            SRL_ERRORf1("Unsupported Sereal protocol version %u", dec->proto_version);
 
-        if ((dec->proto_version_and_flags & SRL_PROTOCOL_ENCODING_MASK) == SRL_PROTOCOL_ENCODING_RAW) {
+        if (dec->encoding_flags == SRL_PROTOCOL_ENCODING_RAW) {
             /* no op */
         }
         else
-        if (
-                ( dec->proto_version_and_flags & SRL_PROTOCOL_ENCODING_MASK ) == SRL_PROTOCOL_ENCODING_SNAPPY
-                ||
-                ( dec->proto_version_and_flags & SRL_PROTOCOL_ENCODING_MASK ) == SRL_PROTOCOL_ENCODING_SNAPPY_INCREMENTAL
-        ) {
+        if (   dec->encoding_flags == SRL_PROTOCOL_ENCODING_SNAPPY
+            || dec->encoding_flags == SRL_PROTOCOL_ENCODING_SNAPPY_INCREMENTAL)
+        {
             if (expect_false( SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_REFUSE_SNAPPY) )) {
                 SRL_ERROR("Sereal document is compressed with Snappy, "
                       "but this decoder is configured to refuse Snappy-compressed input.");
@@ -471,17 +673,25 @@ srl_read_header(pTHX_ srl_decoder_t *dec, SV *header_user_data)
             dec->flags |= SRL_F_DECODER_DECOMPRESS_SNAPPY;
         }
         else
+        if (dec->encoding_flags == SRL_PROTOCOL_ENCODING_ZLIB)
+        {
+            if (expect_false( SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_REFUSE_ZLIB) )) {
+                SRL_ERROR("Sereal document is compressed with ZLIB, "
+                      "but this decoder is configured to refuse ZLIB-compressed input.");
+            }
+            dec->flags |= SRL_F_DECODER_DECOMPRESS_ZLIB;
+        }
+        else
         {
             SRL_ERRORf1( "Sereal document encoded in an unknown format '%d'",
-                     (dec->proto_version_and_flags & SRL_PROTOCOL_ENCODING_MASK)
-                      >> SRL_PROTOCOL_VERSION_BITS);
+                         dec->encoding_flags >> SRL_PROTOCOL_VERSION_BITS);
         }
 
         /* Must do this via a temporary as it modifes dec->pos itself */
         header_len= srl_read_varint_uv_length(aTHX_ dec, " while reading header");
 
-        if (proto_version > 1 && header_len) {
-            /* We have a protocol V2 extensible header:
+        if (dec->proto_version > 1 && header_len) {
+            /* We have a protocol V2+ extensible header:
              *  - 8bit bitfield
              *  - if lowest bit set, we have custom-header-user-data after the bitfield
              *  => Only read header user data if an SV* was passed in to fill. */
@@ -505,8 +715,6 @@ srl_read_header(pTHX_ srl_decoder_t *dec, SV *header_user_data)
              * protocol version. */
             dec->pos += header_len;
         }
-    } else {
-        SRL_ERROR("Bad Sereal header: Does not start with Sereal magic");
     }
 }
 
@@ -514,6 +722,7 @@ SRL_STATIC_INLINE void
 srl_finalize_structure(pTHX_ srl_decoder_t *dec)
 {
     int nobless = SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_NO_BLESS_OBJECTS);
+
     if (dec->weakref_av)
         av_clear(dec->weakref_av);
     if (dec->ref_stashes) {
@@ -546,8 +755,17 @@ srl_finalize_structure(pTHX_ srl_decoder_t *dec)
 #if USE_588_WORKAROUND
                         /* was blessed early, don't rebless */
 #else
-                        if (!nobless)
-                            sv_bless(obj, stash);
+                        if (!nobless) {
+                            if ( SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_READONLY_FLAGS) && SvREADONLY(SvRV(obj))) {
+                                /* the referenced scalar was readonly, temporary 
+                                   set it rw to bless its reference */
+                                SvREADONLY_off(SvRV(obj));
+                                sv_bless(obj, stash);
+                                SvREADONLY_on(SvRV(obj));
+                            } else {
+                                sv_bless(obj, stash);
+                            }
+                        }
 #endif
                     } else {
                         PTABLE_iter_free(it);
@@ -586,18 +804,18 @@ srl_read_varint_uv_safe(pTHX_ srl_decoder_t *dec)
     return uv;
 }
 
-#define SET_UV_FROM_VARINT(uv, from) STMT_START {      \
-    if (*from < 0x80) {                                             \
-        uv= (UV)*from++;                                            \
-    } else {                                                        \
-        unsigned int lshift= 7;                                     \
-        uv= (UV)(*from++ & 0x7f);                                   \
-        while (*from & 0x80){                                       \
-            uv |= ((UV)(*from++ & 0x7F) << lshift);                 \
-            lshift += 7;                                            \
-        }                                                           \
-        uv |= ((UV)(*from++) << lshift);                            \
-    }                                                               \
+#define SET_UV_FROM_VARINT(uv, from) STMT_START {       \
+    if (*from < 0x80) {                                 \
+        uv= (UV)*from++;                                \
+    } else {                                            \
+        unsigned int lshift= 7;                         \
+        uv= (UV)(*from++ & 0x7f);                       \
+        while (*from & 0x80){                           \
+            uv |= ((UV)(*from++ & 0x7F) << lshift);     \
+            lshift += 7;                                \
+        }                                               \
+        uv |= ((UV)(*from++) << lshift);                \
+    }                                                   \
 } STMT_END
 
 SRL_STATIC_INLINE UV
@@ -719,20 +937,9 @@ srl_read_varint(pTHX_ srl_decoder_t *dec, SV* into)
 SRL_STATIC_INLINE void
 srl_read_zigzag(pTHX_ srl_decoder_t *dec, SV* into)
 {
-    UV uv= srl_read_varint_uv(aTHX_ dec);
-    if (uv & 1) {
-        sv_setiv(into, (IV)( -( 1 + (uv >> 1) ) ) );
-    } else {
-        uv = uv >> 1;
-        if (uv <= (UV)IV_MAX) {
-            sv_setiv(into, (IV)uv);
-        } else {
-            /* grr, this is ridiculous! */
-            sv_setiv(into, 0);
-            SvIsUV_on(into);
-            SvUV_set(into, uv);
-        }
-    }
+    UV n= srl_read_varint_uv(aTHX_ dec);
+    IV i= (n >> 1) ^ (-(n & 1));
+    sv_setiv(into, i);
 }
 
 
@@ -755,12 +962,30 @@ srl_read_string(pTHX_ srl_decoder_t *dec, int is_utf8, SV* into)
     dec->pos+= len;
 }
 
+/* declare a union so that we are guaranteed the right alignment
+ * rules - this is required for e.g. ARM */
+union myfloat {
+    U8 c[sizeof(long double)];
+    float f;
+    double d;
+    long double ld;
+};
+
+/* XXX Most (if not all?) non-x86 platforms are strict in their
+ * floating point alignment.  So maybe this logic should be the other
+ * way: default to strict, and do sloppy only if x86? */
 
 SRL_STATIC_INLINE void
 srl_read_float(pTHX_ srl_decoder_t *dec, SV* into)
 {
+    union myfloat val;
     ASSERT_BUF_SPACE(dec, sizeof(float), " while reading FLOAT");
-    sv_setnv(into, (NV)*((float *)dec->pos));
+#if SRL_USE_ALIGNED_LOADS_AND_STORES
+    Copy(dec->pos,val.c,sizeof(float),U8);
+#else
+    val.f= *((float *)dec->pos);
+#endif
+    sv_setnv(into, (NV)val.f);
     dec->pos+= sizeof(float);
 }
 
@@ -768,8 +993,14 @@ srl_read_float(pTHX_ srl_decoder_t *dec, SV* into)
 SRL_STATIC_INLINE void
 srl_read_double(pTHX_ srl_decoder_t *dec, SV* into)
 {
-    ASSERT_BUF_SPACE(dec, sizeof(double)," while reading DOUBLE");
-    sv_setnv(into, (NV)*((double *)dec->pos));
+    union myfloat val;
+    ASSERT_BUF_SPACE(dec, sizeof(double), " while reading DOUBLE");
+#if SRL_USE_ALIGNED_LOADS_AND_STORES
+    Copy(dec->pos,val.c,sizeof(double),U8);
+#else
+    val.d= *((double *)dec->pos);
+#endif
+    sv_setnv(into, (NV)val.d);
     dec->pos+= sizeof(double);
 }
 
@@ -777,8 +1008,14 @@ srl_read_double(pTHX_ srl_decoder_t *dec, SV* into)
 SRL_STATIC_INLINE void
 srl_read_long_double(pTHX_ srl_decoder_t *dec, SV* into)
 {
-    ASSERT_BUF_SPACE(dec, sizeof(long double)," while reading LONG_DOUBLE");
-    sv_setnv(into, (NV)*((long double *)dec->pos));
+    union myfloat val;
+    ASSERT_BUF_SPACE(dec, sizeof(long double), " while reading LONG_DOUBLE");
+#if SRL_USE_ALIGNED_LOADS_AND_STORES
+    Copy(dec->pos,val.c,sizeof(long double),U8);
+#else
+    val.ld= *((long double *)dec->pos);
+#endif
+    sv_setnv(into, (NV)val.ld);
     dec->pos+= sizeof(long double);
 }
 
@@ -789,10 +1026,7 @@ srl_read_array(pTHX_ srl_decoder_t *dec, SV *into, U8 tag) {
     if (tag) {
         SV *referent= (SV *)newAV();
         len= tag & 15;
-        SRL_ASSERT_TYPE_FOR_RV(into);
-        SvTEMP_off(referent);
-        SvRV_set(into, referent);
-        SvROK_on(into);
+        SRL_sv_set_rv_to(into, referent);
         into= referent;
     } else {
         len= srl_read_varint_uv_count(aTHX_ dec," while reading ARRAY");
@@ -805,19 +1039,16 @@ srl_read_array(pTHX_ srl_decoder_t *dec, SV *into, U8 tag) {
 
         ASSERT_BUF_SPACE(dec,len,"while reading array contents, insuffienct remaining tags for specified array size");
 
-        /* we cheat and store undef in the array - we will overwrite it later */
-        av_store((AV*)into, len-1, &PL_sv_undef);
+        /* make sure the array has room */
+        av_extend((AV*)into, len-1);
+        /* set the size */
+        AvFILLp(into)= len - 1;
+
         av_array= AvARRAY((AV*)into);
         av_end= av_array + len;
 
-        for ( ; av_array != av_end ; av_array++) {
-            if ( expect_false( *dec->pos == SRL_HDR_ALIAS ) ) {
-                dec->pos++;
-                *av_array= srl_read_alias(aTHX_ dec);
-            } else {
-                *av_array= newSV_type(SVt_NULL);
-                srl_read_single_value(aTHX_ dec, *av_array);
-            }
+        for ( ; av_array < av_end ; av_array++) {
+            srl_read_single_value_into_container(aTHX_ dec, av_array);
         }
     }
 }
@@ -833,15 +1064,21 @@ srl_read_hash(pTHX_ srl_decoder_t *dec, SV* into, U8 tag) {
     if (tag) {
         SV *referent= (SV *)newHV();
         num_keys= tag & 15;
-        SRL_ASSERT_TYPE_FOR_RV(into);
-        SvTEMP_off(referent);
-        SvRV_set(into, referent);
-        SvROK_on(into);
+        SRL_sv_set_rv_to(into, referent);
         into= referent;
     } else {
         num_keys= srl_read_varint_uv_count(aTHX_ dec," while reading HASH");
         (void)SvUPGRADE(into, SVt_PVHV);
     }
+    /* in some versions of Perl HvRITER() is not properly set on an upgrade SV
+     * so we explicitly set it ourselves */
+#ifdef FIXUP_RITER
+#ifdef HvRITER_set
+    HvRITER_set(into,-1);
+#else
+    HvRITER(into)= -1;
+#endif
+#endif
 
     /* Limit the maximum number of hash keys that we accept to whetever was configured */
     if (expect_false( dec->max_num_hash_entries != 0 && num_keys > dec->max_num_hash_entries )) {
@@ -920,12 +1157,7 @@ srl_read_hash(pTHX_ srl_decoder_t *dec, SV* into, U8 tag) {
         if (expect_false( !fetched_sv )) {
             SRL_ERROR_PANIC(dec,"failed to hv_store");
         }
-        if (expect_false( *dec->pos == SRL_HDR_ALIAS )) {
-            dec->pos++;
-            *fetched_sv= srl_read_alias(aTHX_ dec);
-        } else {
-            srl_read_single_value(aTHX_ dec, *fetched_sv);
-        }
+        srl_read_single_value_into_container(aTHX_ dec, fetched_sv);
     }
 }
 
@@ -935,13 +1167,40 @@ srl_read_refn(pTHX_ srl_decoder_t *dec, SV* into)
 {
     SV *referent;
     ASSERT_BUF_SPACE(dec, 1, " while reading REFN referent");
-    referent= newSV(SVt_NULL);
-
-    SRL_ASSERT_TYPE_FOR_RV(into);
-    SvTEMP_off(referent);
-    SvRV_set(into, referent);
-    SvROK_on(into);
-    srl_read_single_value(aTHX_ dec, referent);
+    U8 tag= *(dec->pos); /* Look ahead for special vars. */
+    if (tag == SRL_HDR_TRUE) {
+        dec->pos++;
+        referent= &PL_sv_yes;
+    }
+    else if (tag == SRL_HDR_FALSE) {
+        dec->pos++;
+        referent= &PL_sv_no;
+    }
+    /*
+     * Note the below is guarded by an option as we have use SRL_HDR_UNDEF
+     * also to represent "any SV which is undef", and using to represent
+     * true PL_sv_undef will break things.
+     *
+     * We need a new, different tag for true perl undef.
+     *
+     */
+    else
+    if (
+        ( tag == SRL_HDR_CANONICAL_UNDEF )
+        ||
+        ( SRL_DEC_HAVE_OPTION(dec,SRL_F_DECODER_USE_UNDEF) && tag == SRL_HDR_UNDEF )
+    ) {
+        dec->pos++;
+        referent= &PL_sv_undef;
+    }
+    else {
+        referent= newSV(SVt_NULL);
+        SvTEMP_off(referent);
+        tag = 0;
+    }
+    SRL_sv_set_rv_to(into, referent);
+    if (!tag)
+        srl_read_single_value(aTHX_ dec, referent);
 }
 
