Revision history for Perl extension AnyEvent.

TODO: document TCP_*** constants
TODO: add debug/trace logging to some modules
TODO: use the eval-with-cb-call trick also to speed up JSON?
TODO: FAQ: common pitfalls? invoke-before-return
TODO: FAQ: scope-ids? hosts file order?
TODO: AnyEvent::Handle with AnyEvent::IO?
TODO: invalid. and localhost. specialcasing inside AEDNS and not AESocket (rfc6761)
TODO: maybe implement env variable to give hosts precedence
TODO: more formats for parse_ipv6 (single ipv6 address without port, ...p80, ...#80)

TODO: inet_aton maybe pack C4? no, add a comment why not
TODO: hosts read multiple times for initial concurrent tcp_connect's?

7.12 Wed Jan 27 19:12:26 CET 2016
	- use common name as hostname for tls connects, if Net::SSLeay
          supports SNI.
        - fix documentation of tls_autostart read type in AnyEvent::Handle,
          analyzed by Felix Ostmann.

7.11 Thu Jul 16 14:36:00 CEST 2015
	- AnyEvent::Socket::parse_ipv6 could accept malformed ipv6
          addresses (extra "::" at end and similar cases).
	- add a more explicit warning to AnyEvent::Handle that it doesn't
          work on files, people keep getting confused.
        - new function AnyEvent::Socket::tcp_bind.
        - new functions AnyEvent::fh_block and AnyEvent::fh_unblock.
	- aligned ipv6 address formatting with RFC 5952 (by not shortening
          a single :0: to ::).
        - added stability canary support.

7.09 Sat May  2 16:38:53 CEST 2015
	- AnyEvent::Debug called an internal function (AnyEvent::Log::ft)
          that was renamed to AnyEvent::Log:format_time. uinder its old name
          (analyzed by Michael Stovenour).
        - update AnyEvent::DNS fallback resolver addresses:
          seems google effectively killed most other free dns resolvers,
          so remove them, but add cable and wireless (ecrc) since it was
          stable for 20 years or so, official or not, and there should be
          an alternative to google.
        - perl5porters broke windows error codes in 5.20, and mapped
          WSAEWOULDBLOCK on the (different) EWOULDBLOCK error code, and
          WSAEINPROGRESS into the incompatible ERINPROGRESS code, probably
          because they were so cool. They probably broke other error codes
          for no reason, too, but I didn't care for checking, it's too
          depressing. This version only works around the WSAEWOULDBLOCK
          issue, because I don't have a nice way to work around the
          WSAEINPROGRESS bug.

7.08 Wed Dec 10 05:27:17 CET 2014
        - work around a newly introduced bug in Socket 2.011 (an
          errornous sun_length check) (analyzed by Maxime Soulé).
	- AnyEvent::TLS didn't load (but refer to) AnyEvent::Socket
          (analyzed by Ben Magistro).
        - AnyEvent::Strict will now confess, not croak. This is in line with
          it being a development/debugging tool.
        - work around a number of libglib bugs (debug builds of libglib enforce
          certain undocumented behaviour patterns such as not being able to
          remove a child watch source after it has fired, which we will try
          to emulate to avoid "criticals". what where they thinking...).
        - mention json security issues in AnyEvent::Handle, now that Douglas
          Crockford has foolishly and incompatibly changed JSON.
        - changed default dns resolver "max_outstanding" value from 1 to 10,
          the latter beinfg the intended value all along
          (reported by Ilya Chesnokov).
        - added new "AnyEvent::Impl::UV" interface module to the UV event lib
          (written by Mike Lowell).