 SRL_STATIC_INLINE void
@@ -958,10 +1217,7 @@ srl_read_refp(pTHX_ srl_decoder_t *dec, SV* into)
     referent= srl_fetch_item(aTHX_ dec, item, "REFP");
     (void)SvREFCNT_inc(referent);
 
-    SRL_ASSERT_TYPE_FOR_RV(into);
-    SvTEMP_off(referent);
-    SvRV_set(into, referent);
-    SvROK_on(into);
+    SRL_sv_set_rv_to(into, referent);
 
 #if USE_588_WORKAROUND
     /* See 'define USE_588_WORKAROUND' above for a discussion of what this does. */
@@ -1097,9 +1353,16 @@ srl_read_object(pTHX_ srl_decoder_t *dec, SV* into, U8 obj_tag)
     if (tag == SRL_HDR_COPY) {
         ofs= srl_read_varint_uv_offset(aTHX_ dec, " while reading COPY class name");
         storepos= ofs;
-        if (expect_true( dec->ref_seenhash != NULL )) {
-            class_stash= PTABLE_fetch(dec->ref_seenhash, (void *)ofs);
+        /* if this string was seen before as part of a classname then we expect
+         * a stash available below. However it might have been serialized as a key
+         * or something like that, which would mean we dont have an entry in ref_stashes
+         * anymore. So first we check if we have a stash. If we do, then we can avoid
+         * some work. */
+        if (expect_true( dec->ref_stashes != NULL )) {
+            class_stash= PTABLE_fetch(dec->ref_stashes, (void *)ofs);
         }
+        /* Check if we actually got a class_stash back. If we didn't then we need
+         * to deserialize the class name */
         if (!class_stash) {
             from= dec->body_pos + ofs;
             tag= *from++;
@@ -1121,28 +1384,33 @@ srl_read_object(pTHX_ srl_decoder_t *dec, SV* into, U8 obj_tag)
                 SRL_ERROR_BAD_COPY(dec, SRL_HDR_OBJECT);
             }
         }
-        /* NOTREACHED */
     } else {
         SRL_ERROR_UNEXPECTED(dec,tag, "a class name");
     }
 
+    /* At this point we may or may not have a class stash. If they used a Copy there
+     * is a decent chance we do. */
     SRL_ASSERT_REF_PTR_TABLES(dec);
     if (!class_stash) {
+        /* no class stash - so we need to look it up and then store it away for future use */
         class_stash= gv_stashpvn((char *)from, key_len, flags);
         PTABLE_store(dec->ref_stashes, (void *)storepos, (void *)class_stash);
+        /* Since this is the first time we have seen this stash then it is the first time
+         * that we have stored an item in the ref_bless_av hash as well. So create a new one
+         * and store it away. */
         av= newAV();
         sv_2mortal((SV*)av);
         PTABLE_store(dec->ref_bless_av, (void *)storepos, (void *)av);
     } else {
-        if (NULL == (av= (AV *)PTABLE_fetch(dec->ref_bless_av, (void *)storepos)) )
+        /* we have a class stash so we should have a ref_bless_av as well. */
+        av= (AV *)PTABLE_fetch(dec->ref_bless_av, (void *)storepos);
+        if ( !av )
             SRL_ERRORf1("Panic, no ref_bless_av for %lu", (unsigned long)storepos);
     }
 
     if (expect_false( obj_tag == SRL_HDR_OBJECT_FREEZE )) {
         srl_read_frozen_object(aTHX_ dec, class_stash, into);
     }  else {
-
-
         /* We now have a stash so we /could/ bless... except that
          * we don't actually want to do so right now. We want to defer blessing
          * until the full packet has been read. Yes it is more overhead, but
@@ -1393,16 +1661,6 @@ srl_read_extend(pTHX_ srl_decoder_t *dec, SV* into)
     return into;
 }
 
-/* these are all special */
-
-SRL_STATIC_INLINE SV *
-srl_read_alias(pTHX_ srl_decoder_t *dec)
-{
-    UV item= srl_read_varint_uv_offset(aTHX_ dec," while reading ALIAS tag");
-    SV *referent= srl_fetch_item(aTHX_ dec, item, "ALIAS");
-    return SvREFCNT_inc(referent);
-}
-
 SRL_STATIC_INLINE void
 srl_read_copy(pTHX_ srl_decoder_t *dec, SV* into)
 {
@@ -1420,6 +1678,98 @@ srl_read_copy(pTHX_ srl_decoder_t *dec, SV* into)
     dec->save_pos= 0;
 }
 
+
+SRL_STATIC_INLINE void
+srl_read_single_value_into_container(pTHX_ srl_decoder_t *dec, SV** container)
+{
+    SV *alias;
+    U32 item;
+    IV iv;
+    U8 tag = *dec->pos;
+    U8 *tag_start= dec->pos;
+
+    /* it helps to think of this somewhat like a switch, except it does
+     * more complicated checks than a single integer expression lookup */
+
+    if (expect_false( tag == SRL_HDR_ALIAS )) {
+        dec->pos++;
+        item= srl_read_varint_uv_offset(aTHX_ dec," while reading ALIAS tag");
+        alias= srl_fetch_item(aTHX_ dec, item, "ALIAS");
+        /* jump forward to the shared aliasing logic */
+        goto do_refcnt_inc_alias;
+    }
+    else
+    if (
+        SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_ALIAS_CHECK_FLAGS)
+    ) {
+
+        if (
+            SRL_DEC_HAVE_OPTION(dec,SRL_F_DECODER_USE_UNDEF) &&
+            tag == SRL_HDR_UNDEF
+        ) {
+            dec->pos++;
+            alias= &PL_sv_undef;
+            /* jump forward to the shared aliasing logic */
+            goto do_alias;
+        }
+        else
+        if (
+            SRL_DEC_HAVE_OPTION(dec,SRL_F_DECODER_ALIAS_SMALLINT) &&
+            tag <= SRL_HDR_NEG_HIGH
+        ) {
+            dec->pos++;
+            if ( tag <= SRL_HDR_POS_HIGH ) {
+                iv= tag;
+            } else {
+                /* must be a SRL_HDR_NEG tag, subtract 32 to get real value */
+                iv= tag - 32;
+            }
+            /* jump forward to the shared iv caching logic */
+            goto do_aliased_iv;
+        }
+        else
+        if (
+            SRL_DEC_HAVE_OPTION(dec,SRL_F_DECODER_ALIAS_VARINT) &&
+            tag == SRL_HDR_VARINT
+        ) {
+            dec->pos++;
+            item= srl_read_varint_uv(aTHX_ dec);
+            if ( item < dec->alias_varint_under ) {
+                iv= (IV)item;
+
+              do_aliased_iv:
+                item = iv + 16; /* we always cover from -16 up so we add 16 */
+                if (!AvARRAY(dec->alias_cache)[item] || AvARRAY(dec->alias_cache)[item] == &PL_sv_undef) {
+                    alias= newSViv(iv);
+                    /* mark it as readonly so people dont try to modify it */
+                    SvREADONLY_on(alias);
+                    /* store it in the alias_cache array */
+                    AvARRAY(dec->alias_cache)[item]= alias;
+                } else {
+                    alias= AvARRAY(dec->alias_cache)[item];
+                }
+
+              do_refcnt_inc_alias:
+                SvREFCNT_inc(alias);
+
+              do_alias:
+                if (*container && *container != &PL_sv_undef)
+                    SvREFCNT_dec(*container);
+                *container= alias;
+                return;
+            }
+            else {
+                /* reset parse pointer and fallthrough */
+                dec->pos= tag_start;
+            }
+        }
+    }
+    if (!*container || *container == &PL_sv_undef)
+        *container = newSV_type(SVt_NULL);
+    srl_read_single_value(aTHX_ dec, *container);
+    return;
+}
+
 /****************************************************************************
  * MAIN DISPATCH SUB - ALL ROADS LEAD HERE                                  *
  ****************************************************************************/
@@ -1429,6 +1779,7 @@ srl_read_single_value(pTHX_ srl_decoder_t *dec, SV* into)
 {
     STRLEN len;
     U8 tag;
+    int is_ref = 0;
     if (++dec->recursion_depth > dec->max_recursion_depth) {
         SRL_ERRORf1("Reached recursion limit (%lu) during deserialization",
                 (unsigned long)dec->max_recursion_depth);
@@ -1462,38 +1813,41 @@ srl_read_single_value(pTHX_ srl_decoder_t *dec, SV* into)
     else
     if ( IS_SRL_HDR_HASHREF(tag) ) {
         srl_read_hash(aTHX_ dec, into, tag);
+        is_ref = 1;
     }
     else
     if ( IS_SRL_HDR_ARRAYREF(tag) ) {
         srl_read_array(aTHX_ dec, into, tag);
+        is_ref = 1;
     }
     else {
         switch (tag) {
-            case SRL_HDR_VARINT:        srl_read_varint(aTHX_ dec, into);           break;
-            case SRL_HDR_ZIGZAG:        srl_read_zigzag(aTHX_ dec, into);           break;
-
-            case SRL_HDR_FLOAT:         srl_read_float(aTHX_ dec, into);            break;
-            case SRL_HDR_DOUBLE:        srl_read_double(aTHX_ dec, into);           break;
-            case SRL_HDR_LONG_DOUBLE:   srl_read_long_double(aTHX_ dec, into);      break;
-
-            case SRL_HDR_TRUE:          sv_setsv(into, &PL_sv_yes);                 break;
-            case SRL_HDR_FALSE:         sv_setsv(into, &PL_sv_no);                  break;
-            case SRL_HDR_UNDEF:         sv_setsv(into, &PL_sv_undef);               break;
-            case SRL_HDR_BINARY:        srl_read_string(aTHX_ dec, 0, into);        break;
-            case SRL_HDR_STR_UTF8:      srl_read_string(aTHX_ dec, 1, into);        break;
-
-            case SRL_HDR_WEAKEN:        srl_read_weaken(aTHX_ dec, into);           break;
-            case SRL_HDR_REFN:          srl_read_refn(aTHX_ dec, into);             break;
-            case SRL_HDR_REFP:          srl_read_refp(aTHX_ dec, into);             break;
+            case SRL_HDR_VARINT:        srl_read_varint(aTHX_ dec, into);                 break;
+            case SRL_HDR_ZIGZAG:        srl_read_zigzag(aTHX_ dec, into);                 break;
+
+            case SRL_HDR_FLOAT:         srl_read_float(aTHX_ dec, into);                  break;
+            case SRL_HDR_DOUBLE:        srl_read_double(aTHX_ dec, into);                 break;
+            case SRL_HDR_LONG_DOUBLE:   srl_read_long_double(aTHX_ dec, into);            break;
+
+            case SRL_HDR_TRUE:          sv_setsv(into, &PL_sv_yes);                       break;
+            case SRL_HDR_FALSE:         sv_setsv(into, &PL_sv_no);                        break;
+            case SRL_HDR_CANONICAL_UNDEF: /* fallthrough */
+            case SRL_HDR_UNDEF:         sv_setsv(into, &PL_sv_undef);                     break;
+            case SRL_HDR_BINARY:        srl_read_string(aTHX_ dec, 0, into);              break;
+            case SRL_HDR_STR_UTF8:      srl_read_string(aTHX_ dec, 1, into);              break;
+
+            case SRL_HDR_WEAKEN:        srl_read_weaken(aTHX_ dec, into);       is_ref=1; break;
+            case SRL_HDR_REFN:          srl_read_refn(aTHX_ dec, into);         is_ref=1; break;
+            case SRL_HDR_REFP:          srl_read_refp(aTHX_ dec, into);         is_ref=1; break;
             case SRL_HDR_OBJECT_FREEZE:
-            case SRL_HDR_OBJECT:        srl_read_object(aTHX_ dec, into, tag);      break;
+            case SRL_HDR_OBJECT:        srl_read_object(aTHX_ dec, into, tag);  is_ref=1; break;
             case SRL_HDR_OBJECTV_FREEZE:
-            case SRL_HDR_OBJECTV:       srl_read_objectv(aTHX_ dec, into, tag);     break;
-            case SRL_HDR_COPY:          srl_read_copy(aTHX_ dec, into);             break;
-            case SRL_HDR_EXTEND:        srl_read_extend(aTHX_ dec, into);           break;
-            case SRL_HDR_HASH:          srl_read_hash(aTHX_ dec, into, 0);          break;
-            case SRL_HDR_ARRAY:         srl_read_array(aTHX_ dec, into, 0);         break;
-            case SRL_HDR_REGEXP:        srl_read_regexp(aTHX_ dec, into);           break;
+            case SRL_HDR_OBJECTV:       srl_read_objectv(aTHX_ dec, into, tag); is_ref=1; break;
+            case SRL_HDR_COPY:          srl_read_copy(aTHX_ dec, into);                   break;
+            case SRL_HDR_EXTEND:        srl_read_extend(aTHX_ dec, into);                 break;
+            case SRL_HDR_HASH:          srl_read_hash(aTHX_ dec, into, 0);                break;
+            case SRL_HDR_ARRAY:         srl_read_array(aTHX_ dec, into, 0);               break;
+            case SRL_HDR_REGEXP:        srl_read_regexp(aTHX_ dec, into);                 break;
 
             case SRL_HDR_PAD:           /* no op */
                 while (BUF_NOT_DONE(dec) && *dec->pos == SRL_HDR_PAD)
@@ -1506,5 +1860,11 @@ srl_read_single_value(pTHX_ srl_decoder_t *dec, SV* into)
         }
     }
 
+    /* they want us to set all SVs readonly, or only the non-ref */
+    if (  SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_SET_READONLY) ||
+          (SRL_DEC_HAVE_OPTION(dec, SRL_F_DECODER_SET_READONLY_SCALARS) && !is_ref) ) {
+        SvREADONLY_on(into);
+    }
+
     dec->recursion_depth--;
 }
@@ -14,7 +14,7 @@ typedef struct {
     unsigned char *body_pos;            /* in Sereal V2, all offsets are relative to the body */
     STRLEN buf_len;
 
-    U32 flags;                          /* flag-like options: See F_* defines in srl_decoder.c */
+    U32 flags;                          /* flag-like options: See SRL_F_DECODER_* defines in srl_decoder.c */
     UV max_recursion_depth;             /* Configurable limit on the number of recursive calls we're willing to make */
     UV max_num_hash_entries;            /* Configured maximum number of acceptable entries in a hash */
     ptable_ptr ref_seenhash;            /* ptr table for avoiding circular refs */
@@ -23,11 +23,18 @@ typedef struct {
     ptable_ptr ref_bless_av;            /* ptr table for tracking which objects need to be bless - key: ofs, value: mortal AV (of refs)  */
     AV* weakref_av;
 
+    AV* alias_cache; /* used to cache integers of different sizes. */
+    UV alias_varint_under;
+
     UV bytes_consumed;
     UV recursion_depth;                 /* Recursion depth of current decoder */
-    U8 proto_version_and_flags;
+    U8 proto_version;
+    U8 encoding_flags;
 } srl_decoder_t;
 
+/* utility routine */
+IV srl_validate_header_version_pv_len(pTHX_ char *strdata, STRLEN len);
+
 /* constructor; don't need destructor, this sets up a callback */
 srl_decoder_t *srl_build_decoder_struct(pTHX_ HV *opt);
 
@@ -65,6 +72,36 @@ void srl_decoder_destructor_hook(pTHX_ void *p);
         }                                                                           \
     } STMT_END
 
+/* Macro to assert that the type of an SV is complex enough to
+ * be an RV. Differs on old perls since there used to be an RV type.
+ */
+#if PERL_VERSION < 12
+#   define SVt_RV_FAKE SVt_RV
+#else
+#   define SVt_RV_FAKE SVt_IV
+#endif
+
+/* this is from sv.h in Perl core, which is for some reason guarded
+ * by an ifdef PERL_CORE, which I am loathe to enable. */
+
+#define SRL_prepare_SV_for_RV(sv)                                   \
+    STMT_START {                                                    \
+        if (SvTYPE(sv) < SVt_PV && SvTYPE(sv) != SVt_RV_FAKE)       \
+            sv_upgrade(sv, SVt_RV_FAKE);                            \
+        else if (SvTYPE(sv) >= SVt_PV) {                            \
+            SvPV_free(sv);                                          \
+            SvLEN_set(sv, 0);                                       \
+            SvCUR_set(sv, 0);                                       \
+        }                                                           \
+    } STMT_END
+
+#define SRL_sv_set_rv_to(into,referent)             \
+    STMT_START {                                    \
+        SRL_prepare_SV_for_RV(into);                \
+        SvTEMP_off(referent);                       \
+        SvRV_set(into, referent);                   \
+        SvROK_on(into);                             \
+    } STMT_END
 
 #define SRL_BASE_ERROR_FORMAT "Sereal: Error in %s line %u and char %i of input: "
 #define SRL_BASE_ERROR_ARGS __FILE__, __LINE__, (int)(1 + dec->pos - dec->buf_start)
@@ -75,9 +112,9 @@ void srl_decoder_destructor_hook(pTHX_ void *p);
 #define SRL_ERRORf3(fmt,var1,var2,var3)         croak(SRL_BASE_ERROR_FORMAT fmt, SRL_BASE_ERROR_ARGS, (var1),(var2),(var3))
 #define SRL_ERRORf4(fmt,var1,var2,var3,var4)    croak(SRL_BASE_ERROR_FORMAT fmt, SRL_BASE_ERROR_ARGS, (var1),(var2),(var3),(var4))
 #define SRL_ERROR_UNIMPLEMENTED(dec,tag,str) \
-    SRL_ERRORf3("Tag %u %s is unimplemented at ofs: %d", (tag), (str), BUF_POS_OFS(dec))
-#define SRL_ERROR_UNTERMINATED(dec,tag,str) \
-    SRL_ERRORf4("Tag SRL_HDR_%s %s was not terminated properly at ofs %lu with %lu to go", \
+    SRL_ERRORf3("Tag %u %s is unimplemented at ofs: %lu", (tag), (str), (unsigned long)BUF_POS_OFS(dec))
+#define SRL_ERROR_UNTERMINATED(dec,tag,str)                                                                 \
+    SRL_ERRORf4("Tag SRL_HDR_%s %s was not terminated properly at ofs %lu with %lu to go",                  \
             tag_name[(tag) & 127], (str), (dec)->pos - (dec)->buf_start, (dec)->buf_end - (dec)->pos)
 #define SRL_ERROR_BAD_COPY(dec, tag) \
     SRL_ERRORf1("While processing tag SRL_HDR_%s encountered a bad COPY tag", tag_name[(tag) & 127])
@@ -89,31 +126,49 @@ void srl_decoder_destructor_hook(pTHX_ void *p);
 
 /* If set, the decoder struct needs to be cleared instead of freed at
  * the end of a deserialization operation */
-#define SRL_F_REUSE_DECODER 1UL
-/* If set, then the decoder destructor was already pushed to the
- * callback stack */
-#define SRL_F_DECODER_DESTRUCTOR_OK 2UL
+#define SRL_F_REUSE_DECODER                     0x00000001UL
+/* If set, then the decoder is already in use and srl_decode_into will
+ * clone its own new decoder. */
+#define SRL_F_DECODER_DIRTY                     0x00000002UL
+/* Non-persistent flag! */
+#define SRL_F_DECODER_NEEDS_FINALIZE            0x00000004UL
 /* Non-persistent flag! */
-#define SRL_F_DECODER_NEEDS_FINALIZE 4UL
+#define SRL_F_DECODER_DECOMPRESS_SNAPPY         0x00000008UL
 /* Non-persistent flag! */
-#define SRL_F_DECODER_DECOMPRESS_SNAPPY 8UL
-/* Persistent flag: Make the decoder REFUSE compressed documents */
-#define SRL_F_DECODER_REFUSE_SNAPPY 16UL
+#define SRL_F_DECODER_DECOMPRESS_ZLIB           0x00000010UL
+/* Persistent flag: Make the decoder REFUSE Snappy-compressed documents */
+#define SRL_F_DECODER_REFUSE_SNAPPY             0x00000020UL
+/* Persistent flag: Make the decoder REFUSE zlib-compressed documents */
+#define SRL_F_DECODER_REFUSE_ZLIB               0x00000040UL
 /* Persistent flag: Make the decoder REFUSE objects */
-#define SRL_F_DECODER_REFUSE_OBJECTS 32UL
+#define SRL_F_DECODER_REFUSE_OBJECTS            0x00000080UL
 /* Persistent flag: Make the decoder validate UTT8 strings */
-#define SRL_F_DECODER_VALIDATE_UTF8 64UL
+#define SRL_F_DECODER_VALIDATE_UTF8             0x00000100UL
 /* Persistent flag: Make the encoder forget to bless */
-#define SRL_F_DECODER_NO_BLESS_OBJECTS 128UL
+#define SRL_F_DECODER_NO_BLESS_OBJECTS          0x00000200UL
 /* Persistent flag: Destructive incremental parsing */
-#define SRL_F_DECODER_DESTRUCTIVE_INCREMENTAL 256UL
+#define SRL_F_DECODER_DESTRUCTIVE_INCREMENTAL   0x00000400UL
 /* Non-persistent flag: The current packet is using protocol version 1 */
-#define SRL_F_DECODER_PROTOCOL_V1 512UL
+#define SRL_F_DECODER_PROTOCOL_V1               0x00000800UL
+/* Persistent flag: alias small integer values in Hashes and Arrays */
+#define SRL_F_DECODER_ALIAS_SMALLINT            0x00001000UL
+/* Persistent flag: use PL_sv_undef for undef values in Hashes and Arrays */
+#define SRL_F_DECODER_ALIAS_VARINT              0x00002000UL
+/* Persistent flag: use PL_sv_undef as many places as possible */
+#define SRL_F_DECODER_USE_UNDEF                 0x00004000UL
+/* Persistent flag: set all SV readonly */
+#define SRL_F_DECODER_SET_READONLY              0x00008000UL
+/* Persistent flag: set non-ref SV readonly */
+#define SRL_F_DECODER_SET_READONLY_SCALARS      0x00010000UL
+
+
+#define SRL_F_DECODER_ALIAS_CHECK_FLAGS   ( SRL_F_DECODER_ALIAS_SMALLINT | SRL_F_DECODER_ALIAS_VARINT | SRL_F_DECODER_USE_UNDEF )
+#define SRL_F_DECODER_READONLY_FLAGS   ( SRL_F_DECODER_SET_READONLY | SRL_F_DECODER_SET_READONLY_SCALARS )
 
 #define SRL_DEC_HAVE_OPTION(dec, flag_num) ((dec)->flags & flag_num)
 #define SRL_DEC_SET_OPTION(dec, flag_num) ((dec)->flags |= flag_num)
 #define SRL_DEC_UNSET_OPTION(dec, flag_num) ((dec)->flags &= ~flag_num)
-#define SRL_DEC_VOLATILE_FLAGS (SRL_F_DECODER_NEEDS_FINALIZE|SRL_F_DECODER_DECOMPRESS_SNAPPY|SRL_F_DECODER_PROTOCOL_V1)
+#define SRL_DEC_VOLATILE_FLAGS (SRL_F_DECODER_NEEDS_FINALIZE|SRL_F_DECODER_DECOMPRESS_SNAPPY|SRL_F_DECODER_PROTOCOL_V1|SRL_F_DECODER_DIRTY|SRL_F_DECODER_DECOMPRESS_ZLIB)
 #define SRL_DEC_RESET_VOLATILE_FLAGS(dec) ((dec)->flags &= ~SRL_DEC_VOLATILE_FLAGS)
 
 /* 
@@ -179,7 +234,7 @@ static const char * const tag_name[] = {
 	"RESERVED_2",        /* "6"   54 0x36 0b00110110 */
 	"RESERVED_3",        /* "7"   55 0x37 0b00110111 */
 	"RESERVED_4",        /* "8"   56 0x38 0b00111000 */
-	"RESERVED_5",        /* "9"   57 0x39 0b00111001 */
+	"CANONICAL_UNDEF",   /* "9"   57 0x39 0b00111001 */
 	"FALSE",             /* ":"   58 0x3a 0b00111010 */
 	"TRUE",              /* ";"   59 0x3b 0b00111011 */
 	"MANY",              /* "<"   60 0x3c 0b00111100 */
@@ -44,7 +44,7 @@
     FLOAT             | "\"" |  34 | 0x22 | 0b00100010 | <IEEE-FLOAT>
     DOUBLE            | "#"  |  35 | 0x23 | 0b00100011 | <IEEE-DOUBLE>
     LONG_DOUBLE       | "\$" |  36 | 0x24 | 0b00100100 | <IEEE-LONG-DOUBLE>
-    UNDEF             | "%"  |  37 | 0x25 | 0b00100101 | None - Perl undef
+    UNDEF             | "%"  |  37 | 0x25 | 0b00100101 | None - Perl undef var; eg my $var= undef;
     BINARY            | "&"  |  38 | 0x26 | 0b00100110 | <LEN-VARINT> <BYTES> - binary/(latin1) string
     STR_UTF8          | "'"  |  39 | 0x27 | 0b00100111 | <LEN-VARINT> <UTF8> - utf8 string
     REFN              | "("  |  40 | 0x28 | 0b00101000 | <ITEM-TAG>    - ref to next item
@@ -63,8 +63,8 @@
     RESERVED_1        | "5"  |  53 | 0x35 | 0b00110101 |
     RESERVED_2        | "6"  |  54 | 0x36 | 0b00110110 |
     RESERVED_3        | "7"  |  55 | 0x37 | 0b00110111 |
-    RESERVED_4        | "8"  |  56 | 0x38 | 0b00111000 |
-    RESERVED_5        | "9"  |  57 | 0x39 | 0b00111001 | reserved
+    RESERVED_4        | "8"  |  56 | 0x38 | 0b00111000 | reserved
+    CANONICAL_UNDEF   | "9"  |  57 | 0x39 | 0b00111001 | undef (PL_sv_undef) - "the" Perl undef (see notes)
     FALSE             | ":"  |  58 | 0x3a | 0b00111010 | false (PL_sv_no)
     TRUE              | ";"  |  59 | 0x3b | 0b00111011 | true  (PL_sv_yes)
     MANY              | "<"  |  60 | 0x3c | 0b00111100 | <LEN-VARINT> <TYPE-BYTE> <TAG-DATA> - repeated tag (not done yet, will be implemented in version 3)
@@ -140,11 +140,22 @@
 */
 
 /* magic string, protocol version and encoding information */
-#define SRL_MAGIC_STRING                "=srl"          /* Magic string for header. Every packet starts with this */
 #define SRL_MAGIC_STRLEN                4               /* Length of SRL_MAGIC_STRING */
-#define SRL_MAGIC_STRING_LILIPUTIAN     0x6c72733d      /* SRL_MAGIC_STRING as a little endian integer */
 
-#define SRL_PROTOCOL_VERSION            ( 2 )
+#define SRL_MAGIC_STRING                "=srl"          /* Magic string for header. Every packet starts with this or "=\xF3rl",
+                                                         * which is the high-bit-set-on-the-"s" equivalent. */
+#define SRL_MAGIC_STRING_UINT_LE                0x6C72733D  /* SRL_MAGIC_STRING as a little endian integer */
+#define SRL_MAGIC_STRING_UINT_BE                0x3D73726C  /* SRL_MAGIC_STRING as a big endian integer */
+
+#define SRL_MAGIC_STRING_HIGHBIT                "=\xF3rl"   /* Magic string for header, with high bit set for UTF8 sanity check. */
+#define SRL_MAGIC_STRING_HIGHBIT_UINT_LE        0x6C72F33D  /* SRL_MAGIC_STRING_HIGHBIT as a little endian integer */
+#define SRL_MAGIC_STRING_HIGHBIT_UINT_BE        0x3DF3726C  /* SRL_MAGIC_STRING_HIGHBIT as a big endian integer */
+
+#define SRL_MAGIC_STRING_HIGHBIT_UTF8           "=\xC3\xB3rl"   /* Magic string for header, corrupted by accidental UTF8 encoding */
+#define SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_LE   0x72B3C33D      /* first four bytes of SRL_MAGIC_STRING encoded as UTF8, little endian */
+#define SRL_MAGIC_STRING_HIGHBIT_UTF8_UINT_BE   0x3DC3B372      /* first four bytes of SRL_MAGIC_STRING encoded as UTF8, big endian */
+
+#define SRL_PROTOCOL_VERSION            ( 3 )
 #define SRL_PROTOCOL_VERSION_BITS       ( 4 )           /* how many bits we use for the version, the rest go to the encoding */
 #define SRL_PROTOCOL_VERSION_MASK       ( ( 1 << SRL_PROTOCOL_VERSION_BITS ) - 1 )
 