7.07 Tue Dec 17 17:45:02 CET 2013
	- the documentation for custom tls verify schemes was wrong. make it agree
          with the code (reported by Maxime Soulé).
	- added cbor read and write types to AnyEvent::Handle (using CBOR::XS).
        - work around an API change in openssl that could cause wrong tls connection
          aborts, likely on windows only (analyzed by sten).
        - calling AnyEvent->now_update with AnyEvent::Impl::Perl caused an
          endless loop (reported by Dietrich Rebmann).
        - add tlsv1_1 and tlsv1_2 protocols to AnyEvent::TLS
          (patch by Maxime Soulé).
        - document AnyEvent::Impl::IOAsync::set_loop and
          $AnyEvent::Impl::IOAsync::LOOP. Though only documented now, this
          functionality has _always_ been available.
	- force a toplevel domain name in t/81_hosts.t (analyzed by
          David Jack Wange Olrik).
        - document that AnyEvent::Log uses AnyEvent::IO.
        - warn about AnyEvent::Filesys::Notify performance.
        - praise the joys of AnyEvent::Fork::*.
        - time for an =encoding directive.
        - do no longer use JSON to create a default json coder, use
          JSON::XS or JSON::PP directly.

7.05 Wed Aug 21 10:38:08 CEST 2013
	- uts46data.pl couldn't be found due to wrong naming of the file
          (reported by Fulko Hew).
        - handle lone \015's properly in AE::Handle's default line read
          (reported by various people).
        - untaint ip addresses found in /etc/hosts (patch by José Micó).
        - the memleak fix in 7.03 caused resolving via /etc/hosts to always
          fail on first use (reported and testcase by Andrew Whatson).
        - expose AnyEvent::Log::format_time, and allow users to redefine it.
        - expose AnyEvent::Log::default_format, and allow redefinition.
        - expose AnyEvent::Log::fatal_exit, to allow redefinition.
	- AnyEvent::Debug shell can now run coro shell commands, if available.
	- t/63* tests were wrongly in MANIFEST.
        - kernel.org's finger server went MIA, switch to freebsd.org and
          icculus.org.
        - clarify that IO::AIO and AnyEvent::AIO are needed for AnyEvent::IO
          to function asynchronously (reported by Daniel Carosone).
        - hard-disable $^W in most tests, it generates too much garbage output.
        - use a (hopefully) more future-proof method to emulate common::sense.
        - upgrade to UTS-46:6.2.0.
        - switch to INSTLIB from INSTLIBDIR, as INSTLIBDIR was wrongly
          documented. should not affect anything.

7.04 Thu Nov 15 02:23:58 CET 2012
	- debugging code left in.

7.03 Thu Nov 15 02:16:11 CET 2012
	- AnyEvent::Socket::inet_aton did not work when dns resolution was
          used to find the addresses (analyzed and patch by Maxime Soulé).
        - fix a memory leak in the /etc/hosts lookup code when hosts don't
          resolve and are not in hosts (analyzed by sten).

7.02 Tue Aug 14 04:27:58 CEST 2012
	- AnyEvent::Util::run_cmd could block indefinitely (analyzed and test
          program by Yorhel).
	- verified that AnyEvent::Socket follows RFC5952.
        - try to parse "ADDR#PORT" in addition to "ADDR PORT".

7.01 Sun May 13 01:03:17 CEST 2012
	- fail with EPROTO in AnyEvent::Handle wqhen TLS is requested but
          not available, instead of throwing an exception.
	- use File::Spec to get the tmpdir in t/*, to avoid needless
          failures on (most, not mine :) windows boxes.
        - new handle read types: tls_detect and tls_autostart.