@@ -152,6 +163,7 @@
 #define SRL_PROTOCOL_ENCODING_RAW       ( 0 << SRL_PROTOCOL_VERSION_BITS )
 #define SRL_PROTOCOL_ENCODING_SNAPPY    ( 1 << SRL_PROTOCOL_VERSION_BITS )
 #define SRL_PROTOCOL_ENCODING_SNAPPY_INCREMENTAL    ( 2 << SRL_PROTOCOL_VERSION_BITS )
+#define SRL_PROTOCOL_ENCODING_ZLIB      ( 3 << SRL_PROTOCOL_VERSION_BITS )
 
 /* Bits in the header bitfield */
 #define SRL_PROTOCOL_HDR_USER_DATA      ( 1 )
@@ -179,7 +191,7 @@
 #define SRL_HDR_FLOAT           ((char)34)      /* <IEEE-FLOAT> */
 #define SRL_HDR_DOUBLE          ((char)35)      /* <IEEE-DOUBLE> */
 #define SRL_HDR_LONG_DOUBLE     ((char)36)      /* <IEEE-LONG-DOUBLE> */
-#define SRL_HDR_UNDEF           ((char)37)      /* None - Perl undef */
+#define SRL_HDR_UNDEF           ((char)37)      /* None - Perl undef var; eg my $var= undef; */
 #define SRL_HDR_BINARY          ((char)38)      /* <LEN-VARINT> <BYTES> - binary/(latin1) string */
 #define SRL_HDR_STR_UTF8        ((char)39)      /* <LEN-VARINT> <UTF8> - utf8 string */
 
@@ -202,8 +214,9 @@
  *       them, might have to explicit == check later. */
 #define SRL_HDR_RESERVED        ((char)52)      /* reserved */
 #define SRL_HDR_RESERVED_LOW    ((char)52)
-#define SRL_HDR_RESERVED_HIGH   ((char)57)
+#define SRL_HDR_RESERVED_HIGH   ((char)56)
 
+#define SRL_HDR_CANONICAL_UNDEF ((char)57)      /* undef (PL_sv_undef) - "the" Perl undef (see notes) */
 #define SRL_HDR_FALSE           ((char)58)      /* false (PL_sv_no)  */
 #define SRL_HDR_TRUE            ((char)59)      /* true  (PL_sv_yes) */
 
@@ -233,6 +246,6 @@
 
 /* TODO */
 
-#define SRL_SET_FBIT(where) ((where) |= SRL_HDR_TRACK_FLAG)
+#define SRL_SET_TRACK_FLAG(where) ((where) |= SRL_HDR_TRACK_FLAG)
 
 #endif
@@ -0,0 +1,36 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+# test our test framework
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+# needs more tests
+ok(_deep_cmp(["x"],{}));
+ok(_deep_cmp({"x"=>1},{"y"=>1}));
+ok(_deep_cmp({"x"=>1},{"x"=>2}));
+ok(_deep_cmp({"x"=>1},{"x"=>2,"y"=>1}));
+ok(!_deep_cmp({"x"=>1},{"x"=>1}));
+ok(!_deep_cmp(["x"],["x"]));
+ok(_deep_cmp(["x"],["y","p"]));
+ok(_deep_cmp(["a","x"],["y"]));
+ok(_cmp_str("foo","bar"));
+ok(!_cmp_str("aaa","aaa"));
+ok(_cmp_str("aaacowbbb","aaadogbb"));
+my $l= "ba\xDF";
+my $u= $l;
+utf8::upgrade($u);
+ok(_cmp_str($l,$u));
+pass();
+done_testing();
+
@@ -15,7 +15,7 @@ BEGIN {
 
 use Sereal::TestSet qw(:all);
 
-use Test::More tests => 8 + 6 + ( 31 * 2 );
+use Test::More tests => 8 + 6 + ( 31 * 3 );
 
 # Simple test to see whether we can get the number of bytes consumed
 # and whether offset works
@@ -51,13 +51,16 @@ SCOPE: {
 SKIP: {
     my $have_enc = have_encoder_and_decoder();
     if (not $have_enc) {
-        skip "Need encoder for chunk tests", 31*2;
+        skip "Need encoder for chunk tests", 31 * 3;
     }
     else {
         require Sereal::Encoder;
-        Sereal::Encoder->import("encode_sereal");
+        Sereal::Encoder->import("encode_sereal", "SRL_ZLIB");
 
-        for my $tuple ( ['raw' => [] ], [ snappy_incr => [ { snappy_incr => 1 } ] ] ) {
+        for my $tuple ( [ raw         => [] ],
+                        [ snappy_incr => [ { snappy_incr => 1 } ] ],
+                        [ zlib        => [ { compress => SRL_ZLIB() } ] ] )
+        {
             my ($name, $opts)= @$tuple;
             my $data;
             my $n = 30;
@@ -76,7 +79,7 @@ SKIP: {
             };
             my $err = $@ || 'Zombie error';
             ok($ok, "incremental decoder ($name) had no hissy fit")
-                or note("Error: $err");
+                or note("Error: $err. Data structures decoded up to that point:\n" . Data::Dumper::Dumper(\@out));
 
             is($out[$_-1], $_, "Decoding multiple packets from single string works ($name: $_)")
                 for 1..$n;
@@ -1,7 +1,7 @@
 #!perl
 use strict;
 use warnings;
-use Sereal::Decoder qw(decode_sereal looks_like_sereal);
+use Sereal::Decoder qw(decode_sereal looks_like_sereal scalar_looks_like_sereal);
 use Sereal::Decoder::Constants qw(:all);
 use Data::Dumper;
 use File::Spec;
@@ -9,25 +9,61 @@ use Devel::Peek;
 
 use Test::More;
 
+sub doc {
+    my ($high, $version, $good)= @_;
+
+    return(
+        ($high eq "utf8" ? SRL_MAGIC_STRING_HIGHBIT_UTF8 :
+         $high ? SRL_MAGIC_STRING_HIGHBIT : SRL_MAGIC_STRING) .
+        chr($version) .
+        chr(0) .
+        ($good ? chr(SRL_HDR_UNDEF) : "")
+    );
+}
+
+
 # Simple tests for looks_like_sereal.
 
 my @tests = (
     # input, bool outcome, name
-    ["", 0, "empty string is not Sereal"],
-    [undef, 0, "undef string is not Sereal"],
-    [SRL_MAGIC_STRING, 0, "SRL_MAGIC_STRING alone is not Sereal"],
-    [SRL_MAGIC_STRING . chr(1) . chr(0), 0, "SRL_MAGIC_STRING with header is not Sereal"],
-    [SRL_MAGIC_STRING . chr(1) . chr(0) . chr(SRL_HDR_UNDEF), 1, "SRL_MAGIC_STRING with header and small payload is Sereal"],
-    [SRL_MAGIC_STRING . chr(0) . chr(0) . chr(SRL_HDR_UNDEF), 0, "SRL_MAGIC_STRING with bad header is not Sereal"],
-    ["=Srl". chr(1) . chr(0) . chr(SRL_HDR_UNDEF), 0, "wrong magic string is not Sereal"],
+    [ "",                       "", "empty string is not Sereal"],
+    [ undef,                    "", "undef string is not Sereal"],
+    [ {},                       "", "{} is not Sereal"],
+    [ [],                       "", "[] is not Sereal"],
+
+    [ SRL_MAGIC_STRING,         "", "SRL_MAGIC_STRING alone is not Sereal"],
+    [ doc(0, 0, 1),             "", "SRL_MAGIC_STRING with bad protocol is not Sereal"],
+    [ doc(0, 1, 0),             "", "SRL_MAGIC_STRING protocol 1 with short body is not Sereal"],
+    [ doc(0, 1, 1),              1, "SRL_MAGIC_STRING protocol 1 with small payload is Sereal"],
+    [ doc(0, 2, 0),             "", "SRL_MAGIC_STRING protocol 2 with short body is not Sereal"],
+    [ doc(0, 2, 1),              2, "SRL_MAGIC_STRING protocol 2 with small payload is Sereal"],
+    [ doc(0, 3, 0),             "", "SRL_MAGIC_STRING protocol 3 with short body is not Sereal"],
+    [ doc(0, 3, 1),             "", "SRL_MAGIC_STRING protocol 3 with small payload is Sereal"],
+
+    [SRL_MAGIC_STRING_HIGHBIT,  "", "SRL_MAGIC_STRING_HIGHBIT alone is not Sereal"],
+    [ doc(     1, 0, 1),        "", "SRL_MAGIC_STRING_HIGHBIT with bad protocol is not Sereal"],
+    [ doc(     1, 1, 0),        "", "SRL_MAGIC_STRING_HIGHBIT protocol 1 with short body is not Sereal"],
+    [ doc(     1, 1, 1),        "", "SRL_MAGIC_STRING_HIGHBIT protocol 1 with small payload is not Sereal"],
+    [ doc(     1, 2, 0),        "", "SRL_MAGIC_STRING_HIGHBIT protocol 2 with short body is not Sereal"],
+    [ doc(     1, 2, 1),        "", "SRL_MAGIC_STRING_HIGHBIT protocol 2 with small payload is not Sereal"],
+    [ doc(     1, 3, 0),        "", "SRL_MAGIC_STRING_HIGHBIT protocol 3 with short body is not Sereal"],
+    [ doc(     1, 3, 1),         3, "SRL_MAGIC_STRING_HIGHBIT protocol 3 with small payload is Sereal"],
+    [ doc("utf8", 3, 1),         0, "SRL_MAGIC_STRING_HIGHBIT_UTF8 protocol 3 with small payload is identified as utf8"],
+
+    ["=Srl". chr(1) . chr(0) . chr(SRL_HDR_UNDEF), "", "wrong magic string is not Sereal"],
 );
 
-plan tests => @tests * 3;
+plan tests => 2 + @tests * 5;
+
+is(prototype(\&looks_like_sereal), undef);
+is(prototype(\&scalar_looks_like_sereal), "\$");
 
 my $decoder = Sereal::Decoder->new;
 foreach my $t (@tests) {
     my ($input, $outcome, $name) = @$t;
-    ok(looks_like_sereal($input) == $outcome, $name . " (function)");
-    ok($decoder->looks_like_sereal($input) == $outcome, $name . " (object method)");
-    ok(Sereal::Decoder->looks_like_sereal($input) == $outcome, $name . " (class method)");
+    is( scalar_looks_like_sereal($input), $outcome, "$name (new function oppable)" );
+    is( &scalar_looks_like_sereal($input), $outcome, "$name (new function non-oppable)" );
+    is( looks_like_sereal($input), $outcome, "$name (old function)" );
+    is( $decoder->looks_like_sereal($input), $outcome, "$name (object method)" );
+    is( Sereal::Decoder->looks_like_sereal($input), $outcome, "$name (class method)" );
 }
@@ -0,0 +1,43 @@
+use strict;
+use warnings;
+
+use Sereal::Decoder;
+use Test::More;
+use File::Spec;
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+use Sereal::TestSet qw(:all);
+
+if (have_encoder_and_decoder()) {
+    plan tests => 6;
+} else {
+    plan skip_all => 'Did not find right version of encoder';
+}
+my $enc = Sereal::Encoder->new;
+my $dec = Sereal::Decoder->new;
+
+sub desc_special($) {
+        return $_[0] == \undef() ? "undef" :
+                $_[0] == \!1 ? "false" :
+                $_[0] == \!0 ? "true" :
+                "not-special";
+}
+
+foreach(
+        [ "ref undef",     \undef(),                                            ],
+        [ "ref undef var", \do { my $z = undef },                               ],
+        [ "ref false",     \!1,                                                 ],
+        [ "ref false var", \do { my $z = !1 },                                  ],
+        [ "ref true",      \!0,                                                 ],
+        [ "ref true var ", \do { my $z = !0 },                                  ],
+) {
+    my ($name, $var, $todo)= @$_;
+    TODO: {
+        todo_skip $todo, 1 if $todo;
+        is( desc_special($dec->decode($enc->encode($var))), desc_special($var), $name );
+
+    }
+}
@@ -0,0 +1,38 @@
+#!perl
+use strict;
+use warnings;
+use Sereal::Decoder;
+use Test::More;
+use File::Spec;
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+use Sereal::TestSet qw(:all);
+
+if (have_encoder_and_decoder()) {
+    plan tests => 1004;
+}
+else {
+    plan skip_all => 'Did not find right version of encoder';
+}
+
+my $e= Sereal::Encoder->new();
+my $d= Sereal::Decoder->new();
+
+for ( 1 .. 1000, [ 'a' .. 'z' ], [ 'A' .. 'Z' ], [ 0 .. 100 ], [ 10000 .. 10512 ] ) {
+    my %hash;
+    if (ref $_) {
+        $hash{$_}++ for @$_;
+    } else {
+        $hash{rand()}++ for 1..26;
+    }
+    my $undump= $d->decode($e->encode(\%hash));
+    my $count= 0;
+    while( my ($h, $k)= each %$undump ) {
+        $count++;
+    }
+    is($count, keys %hash, "Got the expected count of keys: [ @{[ sort keys %hash ]} ]"); 
+}
+
@@ -0,0 +1,58 @@
+use strict;
+use warnings;
+
+use Sereal::Decoder;
+use Test::More;
+use File::Spec;
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+use Sereal::TestSet qw(:all);
+
+my @tests= (
+    [ 15,  alias_smallint => 1  ],
+    [ 127, alias_varint_under => 128  ],
+);
+
+if (have_encoder_and_decoder()) {
+    my $num_tests= 0;
+    $num_tests += ((16 + $_->[0] + 2) * 2) for @tests;
+    plan tests => $num_tests;
+} else {
+    plan skip_all => 'Did not find right version of encoder';
+}
+foreach my $test (@tests) {
+    my ($up_to, $opt, $opt_val)= @$test;
+    #diag "$up_to: $opt $opt_val";
+
+    my $enc = Sereal::Encoder->new;
+    my $dec = Sereal::Decoder->new( { $opt => $opt_val } );
+
+    my $struct= {
+        array => [-16 .. $up_to],
+        array2 => [reverse -16 .. $up_to],
+        map { $_ => $_ } -16 .. $up_to,
+    };
+    my $got= $dec->decode($enc->encode($struct));
+
+    # undef the decoder to make sure if it blows up on DESTROY it does it before we test.
+    undef $dec;
+    undef $enc;
+
+    # Make sure we get the expected aliases
+    for (-16..$up_to) {
+        ok(\$got->{array}[$_+16] == \$got->{array2}[- 1 - ($_+16)],"$opt: array alias: $_");
+        ok(\$got->{$_} == \$got->{array}[$_+16],"$opt: array alias: $_");
+    }
+
+    # Make sure the aliases are readonly.
+    my $eval_ok= eval {
+        $got->{$up_to}= 123;
+    };
+    my $error= $eval_ok ? "" : ("$@" || "Zombie error");
+    ok(!$eval_ok,"$opt: expect modification of \$got->{$up_to} to die");
+    like($error,qr/read-only/,"$opt: expect an error about read-only values");
+}
+
@@ -0,0 +1,85 @@
+use strict;
+use warnings;
+
+use Sereal::Decoder;
+use Test::More;
+use File::Spec;
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+use Sereal::TestSet qw(:all);
+
+my @tests= (
+    [ set_readonly => 1  ],
+);
+
+if (have_encoder_and_decoder()) {
+    my $num_tests= 26;
+    plan tests => $num_tests;
+} else {
+    plan skip_all => 'Did not find right version of encoder';
+}
+
+my $foo = bless([ 1, 2, 3 ],"foo");
+
+my $struct= {
+    hashref => { a => [ "b", 5, bless({ foo => "bar"}, "SomeClass")] },
+    string => "foobar",
+    arrayref => [ "foobar" ],
+    blessed_arrayref => $foo,
+};
+
+foreach my $name ( keys %$struct ) {
+
+    local $_ = $struct->{$name};
+    my $enc = Sereal::Encoder->new;
+    my $dec = Sereal::Decoder->new( { set_readonly => 1 } );
+    my $dec2 = Sereal::Decoder->new( { set_readonly_scalars => 1 } );
+
+    my $got;
+    $dec->decode($enc->encode($_), $got);
+    my $got2;
+    $dec2->decode($enc->encode($_), $got2);
+
+    # undef the decoder to make sure if it blows up on DESTROY it does it before we test.
+    undef $dec;
+    undef $dec2;
+    undef $enc;
+
+    _recurse($got, '', $name, 0);
+    _recurse($got2, '', $name, 1);
+
+}
+
+sub _recurse {
+    my ($s, $path, $name, $scalars_only) = @_;
+
+    $scalars_only ||= 0;
+    my $should_be_readonly = $scalars_only ? !ref($s) : 1;
+    is(Internals::SvREADONLY($_[0]), $should_be_readonly,
+       "scalar_only: '$scalars_only'. We want ro: '$should_be_readonly'. struct: $name, path: $path"
+      );
+
+    my $ref = ref $s
+      or return;
+
+    if ($ref eq 'ARRAY' || $ref eq 'foo') { 
+        my $i = 0;
+        foreach (@$s) {
+            _recurse($_, $path . '->[' . $i . ']', $name, $scalars_only);
+        }
+    }
+    elsif ($ref eq 'HASH' || $ref eq 'SomeClass') {
+        foreach (keys %$s) {
+            _recurse($s->{$_}, $path . '->{' . $_ . '}', $name, $scalars_only);
+        }
+    } elsif ($ref eq 'SCALAR') {
+        _recurse($$s, '${' . $path . '}', $name, $scalars_only);
+    } else {
+        die "unknown ref value '$ref'";
+    }
+}
+
+
@@ -1,38 +0,0 @@
-#!perl
-use strict;
-use warnings;
-use Sereal::Decoder;
-use Data::Dumper;
-use File::Spec;
-
-# These tests use an installed Decoder (or respectively Encoder) to do
-# round-trip testing. There are two strategies, both with drawbacks:
-# - Test::More's is_deeply is waaaay too lenient to catch all the
-#   subtleties that Sereal is supposed to encode.
-# - Serialize - Deserialize - Serialize, then do a string compare.
-#   This won't catch if the first serialization has bogus output
-#   but the subsequent de- & serialization work for the already
-#   bogus output.
-# These tests can't replace carefully crafted manual tests, I fear.
-
-use lib File::Spec->catdir(qw(t lib));
-BEGIN {
-    lib->import('lib')
-        if !-d 't';
-}
-
-use Sereal::TestSet qw(:all);
-use Test::More;
-
-my $ok = have_encoder_and_decoder();
-if (not $ok) {
-    plan skip_all => 'Did not find right version of encoder';
-}
-else {
-    run_roundtrip_tests(2); # 2 == run only tests for proto v2
-}
-
-
-pass();
-done_testing();
-
@@ -1,38 +0,0 @@
-#!perl
-use strict;
-use warnings;
-use Sereal::Decoder;
-use Data::Dumper;
-use File::Spec;
-
-# These tests use an installed Decoder (or respectively Encoder) to do
-# round-trip testing. There are two strategies, both with drawbacks:
-# - Test::More's is_deeply is waaaay too lenient to catch all the
-#   subtleties that Sereal is supposed to encode.
-# - Serialize - Deserialize - Serialize, then do a string compare.
-#   This won't catch if the first serialization has bogus output
-#   but the subsequent de- & serialization work for the already
-#   bogus output.
-# These tests can't replace carefully crafted manual tests, I fear.
-
-use lib File::Spec->catdir(qw(t lib));
-BEGIN {
-    lib->import('lib')
-        if !-d 't';
-}
-
-use Sereal::TestSet qw(:all);
-use Test::More;
-
-my $ok = have_encoder_and_decoder();
-if (not $ok) {
-    plan skip_all => 'Did not find right version of encoder';
-}
-else {
-    run_roundtrip_tests(1); # 1 == run only tests for proto v1
-}
-
-
-pass();
-done_testing();
-
@@ -19,11 +19,11 @@ use Sereal::TestSet qw(:all);
 # bad input. This obviously shouldn't segfault and neither leak
 # memory.
 
-plan tests => 47;
+plan tests => 56;
 my ($ok, $out, $err);
 
 SCOPE: {
-    check_fail(Header(), qr/unexpected end of input/i, "Cannot decode just header");
+    check_fail(Header(), qr/Not a valid Sereal document/i, "Cannot decode just header");
 
     my $badheaderpacket = "srX".chr(SRL_PROTOCOL_VERSION) . chr(0) . integer(1);
     check_fail($badheaderpacket, qr/Bad Sereal header/i, "Packet with invalid header blows up");
@@ -50,6 +50,20 @@ SCOPE: {
 
     check_fail($hash_packet, qr/Sereal: Error/, "Setting hash limit option (1)", {max_num_hash_entries => 1});
     check_fail($hash_packet, qr/Sereal: Error/, "Setting hash limit option (999)", {max_num_hash_entries => 999});
+
+    my $valid_packet = Header(2) . short_string("foo");
+    my $foo = decode_sereal($valid_packet);
+    is($foo, "foo", "Have valid test packet");
+    $valid_packet =~ s/^=srl/=\xF3rl/;
+    $foo = eval { decode_sereal($valid_packet) };
+    ok(!defined($foo), "SRL_MAGIC_STRING_HIGHBIT implies protocol v3 or higher.");
+
+    substr($valid_packet,4,1,chr(3));
+    $foo = eval { decode_sereal($valid_packet) };
+    is($foo,"foo", "Have valid test packet after asserting high bit in magic with protocol v3");
+
+    utf8::encode($valid_packet);
+    check_fail($valid_packet, qr/UTF-8/, "Sereal determined 'accidental' UTF8 upgrade");
 }
 
 pass("Alive"); # done
@@ -0,0 +1,34 @@
+#!perl
+use strict;
+use warnings;
+use Sereal::Decoder;
+use Data::Dumper;
+use File::Spec;
+
+# These tests use an installed Decoder (or respectively Encoder) to do
+# bulk data testing.
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Sereal::BulkTest qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    my %opt = (
+        bench => scalar(grep /^--bench$/, @ARGV),
+    );
+    run_bulk_tests(%opt, decoder_options => { set_readonly => 1});
+}
+
+pass();
+done_testing();
+
@@ -0,0 +1,34 @@
+use strict;
+use warnings;
+use Test::More tests => 8;
+use Sereal::Decoder;
+
+my $decoder= Sereal::Decoder->new();
+my $enc_ref= "=\363rl\3\0Qcfoo\1";
+my $enc_str= "=\363rl\3\0}blah blah blah blah blah blah";
+
+# Repeatedly decode "into" the $into variable.
+# Overwrting/reusing the data it contains.
+# We alternate between ref and scalar to see if we can trigger a segfault.
+
+my $into;
+$decoder->decode($enc_ref, $into);
+ok(ref $into, "first decode was a reference");
+$decoder->decode($enc_str, $into);
+ok(!ref $into, "second decode was a string");
+
+$decoder->decode($enc_ref, $into);
+ok(ref $into, "third decode was a reference");
+$decoder->decode($enc_str, $into);
+ok(!ref $into, "fourth decode was a string (and did not segfault)");
+
+$decoder->decode($enc_ref, $into);
+ok(ref $into, "fifth decode was a reference - and did not segault");
+$decoder->decode($enc_str, $into);
+ok(!ref $into, "sixth decode was a string  - and did not segfault, probably ok");
+
+$decoder->decode($enc_ref, $into);
+ok(ref $into, "seventh decode was a reference - maybe overkill");
+$decoder->decode($enc_str, $into);
+ok(!ref $into, "eight decode was a string - maybe overkill");
+
@@ -0,0 +1,28 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $version;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests("plain");
+}
+
+pass();
+done_testing();
+
@@ -0,0 +1,27 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+$ok= 0 if $ok and $Sereal::Encoder::VERSION < 3.001006;
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder (want 3.001006)';
+}
+else {
+    run_roundtrip_tests("plain_canon", { canonical => 1 });
+}
+
+pass();
+done_testing();
+
@@ -0,0 +1,27 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests('snappy', { snappy           => 1 } );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,27 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+$ok= 0 if $ok and $Sereal::Encoder::VERSION < 3.001006;
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder (want 3.001006)';
+}
+else {
+    run_roundtrip_tests('snappy_canon', { snappy => 1, canonical => 1 } );
+}
+
+pass();
+done_testing();
+
@@ -0,0 +1,29 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests(
+        'dedupe_strings', { dedupe_strings   => 1 } 
+    );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,29 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests(
+        'freeze-thaw',    { freeze_callbacks => 1 }
+    );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,28 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $version;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests("plain");
+}
+
+pass();
+done_testing();
+
@@ -0,0 +1,27 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+$ok= 0 if $ok and $Sereal::Encoder::VERSION < 3.001006;
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder (want 3.001006)';
+}
+else {
+    run_roundtrip_tests("plain_canon", { canonical => 1 });
+}
+
+pass();
+done_testing();
+
@@ -0,0 +1,29 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests(
+        'readonly',       { set_readonly     => 1 } 
+    );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,27 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests('snappy', { snappy           => 1 } );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,27 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+$ok= 0 if $ok and $Sereal::Encoder::VERSION < 3.001006;
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder (want 3.001006)';
+}
+else {
+    run_roundtrip_tests('snappy_canon', { snappy => 1, canonical => 1 } );
+}
+
+pass();
+done_testing();
+
@@ -0,0 +1,29 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests(
+        'snappy_incr',    { snappy_incr      => 1 } 
+    );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,27 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+$ok= 0 if $ok and $Sereal::Encoder::VERSION < 3.001006;
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder (want 3.001006)';
+}
+else {
+    run_roundtrip_tests('snappy_incr_canon', { snappy_incr => 1, canonical => 1 });
+}
+
+pass();
+done_testing();
+
@@ -0,0 +1,29 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests(
+        'sort_keys',      { sort_keys        => 1 }
+    );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,29 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests(
+        'dedupe_strings', { dedupe_strings   => 1 } 
+    );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,29 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests(
+        'freeze-thaw',    { freeze_callbacks => 1 }
+    );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,28 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $version;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests("plain");
+}
+
+pass();
+done_testing();
+
@@ -0,0 +1,27 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+$ok= 0 if $ok and $Sereal::Encoder::VERSION < 3.001006;
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder (want 3.001006)';
+}
+else {
+    run_roundtrip_tests("plain_canon", { canonical => 1 });
+}
+
+pass();
+done_testing();
+
@@ -0,0 +1,29 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests(
+        'readonly',       { set_readonly     => 1 } 
+    );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,27 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests('snappy', { snappy           => 1 } );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,27 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+$ok= 0 if $ok and $Sereal::Encoder::VERSION < 3.001006;
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder (want 3.001006)';
+}
+else {
+    run_roundtrip_tests('snappy_canon', { snappy => 1, canonical => 1 } );
+}
+
+pass();
+done_testing();
+
@@ -0,0 +1,29 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests(
+        'snappy_incr',    { snappy_incr      => 1 } 
+    );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,27 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+$ok= 0 if $ok and $Sereal::Encoder::VERSION < 3.001006;
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder (want 3.001006)';
+}
+else {
+    run_roundtrip_tests('snappy_incr_canon', { snappy_incr => 1, canonical => 1 });
+}
+
+pass();
+done_testing();
+
@@ -0,0 +1,29 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests(
+        'sort_keys',      { sort_keys        => 1 }
+    );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,32 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests(
+        'zlib', 
+        { 
+            compress => Sereal::Encoder::SRL_ZLIB(),
+        }
+    );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,33 @@
+#!perl
+use strict;
+use warnings;
+use Data::Dumper;
+use File::Spec;
+
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More;
+
+my $ok = have_encoder_and_decoder();
+if (not $ok) {
+    plan skip_all => 'Did not find right version of encoder';
+}
+else {
+    run_roundtrip_tests(
+        'zlib_force',     
+        { 
+            compress => Sereal::Encoder::SRL_ZLIB(), 
+            compress_threshold => 0,
+        } 
+    );
+}
+
+
+pass();
+done_testing();
+
@@ -0,0 +1,15 @@
+#!perl
+use strict;
+use warnings;
+use Sereal::Decoder qw(decode_sereal);
+use Sereal::Decoder::Constants qw(:all);
+use Test::More tests => 4;
+
+for my $ref (\"", [], {}, \*STDERR) {
+    eval {
+        decode_sereal($ref);
+        1;
+    } or do {
+        like($@, qr/We can't decode a reference as Sereal!/, "We'll die on " . ref($ref) . " references");
+    };
+}
@@ -0,0 +1,47 @@
+use strict;
+use warnings;
+use Sereal::Decoder;
+use File::Spec;
+use lib File::Spec->catdir(qw(t lib));
+BEGIN {
+    lib->import('lib')
+        if !-d 't';
+}
+
+use Sereal::TestSet qw(:all);
+use Test::More tests => 2;
+
+# Regression test for RT #93563
+
+# Decoder is (was) not re-entrant.
+
+my $dec;
+package Foo;
+sub FREEZE { my $x = Sereal::Encoder->new->encode($_[0]->{a}); return $x; }
+sub THAW { bless({a => $dec->decode($_[2])}, $_[0]) }
+
+package main;
+
+SKIP: {
+    my $have_enc = have_encoder_and_decoder();
+    if (not $have_enc) {
+        skip "Need encoder for Snappy regression tests", 2;
+    }
+    else {
+        $dec = Sereal::Decoder->new;
+        my $z = [ bless({a=>42},"Foo") ];
+        push @$z, $z;
+        my $a = Sereal::Encoder->new({freeze_callbacks=>1})->encode($z);
+        my $b;
+        my $err;
+        eval {
+            $b = $dec->decode($a);
+            1
+        } or do {
+            $err = $@ || 'Zombie error';
+        };
+        ok(!$err, "Decoding did not barf")
+            or diag("Decoding barfed with '$err'");
+        is_deeply($b, $z, "Output from decoding is correct");
+    }
+}
@@ -66,12 +66,12 @@ sub read_files {
     }
 
     my $count= 0;
-    foreach (@$corpus) {
-        $count++ if $sub->($_);
+    foreach my $test (@$corpus) {
+        $count++ if $sub->($test);
     }
     return $count;
 }
-
+#use Devel::Peek;
 sub run_bulk_tests {
     my %opt = @_;
 