7.0  Fri Apr 13 06:33:30 CEST 2012
        - child watchers are broken in POE 1.352 (also many earlier
          versions) and there seems to be no way to work around it, as POE
          itself is inherently racy. Document this regression and add a delay
          in t/68_poe_03_child.t for the time being.
	- new module AnyEvent::IO, that is a frontend to either a pure-perl
          synchronous I/O implementation (AnyEvent::IO::Perl), or to the
          asynchronous I/O extension IO::AIO (AnyEvent::IO::IOAIO).
        - load /etc/hosts only when DNS has no answer.
        - stat /etc/hosts on every access and reload it if it changed.
        - load /etc/hosts via AnyEvent::IO - potentially asynchronous.
        - fix a buggy croak in the dh parameter reading in AnyEvent::TLS.
        - AnyEvent::Log log_to_file and log_to_path now use AnyEvent::IO. As
          a side effect, they now use true appending as opposed to libc
          appending, and the file might not have been opened when the function
          returns.
        - the default logging level was not properly documented in a variety of
          places, this has hopefully been rectified.
	- updated uts46data.pl for unicode 6.1.0.
        - made log messages generated by AnyEvent submodules not
          include the package name anymore, as it will be logged
          by default already.
        - upgrade to the trick used by common-sense 3.5 to work
          around extra warning torture/breakage under perl 5.15.x.
        - log messages by anyevent are now ucfirst, are usually full
          sentences and do no longer include the package name.
        - the storable read type would sometimes throw an exception instead
          of causing EBADMSG (analyzed by Maxime Soulé).
        - work around a bug in openssl 1.0.1 which enforces a minimum rsa
          keysize (reported by Paul Howarth).
	- documented Rocco fabricating statements about AnyEvent and me.

6.14 Tue Jan 31 20:00:24 CET 2012
	- AnyEvent::Impl::Tk was broken due to a mysterious "die" inside,
          probably an editing mistake (reported by Darin McBride).

6.13 Thu Jan 12 07:27:01 CET 2012
	- AnyEvent::Util::fork_call checked for POSIX availability in the wrong way,
          causing it to fail if POSIX wasn't loaded already (analysed by Rock Power).
	- AnyEvent::Handle::push_read (line => did pass $1 directly, so
          regex-matching inside the callback would change the parameter, despite
          $1 being dynamically scoped per-block. this perl bug is now being
          worked around (testcase by Cindy Wang).

6.12 Mon Dec 12 13:21:10 CET 2011
	- $! was clobbered by subsequent calls in tcp_connect, due to
          the postpone not saving/restoring it (analyzed by Richard Garnish).

6.11 Tue Nov 22 10:36:05 CET 2011
	- Tk cannot create windows when tainted unless you set a title, so
          set a dummy title for AnyEvent's dummy mainwindow. How dummy.
	- escape any nonprintable/nonascii characters when stringifying
          backtraces.
        - log the reason loading an interface module fails to load at level debug.

6.1  Tue Oct  4 19:44:30 CEST 2011
	- INCOMPATIBLE CHANGE: the default log level is now "4" (error
          and above), and some messages inside AnyEvent have been elevated
          to higher log levels to print by default.
        - AnyEvent::log, unlike AnyEvent::Log::log, did not always exit on fatal.
        - work around yet another signal race bug, newly introduced by POE. My
          how that module sucks.
        - add google public ipv6 dns as fallback servers, so AE::DNS can finally
          support ipv6-only operation even with defaults (also use ::1 instead
          of 127.0.0.1).
	- give better guidelines on how to chose a logging level.
        - AnyEvent::Log can now cap (limit) log message priority by context.
        - AE::log will now save and restore $! and $@, so callers don't have to.
        - AE::Strict now checks that registered fds are still valid at regular
          intervals.
        - skip frequently called 'require' calls if the module appears loaded.
	- increase signal latency test timer considerably for poor little osx,
          which can't be bothered with basic posix functionality such as signal
          delivery.
        - mention some debugging aids in the tutorial.
        - new env variable PERL_ANYEVENT_MAX_SIGNAL_LATENCY.
        - private (%namespace) log contexts in PERL_ANYEVENT_LOG now have the name
          of the context as title.
        - AnyEvent::Strict would not restore read-write access to $_ when a callback
          died - this is a user bug, but since AE::Strict exists to help dbeugging...
	- detect the extremely evil IO::Async::Loop::AnyEvent module and refuse
          to work with it - the author is well aware of the issues but refuses
          to fix them. This extreme step has been taken because the module actively
          attacks the goals of AnyEvent and makes it harder for module authors to use
          AnyEvent.