@@ -79,21 +79,30 @@ sub run_bulk_tests {
         my $total= read_files(sub { return 1 });
         my $read= 0;
         my $eval_ok= read_files(sub {
+            my $struct= $_[0];
             diag("read $read\n") unless ++$read % 1000;
-            my ($dump,$undump);
+            my ($dump, $undump);
             my $ok= eval {
                 $dump = Sereal::Encoder::encode_sereal($_[0]);
-                $undump= Sereal::Decoder::decode_sereal($dump);
+                $undump= Sereal::Decoder::decode_sereal($dump, $opt{decoder_options} || {});
                 1;
             };
             my $err = $@ || 'Zombie error';
             ok($ok,"Error return is empty")
                 or diag("Error was: '$err'"), return $ok;
+            if ($ok and ref($struct) eq "HASH") {
+                my $each_count= 0;
+
+                $each_count++ while my($k,$v)= each %$undump;
 
-            my $eval_dump= Data::Dumper->new([ $_[0] ])->Sortkeys(1)->Dump();
-            my $undump_dump= Data::Dumper->new([ $undump ])->Sortkeys(1)->Dump();
-            $ok= is_string($undump_dump, $eval_dump)
-                or diag $_[0];
+                my $keys_count= 0 + keys %$struct;
+                is($each_count,$keys_count,"Number of keys match");
+            }
+
+            my $struct_dd= Data::Dumper->new([ $struct ])->Sortkeys(1)->Dump();
+            my $undump_dd= Data::Dumper->new([ $undump ])->Sortkeys(1)->Dump();
+            $ok= is_string($undump_dd, $struct_dd)
+                or diag $struct_dd;
             return $ok;
         });
         is($total,$eval_ok);
@@ -110,7 +119,7 @@ sub run_bulk_tests {
                     read_files(sub{return 1})
                 },
                 'decode_sereal' => sub{
-                    read_files(sub { return( decode_sereal($_[0]) ); }, 'sereal')
+                    read_files(sub { return( decode_sereal($_[0], $opt{decoder_options} || {} ) ); }, 'sereal')
                 },
                 'eval' => sub{
                     read_files(sub { return( eval $_[0] ); }, 'raw')
@@ -130,5 +139,4 @@ sub run_bulk_tests {
         note join "\n","", map {sprintf"%-20s" . (" %20s" x (@$_-1)), @$_ } @$result;
     }
 }
-
 1;
@@ -10,18 +10,20 @@ use Test::More;
 use Test::LongString;
 #use Data::Dumper; # MUST BE LOADED *AFTER* THIS FILE (BUG IN PERL)
 use Devel::Peek;
-use Encode qw(encode_utf8);
+use Encode qw(encode_utf8 is_utf8);
+use Scalar::Util qw(reftype blessed refaddr);
+use Config;
 
 # Dynamically load constants from whatever is being tested
 our ($Class, $ConstClass);
 BEGIN {
-    if (defined $INC{"Sereal/Encoder.pm"}
-        and $INC{"Sereal/Encoder.pm"} =~ /\bblib\b/)
-    {
+    if (-e "lib/Sereal/Encoder") {
         $Class = 'Sereal::Encoder';
     }
-    else {
+    elsif (-e "lib/Sereal/Decoder") {
         $Class = 'Sereal::Decoder';
+    } else {
+        die "Could not find an applicable Sereal constants location";
     }
     $ConstClass = $Class . "::Constants";
     eval "use $ConstClass ':all'; 1"
@@ -36,7 +38,7 @@ our @ISA = qw(Exporter);
 our @EXPORT_OK = qw(
     @BasicTests $Class $ConstClass
     Header
-    FBIT
+    TRACK_FLAG
     hobodecode
     integer short_string varint array array_fbit
     hash dump_bless
@@ -45,16 +47,20 @@ our @EXPORT_OK = qw(
     write_test_files
     $use_objectv
     setup_tests
+    _deep_cmp
+    _test
+    _cmp_str
 );
 
 our %EXPORT_TAGS = (all => \@EXPORT_OK);
 our $use_objectv = 1;
 
-use constant FBIT => 128;
+use constant TRACK_FLAG => 128;
 
 sub hobodecode {
+    return unless defined $_[0];
     open my $fh, "| $^X -Mblib=../Encoder -Mblib=../Decoder author_tools/hobodecoder.pl -e" or die $!;
-    print $fh @_;
+    print $fh $_[0];
     close $fh;
 }
 
@@ -71,7 +77,7 @@ sub array {
 
 sub array_fbit {
     chr(SRL_HDR_REFN).
-    chr(SRL_HDR_ARRAY+FBIT) . varint(0+@_) . join("", @_)
+    chr(SRL_HDR_ARRAY+TRACK_FLAG) . varint(0+@_) . join("", @_)
 }
 
 sub hash_head {
@@ -110,7 +116,9 @@ sub dump_bless {
 
 sub short_string {
     die if length($_[0]) > SRL_MASK_SHORT_BINARY_LEN;
-    return chr(SRL_HDR_SHORT_BINARY_LOW + length($_[0])) . $_[0];
+    my $tag = SRL_HDR_SHORT_BINARY_LOW + length($_[0]);
+    $tag |= SRL_HDR_TRACK_FLAG if $_[1];
+    return pack("c a*",$tag,$_[0]);
 }
 
 sub integer {
@@ -141,9 +149,10 @@ sub varint {
 our $PROTO_VERSION;
 
 sub Header {
-    my $proto_version = shift || $PROTO_VERSION;
+    my $proto_version = shift || $PROTO_VERSION || SRL_PROTOCOL_VERSION;
     my $user_data_blob = shift;
-    my $hdr_base = SRL_MAGIC_STRING . chr($proto_version||SRL_PROTOCOL_VERSION);
+    my $mgc = $proto_version > 2 ? SRL_MAGIC_STRING_HIGHBIT : SRL_MAGIC_STRING;
+    my $hdr_base = $mgc . chr($proto_version);
     if (defined $user_data_blob) {
         return $hdr_base . varint(1 + length($user_data_blob)) . chr(1) . $user_data_blob;
     }
@@ -172,13 +181,18 @@ sub offseti {
 }
 
 sub debug_checks {
-    my ($data_ref, $encoded_ref, $decoded_ref) = @_;
-    if (defined $ENV{DEBUG_SEREAL}) {
-        note("Original data was: " . Data::Dumper::Dumper($$data_ref)) if defined $data_ref;
-        note("Encoded data is: " . (defined($$encoded_ref) ? $$encoded_ref : "<undef>")) if defined $encoded_ref;
-        note("Decoded data was: " . Data::Dumper::Dumper($$decoded_ref)) if defined $decoded_ref;
+    my ($data_ref, $encoded_ref, $decoded_ref, $debug) = @_;
+    if ($debug or defined $ENV{DEBUG_SEREAL}) {
+        require Data::Dumper;
+        note("Original data was: " . Data::Dumper::Dumper($$data_ref))
+            if defined $data_ref;
+        note("Encoded data is: " . (defined($$encoded_ref) ? Data::Dumper::qquote($$encoded_ref) : "<undef>"))
+            if defined $encoded_ref;
+        note("Decoded data was: " . Data::Dumper::Dumper($$decoded_ref))
+            if defined $decoded_ref;
     }
     if (defined $ENV{DEBUG_DUMP}) {
+        Dump($$data_ref)    if defined $data_ref;
         Dump($$encoded_ref) if defined $encoded_ref;
         Dump($$decoded_ref) if defined $decoded_ref;
     }
@@ -199,7 +213,11 @@ sub setup_tests {
 
     my $unicode1= "Ba\xDF Ba\xDF"; my $unicode2= "\x{168}nix! \x{263a}"; utf8::upgrade($unicode1); utf8::upgrade($unicode2);
 
-
+    # each test is an array:
+    # index 0 is the input to the encoder
+    # index 1 is the output *without* header - or a sub which returns an expected output
+    # index 2 is the name of the test
+    # index 3 and on are alternate outputs (or subs which return alternate output(s))
     @BasicTests = (
         # warning: this hardcodes the POS/NEG headers
         [-16, chr(0b0001_0000), "encode -16"],
@@ -234,7 +252,7 @@ sub setup_tests {
                   my $d = array_head(3);
                   my $pos = offset($d);
                   my $tag = $opt->{aliased_dedupe_strings} ? SRL_HDR_ALIAS : SRL_HDR_COPY;
-                  $d .= short_string("foooo") . chr($tag) . varint($pos)
+                  $d .= short_string("foooo",$opt->{aliased_dedupe_strings} ? 1 : 0) . chr($tag) . varint($pos)
                         . chr($tag) . varint($pos);
                   return $d;
               }
@@ -258,7 +276,7 @@ sub setup_tests {
                   my $tag = $opt->{aliased_dedupe_strings} ? SRL_HDR_ALIAS : SRL_HDR_COPY;
                   my $d = array_head(2) . hash_head(2) . short_string("foooo");
                   my $pos = offset($d);
-                  $d .= short_string("foooo") . hash_head(2)
+                  $d .= short_string("foooo",$opt->{aliased_dedupe_strings} ? 1 : 0) . hash_head(2)
                         . short_string("foooo2")
                         . chr($tag) . varint($pos);
                   return $d;
@@ -310,7 +328,7 @@ sub setup_tests {
         [
             $weak_thing,
             chr(SRL_HDR_REFN) 
-            . chr(SRL_HDR_ARRAY + FBIT) . varint(2)
+            . chr(SRL_HDR_ARRAY + TRACK_FLAG) . varint(2)
                 . chr(SRL_HDR_PAD) . chr(SRL_HDR_REFN) 
                     . chr(SRL_HDR_REFP) . varint(offseti(1))
                 . chr(0b0000_0001)
@@ -320,7 +338,7 @@ sub setup_tests {
         [
             \$weak_thing,
             chr(SRL_HDR_REFN)
-            . chr(SRL_HDR_REFN + FBIT)
+            . chr(SRL_HDR_REFN + TRACK_FLAG)
                 . chr(SRL_HDR_ARRAY) . varint(2)
                     .chr(SRL_HDR_WEAKEN) . chr(SRL_HDR_REFP) . varint(offseti(1))
                     .chr(0b0000_0001)
@@ -329,7 +347,7 @@ sub setup_tests {
         ],
         sub { \@_ } ->(
             $weak_thing,
-            chr(SRL_HDR_REFN + FBIT)
+            chr(SRL_HDR_REFN + TRACK_FLAG)
                 .chr(SRL_HDR_ARRAY).varint(2)
                     .chr(SRL_HDR_WEAKEN).chr(SRL_HDR_REFP).varint(offseti(0))
                     .chr(0b0000_0001)
@@ -342,8 +360,8 @@ sub setup_tests {
                 my $content= array_head(2);
                 my $pos= offset($content);
                 $content
-                . chr(SRL_HDR_REFN + FBIT)
-                . chr(SRL_HDR_REFP + FBIT)
+                . chr(SRL_HDR_REFN + TRACK_FLAG)
+                . chr(SRL_HDR_REFP + TRACK_FLAG)
                 . varint( $pos )
                 . chr(SRL_HDR_ALIAS)
                 . varint($pos + 1)
@@ -356,9 +374,9 @@ sub setup_tests {
                 my $content= array_head(2);
                 my $pos= offset($content);
                 $content
-                . chr(SRL_HDR_WEAKEN + FBIT)
+                . chr(SRL_HDR_WEAKEN + TRACK_FLAG)
                 . chr(SRL_HDR_REFN)
-                . chr(SRL_HDR_WEAKEN + FBIT)
+                . chr(SRL_HDR_WEAKEN + TRACK_FLAG)
                 . chr(SRL_HDR_REFP)
                 . varint($pos)
                 . chr(SRL_HDR_ALIAS)
@@ -380,7 +398,7 @@ sub setup_tests {
                     chr(SRL_HDR_OBJECT),
                     short_string("bar"),
                     chr(SRL_HDR_REFN),
-                    chr(SRL_HDR_REGEXP + FBIT),
+                    chr(SRL_HDR_REGEXP + TRACK_FLAG),
                     short_string("foo"),
                     short_string("ix"),
                     chr(SRL_HDR_REFP),
@@ -396,10 +414,10 @@ sub setup_tests {
                 my $pos= offset($content);
                 join("",$content,
                             short_string("foo"),
-                            chr(SRL_HDR_REFN).chr(SRL_HDR_ARRAY + FBIT),varint(0),
+                            chr(SRL_HDR_REFN).chr(SRL_HDR_ARRAY + TRACK_FLAG),varint(0),
                         chr( SRL_HDR_OBJECT + $use_objectv),
                             $use_objectv ? () : chr(SRL_HDR_COPY), varint($pos),
-                            chr(SRL_HDR_REFN).chr(SRL_HDR_ARRAY  + FBIT), varint(0),
+                            chr(SRL_HDR_REFN).chr(SRL_HDR_ARRAY  + TRACK_FLAG), varint(0),
                         chr(SRL_HDR_REFP),varint($pos + 5),
                         chr(SRL_HDR_REFP),varint($pos + 10),
                     )
@@ -507,11 +525,33 @@ sub setup_tests {
             ),
             "simple unicode hash key and value"
         ],
+        # Test true/false. Due to some edge case behavior in perl these two tests
+        # produce different "expected" results depending on such things as how many
+        # times we perform the test. Therefore we allow various "alternates" to
+        # be produced. An example of the underlying weirdness is that on an unthreaded
+        # linux perl 5.14 the two tests have their expected output first, which
+        # as you will note is different for the first and second call, despite the underlying
+        # code being the same both times.
+        #
+        # So for instance the first test need not have the last two options, at least
+        # on perl 5.14, but the second test requires one of those options. Working around
+        # perl bugs sucks.
         [
             sub { \@_ }->(!1,!0),
+            array(chr(SRL_HDR_FALSE),chr(SRL_HDR_TRUE)),  # this is the "correct" response.
+            "true/false (prefered order)",
+            array(chr(SRL_HDR_FALSE),short_string("1")),  # this is what threaded perls will probably match
+            array(short_string(""),chr(SRL_HDR_TRUE)),    # accept this also (but we dont expect we will)
+            array(short_string(""),short_string("1")),    # accept this also (but we dont expect we will)
+        ],
+        [
+            sub { \@_ }->(!1,!0),
+            array(short_string(""),short_string("1")),    # this is the expected value on perl 5.14 unthreaded
+            "true/false (reversed alternates)",
+            array(short_string(""),chr(SRL_HDR_TRUE)),    # from here we just reverse the order from the first test
+            array(chr(SRL_HDR_FALSE),short_string("1")),  # ....
             array(chr(SRL_HDR_FALSE),chr(SRL_HDR_TRUE)),
-            "true/false"
-        ]
+        ],
     );
 }
 