6.02  Fri Aug 26 20:08:31 CEST 2011
        - log_to_syslog now accepts facility strings.
	- log_to_syslog would not allow facilities - "$facility|$level"
          *literally* means that string. sheesh.
        - the CPAN parser got confused and thought we provide Sys::Syslog, try
          to work around.
        - renamed AnyEvent::Impl::FLTK2 to FLTK.

6.01  Fri Aug 26 07:04:11 CEST 2011
        - INCOMPATIBLE CHANGE: AnyEvent::DNS resource records now include
          the TTL as fourth element - this affects ->resolve, ->request
          and ->dns_unpack indirectly. Convenience functions such as
          AnyEvent::DNS::{a,txt,mx,srv} are not affected, neither is
          AnyEvent::Socket.
        - INCOMPATIBLE CHANGE: backend authors now should not implement
          one_event or loop, but instead the AnyEvent::CondVar::_wait and _poll
          methods.
        - INCOMPATIBLE CHANGE: verbosity levels have been redefined, see
          AnyEvent (PERL_ANYEVENT_VERBOSE) and AnyEvent::Log for details.
        - DEPRECIATION: please change
          "use AnyEvent::Impl::Perl" to "use AnyEvent::Loop" and
          "AnyEvent::Impl::Perl::loop" to "AnyEvent::Loop::run"
          in your programs, for future compatibility.
        - *blush* idle watcher emulation was borked.
        - AnyEvent::Strict now checks the AE::xxx functions as well.
        - querying too long domain names would barf deep inside AnyEvent::DNS - since
          this is hard to avoid for applications, AnyEvent:.DNS::request now checks for
          for too long domain names explicitly and passes an undef to the callback
          instead (reported by James Bromberger).
        - make AnyEvent::Util more robust against EINTR (analyzed by Andrew Suffield).
        - new functions: AnyEvent::postpone, AnyEvent::log.
        - new module: AnyEvent::Log, for simple logging needs.
        - new env variable, PERL_ANYEVENT_LOG, for fine-grained logging config.
        - new env variable, PERL_ANYEVENT_DEBUG_SHELL, to automatically start
          a debug shell.
        - new env variable, PERL_ANYEVENT_DEBUG_WRAP, to automatically enable
          watcher instrumentation/wrapping (see AnyEvent::Debug::wrap).
        - new env variable, PERL_ANYEVENT_HOSTS, to override the place where
          /etc/hosts can be found.
        - an empty PERL_ANYEVENT_RESOLV_CONF now selects the default
          configuration.
        - env variables can now be specified using AE_xyz instead of
          PERL_ANYEVENT_xyz, subject to some rules.
        - lots of new AnyEvent::Debug functionality.
        - AnyEvent::Strict now checks for modifications of $_ by the callback
          (by setting it to readonly while executing the callback, if possible).
        - IO::Async sometimes cancels the wrong timer when you call cancel_timer.
          Invest extra overhead to work around this peculiar behaviour.
        - split the pure perl backend into a loop module and an impl
          module (AnyEvent::Loop and AnyEvent::Impl::Perl).
        - some read types (regex, netstring, json, storable) could
          errornously keep an AE::Handle object alive on their own
          (reported by Mohammad Toossi).
        - AnyEvent::Socket::resolve_sockaddr and all functions using it now
          supports /etc/hosts overrides.
        - add more workarounds around child watcher bugs in IO::Async
          and (newly introduced) in POE.
        - use glib child watchers and try to work around its limitations,
          if possible (based on a patch by Kevin Ryde).
	- add support for FLTK (fltk2), via AnyEvent::Impl::FLTK2.
        - many read types in AnyEvent::Handle would malfunction during
          transitions to TLS because of a "delete $self->{rbuf}".
          (reported by Gerald Galster).
        - improve condvar blocking wait performance (probably only noticable
          with faster backends), also make it easier for event loops that
          do not support blocking waits.
        - PERL_ANYEVENT_MODEL now also allows full module names, although
          the usefulness of this feature is not yet clear.
        - add tests for all supported event loops, conditional on PERL_ANYEVENT_LOOP_TESTS.
        - add t/80_ssltext.t.
        - no longer use AUTOLOAD for the initial loading (for no very good reason).
        - tcp_server now tries to unlink unix domain sockets when it gets destroyed.
        - do not run event loop detection multiple times when programs call their cached
          copies of AnyEvent::detect.
        - parse_hostport now accepts absolute paths als unix domain sockets.
        - suppress the idiotic warning inside POE by patching the POE::Kernel
          run flags instead of asking the user to jump through contortions.
        - add Task::Weaken as dependency (as recommends only at this time).
        - replace PF_UNSPEC by 0, for good old minix.

6.0 (not officially released, but was on CPAN for about 15 minutes :/)

5.34  Thu May 12 09:59:41 CEST 2011
	- AE::TLS didn't support both cert_file and key at the same time,
          due to some copy&paste bug (reported and analyzed by vti).
	- improve AE::Handle timeout documentation slightly.

5.33  Mon Feb  7 21:51:41 CET 2011
	- never appeared on CPAN, for unknown reasons.
	- AnyEvent::Handle on_connect_error - do not try to destroy a handle
          object when it's already gone (analyzed by Ryan Bullock).
	- update idna algorithm(s) to UTS#46 version 6.0.0.
        - preserve trailing dot in idn_to_ascii - testsuite says to preserve,
          uts46 says it's an error, and testsuite makes more sense.
        - remove leading dots in idn_to_ascii - testsuite says to remove,
          uts46 says it's an error, and testsuite makes more sense.
        - fixed a bug in idn ignored character processing.
        - passes the uts46 6.0.0 testsuite except for three cases which
          apparently are bugs in the testsuite itself.
        - passes the uts46 6.0.1 testsuite except for eleven cases which
          apparently are bugs in the testsuite itself.

5.31  Sun Jan 23 11:49:19 CET 2011
	- work around a POE bug where POE would first poll for new events and
          invoke handlers BEFORE handling the message queue.
	- fix parsing of "domain" statement in resolv.conf (Johannes Plunien).
	- stop_read now does just that in TLS mode. see its
          documentation for caveats.
	- update to common::sense 3.4 warning set (some warnings in perl impose
          an arbitrary 32kb limit on regex matches, which affects
          AnyEvent::Handle).
        - AE::Handle had a broken rbuf_max method and missing wbuf_max method.
        - new experimental AE::Handle::resettls method.

5.3    Fri Dec 31 05:48:13 CET 2010
	- major IOAsync patch by Paul Evans - 0.33 is required, as the major
          issues with IO::Async have been solved in that version, including
          the availabilty of a default loop!
	- implement new wbuf_max parameter for AnyEvent::Handle.
        - added Cocoa::EventLoop backend by Daisuke Murase.
        - the default testsuite now honors PERL_ANYEVENT_MODEL (Paul Evans).
        - clarify handle on_prepare documentation.
	- set CLOEXEC flag on parent-side fd's in AnyEvent::Util::run_cmd,
          to avoid leaking them to any child processes.
        - try to force files into the arch-dependend lib directory, for the
          half-"turly obsessive compulsive person"(s) that depend on it.

5.29   Sun Dec  5 10:49:21 CET 2010
        - convert EV backend to EV 4.00 API (so better upgrade EV too).
        - AE::Handle: implement dynamic read_size adjustment, add
          max_read_size parameter, reduce default read_size to 2048.
        - add some FAQ entries for servers and/or on_eof vs. on_error.
	- work around OS bugs (cygwin again) when lingering in AE::Handle.
        - reduce memory usage slightly when connect was used in AE::Handle.
        - wrok around more segfaults in Tk.
        - document tls_ctx => undef as valid.
        - detect WNOHANG value at compiletime.
        - include some probably linux-only support for building AnyEvent
          as part of the perl core.
        - improve compatibility of t/08_idna.t to perl 5.8.
        - make AnyEvent::Strict truly optional (in case somebody builds a
          minimal perl...).