@@ -528,6 +568,7 @@ sub get_git_top_dir {
 }
 
 sub have_encoder_and_decoder {
+    my ($min_v)= @_;
     # $Class is the already-loaded class, so the one we're testing
     my $need = $Class =~ /Encoder/ ? "Decoder" : "Encoder";
     my $need_class = "Sereal::$need";
@@ -547,6 +588,11 @@ sub have_encoder_and_decoder {
         return();
     };
     my $cmp_v = $need_class->VERSION;
+    if ($min_v and $cmp_v <= $min_v) {
+        note("Could not load correct version of $need_class for testing "
+             ."(got: $cmp_v, needed at least $min_v)");
+        return;
+    }
     $cmp_v =~ s/_//;
     $cmp_v = sprintf("%.2f", int($cmp_v*100)/100);
     if (not defined $cmp_v or not exists $compat_versions{$cmp_v}) {
@@ -554,7 +600,6 @@ sub have_encoder_and_decoder {
              ."(got: $cmp_v, needed any of ".join(", ", keys %compat_versions).")");
         return();
     }
-
     return 1;
 }
 
@@ -576,16 +621,24 @@ my $lots_of_9C = do {
 my $max_iv = ~0 >> 1;
 my $min_iv = do {use integer; -$max_iv-1}; # 2s complement assumption
 
+my $eng0e0= "0e0";
+my $eng0e1= "0e1";
+my $eng2= "1e3";
+
+my $sum= $eng0e0 + $eng0e1 + $eng2;
+
 our @ScalarRoundtripTests = (
     # name, structure
     ["undef", undef],
     ["small int", 3],
     ["small negative int", -8],
     ["largeish int", 100000],
-    ["largeish negative int", -302001],
+    ["largeish negative int -302001",   -302001],
+    ["largeish negative int -1234567",  -1234567],
+    ["largeish negative int -12345678", -12345678],
 
     (
-        map {["integer: $_", $_]} (
+        map {["integer: $_", 0+$_]} (
             # IV bounds of 8 bits
             -1, 0, 1, -127, -128, -129, 42, 126, 127, 128, 129, 254, 255, 256, 257,
             # IV bounds of 32 bits
@@ -597,25 +650,47 @@ our @ScalarRoundtripTests = (
             0x7FFFFFFF, 0x80000000, 0x80000001, 0xFFFFFFFF, 0xDEADBEEF,
             # UV bounds
             $max_iv_p1, $max_uv_m1, $max_uv, $lots_of_9C,
+            $eng0e0, $eng0e1, $eng2,
         )
     ),
-
-    ["float", 0.2],
+    (map { ["float $_", 0+$_] } (0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9)),
     ["short ascii string", "fooo"],
     ["short latin1 string", "Müller"],
-    ["short utf8 string", do {use utf8; " עדיין ח"}],
-    ["long ascii string", do{"abc" x 1000}],
-    ["long latin1 string", "üll" x 1000],
-    ["long utf8 string", do {use utf8; " עדיין חשב" x 1000}],
-    ["long utf8 string with only ascii", do {use utf8; "foo" x 1000}],
-    ["long utf8 string with only latin1 subset", do {use utf8; "üll" x 1000}],
+    ["short utf8 string", do {use utf8; " עדיין ח"} ],
+
+    (map { [ "long ascii string 'a' x $_", do{"a" x $_} ] } (
+        9999,10000,10001,
+        1023,1024,1025,
+        8191,8192,8193,
+    )),
+    (map { [ "long ascii string 'ab' x $_", do{"ab" x $_} ] } (
+        9999,10000,10001,
+        1023,1024,1025,
+        8191,8192,8193,
+    )),
+    (map { [ "long ascii string 'abc' x $_", do{"abc" x $_} ] } (
+        9999,10000,10001,
+        1023,1024,1025,
+        8191,8192,8193,
+    )),
+    (map { [ "long ascii string 'abcd' x $_", do{"abcd" x $_} ] } (
+        9999,10000,10001,
+        1023,1024,1025,
+        8191,8192,8193,
+    )),
+
+    ["long latin1 string", "üll" x 10000],
+    ["long utf8 string", do {use utf8; " עדיין חשב" x 10000}],
+    ["long utf8 string with only ascii", do {use utf8; "foo" x 10000}],
+    ["long utf8 string with only latin1 subset", do {use utf8; "üll" x 10000}],
+
     ["simple regexp", qr/foo/],
     ["regexp with inline modifiers", qr/(?i-xsm:foo)/],
     ["regexp with modifiers", qr/foo/i],
     ["float", 123013.139],
     ["negative float",-1234.59],
-    ["small float",0.41],
-    ["negative small float",-0.13],
+    ["small float 0.41",0.41],
+    ["negative small float -0.13",-0.13],
     ["small int", 123],
     ["empty string", ''],
     ["simple array", []],
@@ -627,7 +702,16 @@ our @ScalarRoundtripTests = (
     ["deep nest", [1,2,[3,4,{5=>6,7=>{8=>[]},9=>{}},{},[]]]],
     ["complex hash", {
         foo => 123,
-        bar => -159.23 ,
+        bar => -159, pi => 3,
+        'baz' =>"foo",
+        'bop \''=> "\10"
+        ,'bop \'\\'=> "\x{100}" ,
+        'bop \'x\\x'    =>"x\x{100}"   , 'bing' =>   "x\x{100}",
+        x=>'y', z => 'p', i=> '1', l=>" \10", m=>"\10 ", n => " \10 ",
+    }],
+    ["complex hash with float", {
+        foo => 123,
+        bar => -159.23, a_pi => 3.14159,
         'baz' =>"foo",
         'bop \''=> "\10"
         ,'bop \'\\'=> "\x{100}" ,
@@ -636,12 +720,21 @@ our @ScalarRoundtripTests = (
     }],
     ["more complex", {
         foo => [123],
-        "bar" => [-159.23 , { 'baz' => "foo", }, ],
+        "bar" => [-159, n => 3, { 'baz' => "foo", }, ],
         'bop \''=> { "\10" => { 'bop \'\\'=> "\x{100}", h=>{
         'bop \'x\\x'    =>"x\x{100}"   , 'bing' =>   "x\x{100}",
         x=>'y',}, z => 'p' ,   }   ,
         i    =>  '1' ,}, l=>" \10", m=>"\10 ", n => " \10 ",
-        o => undef ,p=>undef,
+        o => undef ,p=>undef, q=>\undef, r=>\$eng0e0, u => \$eng0e1, w=>\$eng2
+    }],
+    ["more complex with float", {
+        foo => [123],
+        "bar" => [-159.23, a_pi => 3.14159, { 'baz' => "foo", }, ],
+        'bop \''=> { "\10" => { 'bop \'\\'=> "\x{100}", h=>{
+        'bop \'x\\x'    =>"x\x{100}"   , 'bing' =>   "x\x{100}",
+        x=>'y',}, z => 'p' ,   }   ,
+        i    =>  '1' ,}, l=>" \10", m=>"\10 ", n => " \10 ",
+        o => undef ,p=>undef, q=>\undef, r=>\$eng0e0, u => \$eng0e1, w=>\$eng2
     }],
     ['var strings', [ "\$", "\@", "\%" ]],
     [ "quote keys", { "" => '"', "'" => "" }],
@@ -656,6 +749,8 @@ our @ScalarRoundtripTests = (
     [ "ref to undef", \undef],
     [ "negative big num", -4123456789],
     [ "positive big num", 4123456789],
+    [ "eng-ref", [\$eng0e0, \$eng0e1, \$eng2] ],
+    [ "undef", [\undef, \undef] ],
 );
 
 use Storable qw(dclone);
@@ -669,7 +764,13 @@ our @RoundtripTests = (
     (map {["nested scalar ref to " . $_->[0], (\\($_->[1]))]} @ScalarRoundtripTests),
     (map {["array ref to " . $_->[0], ([$_->[1]])]} @ScalarRoundtripTests),
     (map {["hash ref to " . $_->[0], ({foo => $_->[1]})]} @ScalarRoundtripTests),
+    # ---
     (map {["array ref to duplicate " . $_->[0], ([$_->[1], $_->[1]])]} @ScalarRoundtripTests),
+    (map {[
+            "AoA of duplicates " . $_->[0],
+            ( [ $_->[1], [ $_->[1], $_->[1] ], $_->[1], [ $_->[1], $_->[1], $_->[1] ], $_->[1] ] )
+         ]} @ScalarRoundtripTests),
+    # ---
     (map {["array ref to aliases " . $_->[0], (sub {\@_}->($_->[1], $_->[1]))]} @ScalarRoundtripTests),
     (map {["array ref to scalar refs to same " . $_->[0], ([\($_->[1]), \($_->[1])])]} @ScalarRoundtripTests),
 );
@@ -689,76 +790,233 @@ if (eval "use Array::RefElem (av_store hv_store); 1") {
 
 
 sub run_roundtrip_tests {
-    my ($proto_version) = @_;
-    my @proto_versions = ($proto_version ? ($proto_version) : qw(2 1));
-
-    for my $proto_version ($proto_version) {
-        my $suffix = $proto_version == 1 ? "_v1" : "";
-
-        for my $opt (
-            ['plain',          {                  } ],
-            ['snappy',         { snappy           => 1 } ],
-            ['snappy_incr',    { snappy_incr      => 1 } ],
-            ['sort_keys',      { sort_keys        => 1 } ],
-            ['dedupe_strings', { dedupe_strings   => 1 } ],
-            ['freeze/thaw',    { freeze_callbacks => 1 } ],
-        ) {
-            my ($name, $opts) = @$opt;
-            $name .= $suffix;
-            $opts->{use_protocol_v1} = 1 if $proto_version == 1;
-            $PROTO_VERSION= $proto_version;
-            setup_tests();
-            run_roundtrip_tests_internal($name, $opts);
+    my ($name, $opts) = @_;
+
+    my $proto_version;
+    if ( $0 =~ m![\\/]v(\d+)[\\/]!) {
+        $proto_version= $1;
+    } else {
+        die "Failed to detect version\n";
+    }
+
+    my $suffix = "_v$proto_version";
+    if ($proto_version == 1) {
+        $opts->{use_protocol_v1} = 1;
+    }
+    else {
+        # v2 ignores this, but will output v2 by default
+        $opts->{protocol_version} = $proto_version;
+    }
+    setup_tests($proto_version);
+    run_roundtrip_tests_internal($name . $suffix, $opts);
+}
+
+sub _test {
+    my ($msg, $v1, $v2)= @_;
+    # require Data::Dumper not needed, called in parent frame
+    if ($v1 ne $v2) {
+        my $q1= Data::Dumper::qquote($v1);
+        my $q2= Data::Dumper::qquote($v2);
+        return "$msg: $q1 ne $q2"
+    }
+    return;
+}
+
+sub _cmp_str {
+    my ($v1, $v2)= @_;
+    my $v1_is_utf8= is_utf8($v1);
+    my $v2_is_utf8= is_utf8($v2);
+
+    Encode::_utf8_off($v1); # turn off utf8, in case it is corrupt
+    Encode::_utf8_off($v2); # turn off utf8, in case it is corrupt
+    if ($v1 eq $v2) {
+        return;
+    }
+    my $diff_start= 0;
+    $diff_start++ while $diff_start < length($v1)
+                    and $diff_start < length($v2)
+                    and substr($v1, $diff_start,1) eq substr($v2, $diff_start,1);
+    my $diff_end= length($v1) < length($v2) ? length($v1) : length($v2);
+
+    $diff_end-- while $diff_end > $diff_start
+                  and $diff_end > $diff_start
+                  and substr($v1, $diff_end-1,1) eq substr($v2, $diff_end-1,1);
+    my $length_to_show= $diff_end - $diff_start;
+
+    my $max_context_len= 10;
+    my $max_diff_len= 30;
+
+    $length_to_show= $max_diff_len if $length_to_show > $max_diff_len;
+
+    # require Data::Dumper not needed, called in parent frame
+    my $q1= Data::Dumper::qquote(substr($v1, $diff_start, $length_to_show ));
+    my $q2= Data::Dumper::qquote(substr($v2, $diff_start, $length_to_show ));
+    my $context_start= $diff_start > $max_context_len ? $diff_start - $max_context_len : 0;
+
+    if ($context_start < $diff_start) {
+        $q1 = Data::Dumper::qquote(substr($v1,$context_start, $diff_start - $context_start)) . " . " . $q1;
+        $q2 = Data::Dumper::qquote(substr($v2,$context_start, $diff_start - $context_start)) . " . " . $q2;
+    }
+
+    if ($context_start > 0) {
+        $q1 = "...$q1";
+        $q2 = "...$q2";
+    }
+    if ($length_to_show < $max_diff_len) {
+        $q1 .= " . " . Data::Dumper::qquote(substr($v1, $diff_start + $length_to_show, $max_diff_len - $length_to_show))
+            if $diff_start + $length_to_show < length($v1);
+        $q2 .= " . " . Data::Dumper::qquote(substr($v2, $diff_start + $length_to_show, $max_diff_len - $length_to_show))
+            if $diff_start + $length_to_show < length($v2);
+    }
+    if ( $diff_start + $max_diff_len <= length($v1) ) {
+        $q1 .= "..."
+    }
+    if ( $diff_start + $max_diff_len <= length($v2) ) {
+        $q2 .= "..."
+    }
+    my $pad= length($q1) > length($q2) ? length($q1) : length($q2);
+    my $lpad= length(length($v1)) > length(length($v2)) ? length(length($v1)) : length(length($v2));
+
+    my $issues= "";
+    $issues .="; utf8 mismatch" if $v1_is_utf8 != $v2_is_utf8;
+    $issues .="; length mismatch" if length($v1) != length($v2);
+
+    my $ret= sprintf(  "strings different\n"
+                     . "first string difference at octet offset %d%s\n"
+                     . " got-octets = %*s (octets: %*d, utf8-flag: %d)\n"
+                     . "want-octets = %*s (octets: %*d, utf8-flag: %d)\n"
+        ,$diff_start, $issues,
+        -$pad, $q1, $lpad, length($v1), $v1_is_utf8,
+        -$pad, $q2, $lpad, length($v2), $v2_is_utf8,
+    );
+    return $ret;
+}
+
+sub _deep_cmp {
+    my ($x, $y, $seenx, $seeny)= @_;
+    $seenx ||= {};
+    $seeny ||= {};
+    my $cmp;
+
+    $cmp= _test("defined mismatch",defined($x),defined($y))
+        and return $cmp;
+    defined($x)
+        or return "";
+    $cmp=  _test("seen scalar ", ++$seenx->{refaddr \$_[0]}, ++$seeny->{refaddr \$_[1]})
+        || _test("boolean mismatch",!!$x, !!$y)
+        || _test("isref mismatch",!!ref($x), !!ref($y))
+        and return $cmp;
+
+    if (ref $x) {
+        $cmp=  _test("seen ref", ++$seenx->{refaddr $x}, ++$seeny->{refaddr $y})
+            || _test("reftype mismatch",reftype($x), reftype($y))
+            || _test("class mismatch", !blessed($x), !blessed($y))
+            || _test("class different", blessed($x) || "", blessed($y) || "")
+            and return $cmp;
+        return "" if $x == $y
+                  or $seenx->{refaddr $x} > 1;
+
+        if (reftype($x) eq "HASH") {
+            $cmp= _test("keycount mismatch",0+keys(%$x),0+keys(%$y))
+                and return $cmp;
+            foreach my $key (keys %$x) {
+                return "key missing '$key'" unless exists $y->{$key};
+                $cmp= _deep_cmp($x->{$key},$y->{$key}, $seenx, $seeny)
+                    and return $cmp;
+            }
+        } elsif (reftype($x) eq "ARRAY") {
+            $cmp= _test("arraysize mismatch",0+@$x,0+@$y)
+                and return $cmp;
+            foreach my $idx (0..$#$x) {
+                $cmp= _deep_cmp($x->[$idx], $y->[$idx], $seenx, $seeny)
+                    and return $cmp;
+            }
+        } elsif (reftype($x) eq "SCALAR" or reftype($x) eq "REF") {
+            return _deep_cmp($$x, $$y, $seenx, $seeny);
+        } elsif (reftype($x) eq "REGEXP") {
+            $cmp= _test("regexp different","$x","$y")
+                and return $cmp;
+        } else {
+            die "Unknown reftype '",reftype($x)."'";
         }
+    } else {
+        $cmp= _cmp_str($x,$y)
+            and return $cmp;
+    }
+    return ""
+}
+
+sub deep_cmp {
+    my ($v1, $v2, $name)= @_;
+    my $diff= _deep_cmp($v1, $v2);
+    if ($diff) {
+        my ($reason,$diag)= split /\n/, $diff, 2;
+        fail("$name - $reason");
+        diag("$name - $diag") if $diag;
+        return;
     }
+    return 1;
 }
 
+
 sub run_roundtrip_tests_internal {
     my ($ename, $opt, $encode_decode_callbacks) = @_;
+    require Data::Dumper;
+
     my $decoder = Sereal::Decoder->new($opt);
     my $encoder = Sereal::Encoder->new($opt);
+    my %seen_name;
+
+    foreach my $rt (@RoundtripTests) {
+        my ($name, $data) = @$rt;
+
+        foreach my $meth (
+              ['object-oriented',
+                sub {$encoder->encode($_[0])},
+                sub {$decoder->decode($_[0])}],
+              ['functional simple',
+                sub {Sereal::Encoder::encode_sereal($_[0], $opt)},
+                sub {Sereal::Decoder::decode_sereal($_[0], $opt)}],
+              ['functional with object',
+                  sub {Sereal::Encoder::sereal_encode_with_object($encoder, $_[0])},
+                  sub {Sereal::Decoder::sereal_decode_with_object($decoder, $_[0])}],
+              ['header-body',
+                sub {$encoder->encode($_[0], 123456789)}, # header data is abitrary to stand out for debugging
+                sub {$decoder->decode($_[0])}],
+              ['header-only',
+                sub {$encoder->encode(987654321, $_[0])}, # body data is abitrary to stand out for debugging
+                sub {$decoder->decode_only_header($_[0])}],
+        ) {
+            my ($mname, $enc, $dec) = @$meth;
+
+            next if $mname =~ /header/ and $opt->{use_protocol_v1};
 
-    foreach my $meth (
-                      ['functional',
-                        sub {Sereal::Encoder::encode_sereal(shift, $opt)},
-                        sub {Sereal::Decoder::decode_sereal(shift, $opt)}],
-                      ['object-oriented',
-                        sub {$encoder->encode(shift)},
-                        sub {$decoder->decode(shift)}],
-                      ['header-body',
-                        sub {$encoder->encode(shift, 123456789)}, # header data is abitrary to stand out for debugging
-                        sub {$decoder->decode(shift)}],
-                      ['header-only',
-                        sub {$encoder->encode(987654321, shift)}, # body data is abitrary to stand out for debugging
-                        sub {$decoder->decode_only_header(shift)}],
-                      )
-    {
-        my ($mname, $enc, $dec) = @$meth;
-        next if $mname =~ /header/ and $opt->{use_protocol_v1};
-
-        foreach my $rt (@RoundtripTests) {
-            my ($name, $data) = @$rt;
             my $encoded;
             eval {$encoded = $enc->($data); 1}
                 or do {
                     my $err = $@ || 'Zombie error';
                     diag("Got error while encoding: $err");
                 };
-            ok(defined $encoded, "$name ($ename, $mname, encoded defined)")
+
+            defined($encoded)
                 or do {
+                    fail("$name ($ename, $mname, encoded defined)");
                     debug_checks(\$data, \$encoded, undef);
-                    next;
+                    next; #test
                 };
+
             my $decoded;
             eval {$decoded = $dec->($encoded); 1}
                 or do {
                     my $err = $@ || 'Zombie error';
                     diag("Got error while decoding: $err");
                 };
-            ok( defined($decoded) == defined($data), "$name ($ename, $mname, decoded definedness)")
+
+            defined($decoded) == defined($data)
                 or do {
+                    fail("$name ($ename, $mname, decoded definedness)");
                     debug_checks(\$data, \$encoded, undef);
-                    next;
+                    next; #test
                 };
 
             # Second roundtrip
@@ -768,10 +1026,12 @@ sub run_roundtrip_tests_internal {
                     my $err = $@ || 'Zombie error';
                     diag("Got error while encoding the second time: $err");
                 };
-            ok(defined $encoded2, "$name ($ename, $mname, encoded2 defined)")
+
+            defined $encoded2
                 or do {
+                    fail("$name ($ename, $mname, encoded2 defined)");
                     debug_checks(\$data, \$encoded, \$decoded);
-                    next;
+                    next; #test
                 };
 
             my $decoded2;
@@ -781,42 +1041,85 @@ sub run_roundtrip_tests_internal {
                     diag("Got error while encoding the second time: $err");
                 };
 
-            ok(defined($decoded2) == defined($data), "$name ($ename, $mname, decoded2 defined)")
-                or next;
-            is_deeply($decoded, $data, "$name ($ename, $mname, decoded vs data)")
+            defined($decoded2) == defined($data)
+                or do {
+                    fail("$name ($ename, $mname, decoded2 defined)");
+                    next; #test
+                };
+
+            # Third roundtrip
+            my $encoded3;
+            eval {$encoded3 = $enc->($decoded2); 1}
+                or do {
+                    my $err = $@ || 'Zombie error';
+                    diag("Got error while encoding the third time: $err");
+                };
+
+            defined $encoded3
                 or do {
-                    debug_checks(\$data, \$encoded2, \$decoded2);
+                    fail("$name ($ename, $mname, encoded3 defined)");
+                    debug_checks(\$data, \$encoded, \$decoded);
+                    next; #test
                 };
-            is_deeply($decoded2, $data, "$name ($ename, $mname, decoded2 vs data)")
+
+            my $decoded3;
+            eval {$decoded3 = $dec->($encoded3); 1}
                 or do {
-                    debug_checks(\$data, \$encoded2, \$decoded2);
+                    my $err = $@ || 'Zombie error';
+                    diag("Got error while encoding the third time: $err");
                 };
-            is_deeply($decoded, $decoded2, "$name ($ename, $mname, decoded vs decoded2)")
+
+            defined($decoded3) == defined($data)
                 or do {
-                    debug_checks(\$data, \$encoded2, \$decoded2);
+                    fail("$name ($ename, $mname, decoded3 defined)");
+                    next; #test
+                };
+
+            deep_cmp($decoded, $data,       "$name ($ename, $mname, decoded vs data)") or next; #test
+            deep_cmp($decoded2, $data,      "$name ($ename, $mname, decoded2 vs data)") or next; #test
+            deep_cmp($decoded2, $decoded,   "$name ($ename, $mname, decoded2 vs decoded)") or next; #test
+
+            deep_cmp($decoded3, $data,      "$name ($ename, $mname, decoded3 vs data)") or next; #test
+            deep_cmp($decoded3, $decoded,   "$name ($ename, $mname, decoded3 vs decoded)") or next; #test
+            deep_cmp($decoded3, $decoded2,  "$name ($ename, $mname, decoded3 vs decoded2)") or next; #test
+
+            if ( $ename =~ /canon/ ) {
+                deep_cmp($encoded2, $encoded,  "$name ($ename, $mname, encoded2 vs encoded)") or do {
+                    diag Dumper($encoded2);
+                    diag Dumper($encoded);
+                    next; #test
                 };
+                deep_cmp($encoded3, $encoded2, "$name ($ename, $mname, encoded3 vs encoded2)") or next; #test
+                deep_cmp($encoded3, $encoded,  "$name ($ename, $mname, encoded3 vs encoded)") or next; #test
 
-            if (0) {
-                # It isnt really safe to test this way right now. The exact output
-                # of two runs of Sereal is not guaranteed to be the same due to the effect of
-                # refcounts. We could disable ARRAYREF/HASHREF as an option,
-                # and then skip these tests. We should probably do that just to test
-                # that we can handle both representations properly at all times.
-                my $ret;
-                if ($name=~/complex/) {
-                    SKIP: {
-                        skip "Encoded string length tests for complex hashes and compression depends on hash key ordering", 1 if $opt->{snappy};
-                        $ret = is(length($encoded2), length($encoded),"$name ($ename, $mname, length encoded2 vs length encoded)");
+                if ($ENV{SEREAL_TEST_SAVE_OUTPUT} and $mname eq 'object-oriented') {
+                    use File::Path;
+                    my $combined_name= "$ename - $name";
+                    if (!$seen_name{$combined_name}) {
+                        my @clean= ($ename, $name);
+                        s/[^\w.-]+/_/g, s/__+/_/g for @clean;
+                        my $cleaned= join "/", @clean;
+                        my $dir= $0;
+                        $dir=~s!/[^/]+\z!/data/$clean[0]!;
+                        mkpath $dir unless -d $dir;
+                        my $base= "$dir/$clean[1].enc";
+                        $seen_name{$combined_name}= $base;
+                        for my $f ( [ "", $encoded ], $encoded ne $encoded2 ? [ "2", $encoded2 ] : ()) {
+                            my $file= $base . $f->[0];
+                            next if -e $file;
+                            open my $fh, ">", $file
+                                or die "Can't open '$file' for writing: $!";
+                            binmode($fh);
+                            print $fh $f->[1];
+                            close $fh;
+                        }
+                        diag "Wrote sample files for '$combined_name' to $base";
                     }
-                } else {
-                    $ret = is_string($encoded2, $encoded, "$name ($ename, $mname, encoded2 vs encoded)");
                 }
-                $ret or do {
-                    debug_checks(\$data, \$encoded, \$decoded);
-                };
             }
-        }
-    } # end serialization method iteration
+            pass("$name ($ename, $mname)");
+        } # end method type
+    } # end test type
 }
 
 
@@ -831,6 +1134,7 @@ sub _write_file {
 }
 
 # For bootstrapping other language implementations' tests
+our $COMPRESS;
 sub write_test_files {
     my $dir = shift;
     require File::Path;
@@ -847,7 +1151,10 @@ sub write_test_files {
         _write_file($make_name_file_name->($testno), $t->[2] . "\n");
     }
 
-    my $encoder = Sereal::Encoder->new;
+    my $encoder = Sereal::Encoder->new({
+        protocol_version => $PROTO_VERSION,
+        compress => $COMPRESS || Sereal::Encoder::SRL_UNCOMPRESSED(),
+    });
     foreach my $i (0..$#RoundtripTests) {
         my $testno = @BasicTests + $i + 1;
         my $t = $RoundtripTests[$i];
@@ -1,5 +1,3 @@
-# from "perlobject.map"  Dean Roehrich, version 19960302
-
 # O_OBJECT	-> link an opaque C or C++ object to a blessed Perl object.
 srl_encoder_t * O_OBJECT
 srl_decoder_t * O_OBJECT
@@ -17,7 +15,7 @@ INPUT
 
 O_OBJECT
 	if( sv_isobject($arg) && (SvTYPE(SvRV($arg)) == SVt_PVMG) )
-		$var = ($type)SvIV((SV*)SvRV( $arg ));
+		$var = INT2PTR($type, SvIV((SV*)SvRV( $arg )));
 	else{
 		warn( \"${Package}::$func_name() -- $var is not a blessed SV reference\" );
 		XSRETURN_UNDEF;