5.28   Wed Oct 13 04:14:23 CEST 2010
        - due to a glitch, AnyEvent's internal getprotobyname wasn't
          actually used in tcp_server.
        - implement AnyEvent::CondVar->new method, which was documented but
          missing.
        - support autoloading of read types in unshift_read.
        - AnyEvent::Strict: do the test for invalid fh arguments differently,
          resulting in more informative error messages.
        - new AnyEvent::FAQ manpage.
	- clarify that you can actually call push_read etc. in on_read.
        - put keys and certs in /tmp, not into the current directory,
          and document this in the SECURITY CONSIDERATIONS sections
          of AnyEvent::TLS.
        - support ";" as resolv.conf comment character in AnyEvent::DNS, as
          requested by Ron Isaacson.
        - document $AnyEvent::DNS::RESOLVER variable.
        - incorporated major typo patches by Abhijit Menon-Sen.
        - AnyEvent::Handle now croaks when negative timeouts are passed.
        - add a shitload of TCP_xxx sockopt name constants to AnyEvent::Util,
          undocumented, but who knows when they might come in handy.

5.271  Tue Jun  8 12:05:46 CEST 2010
	- backport to perl 5.8.x.

5.27   Sun Jun  6 12:12:05 CEST 2010
	- postpone differently in AnyEvent::Socket now, as
          when not, canceling the connection attempt might fail
          (found by Felix Antonius Wilhelm Ostmann).
	- explicitly check for non-stream sockets in AE::Handle, too many
          clueless people fell into the trap of this somehow working.
	- simplified and reworked the "OTHER MODULES" section.
        - better/more condvar examples.

5.261 Wed Apr 28 16:13:36 CEST 2010
	- AF_INET6 was not properly used from Socket6 during configuration
          time (found by Andy Grundman).

5.26  Mon Apr 12 04:49:35 CEST 2010
	- don't generate spurious readyness notifications when select
          returns EINTR in the pure perl backend (analysed and testcase
          by Bas Denissen).
        - give same examples in SYNOPSIS sections of AnyEvent and AE.
        - provider faster implementations for AE::time/now/now_update for
          EV and Perl backends.
        - provide faster AE::cv and AE::time implementations for all
          backends.
        - no longer support lower/mixed-case signal names (this was
          never documented nor universally supported).
        - some more memory µ-optimisations, and somewhat less messy
          code to cope with both AnyEvent and AE APIs.

5.251 Sat Mar 13 00:58:21 CET 2010
        - make test failed if it was run before make install on perl <=
          5.8.8, and older versions of perl make our live very difficult,
          so write constants.pl during Makefile.PL time.

5.25  Sat Mar 13 00:23:14 CET 2010
	- fix a race condition in AnyEvent::Handle that would cause
          a "bio_read: unsupported method" error (in _another_
          ssl connection) after user code threw an exception.
        - added AnyEvent::Handle->destroyed method.
	- speed up AnyEvent::Socket::format_address and ::format_ipv6.
        - the AnyEvent::Util::fh_nonblocking function worked only
          by lucky accident on win32.
        - smaller and faster AnyEvent::Util::fh_nonblocking.
        - when the (required!) Time::HiRes module is missing, AnyEvent
          did not fall back to built-in time properly.
        - do not load Fcntl at runtime, saving memory and loading time.
        - precompile a number of constants and use them instead of runtime
          detection and eval.
        - free detection code after detection and similar memory optimisations.
        - Perl backend timer interval best effort drift has been improved
          (same algorithm as EV).
        - update unicode idna mapping table.

5.24  Tue Jan  5 11:39:43 CET 2010
	- cygwin never reports errors from failed connects. 1.5 just gives
          you continous readyness and EAGAIN, 1.7 is even more broken and
          just hangs. work around both issues in a cygwin-specific hack.
	- improve idle watcher documentation slightly.

5.23  Sun Dec 20 23:48:00 CET 2009
	- support IDNs in resolve_sockaddr, and therefore in tcp_connect.
        - implement punycode_encode/decode, idn_nameprep,
          idn_to_ascii and idn_to_unicode operations in AnyEvent::Util.
	- provide $AE::VERSION.
        - removed traces of "no strict 'refs'".

5.22  Sat Dec  5 03:51:13 CET 2009
	- downgrade-or-fail in AnyEvent::Handle::push_write, to
          diagnose encoding failures earlier and more succinctly.
          (this works around bugs in perl, throwing away encoding info
          when passing scalar data to extensions).
        - add more examples to AnyEvent::Socket manpage.
        - upgrade internal warning set to the same as common::sense 2.03.
        - use pack "n/a*" for pre-5.8.9 perl compatibility in AnyEvent::DNS
          (John Beppu).
        - AnyEvent::Socket::inet_aton now properly supports ipv6, as documented.
        - add google public dns servers to fallback server set and make sure
          we load-balance properly between the three sets. also add all
          fallback dns servers, not just a random one, to each dns config.

5.21  Thu Nov 19 02:48:47 CET 2009
	- fix a problem where socket constants were called with parameters
          (spotted by David Friedland).
        - fork_call never use'd POSIX (reported by Daisuke Maki).
        - improve perl 5.6 compatibility further (but it still won't work
          unless you rip out everything but the core).
        - prefer Net::DNS::Resolver over ipconfig, if installed, on win32.
          uses 10MB of RAM, but doesn't open a console window. *sigh*.

5.202 Wed Oct 14 22:35:44 CEST 2009
	- AnyEvent::DNS would unexpectedly clobber $_ under windows
          (analysed by Matthias Waldorf).
        - AnyEvent::Handle::run_cmd can now pass the PID of the
          newly-created process, which is much less useful than it might
          sound (based on patch by Yann Kerherve).

5.201 Tue Sep 29 12:09:25 CEST 2009
	- AnyEvent:Handle::on_starttls/on_stoptls methods were broken
          (reported by Torsten Foertsch).
        - common::sense 2.0 could cause tcp_server to throw an exception
          (analysed by elmex).

5.2  Mon Sep 14 07:04:49 CEST 2009
	- INCOMPATIBLE CHANGE: do no longer support register_read_type
          and register_write_type in AnyEvent::Handle, instead support
          package names (the facility was mostly abused).
        - implement "packagename-as-read/write type" support in
          AnyEvent::Handle.
	- AnyEvent::Handle: new options "keepalive" and "oobinline".
        - oobinline set by default to avoid security issues.
        - the pure-perl event loop backend wrongly detected times() fallback
          support (spotted by Pavel Boldin).

5.12 Tue Sep  1 20:26:50 CEST 2009
	- be more lenient when parsing resolv.conf files, as some people
          use hashmarks as comment indicator inside directives (reported
          by Michael S. Fischer).
        - use same set of warnings as common::sense 2.0.
	- fix a potential 32 bit overflow issue due to perl having problems
          with large hex constants in the Perl backend.
        - do not use the slower 5.6 version fo the enc_name code in AnyEvent::DNS
          by default.
        - fix some prototypes.
	- (optionally) require Async::Interrupt 1.02.
        - replace opendns fallbacks by verizon/level3 public dns servers.

5.112 Fri Aug 21 13:59:04 CEST 2009
        - AnyEvent::Handle could cause hard crashes in Net::SSLeay when
          using starttls with an illegal/uncreatable tls context.
        - port to broken win32 perls.

5.111 Sun Aug 16 18:44:36 CEST 2009
	- perl errornously requires sockaddr_un structures returned by the
          kernel to have a specific length. work around this bug by padding
          sockaddr_un structures (found to fail on OS X, as perl only uses
          the correct code for GNU/Linux, but likely to fail on other
          OSes as well).

5.11 Wed Aug 12 17:49:37 CEST 2009
        - fixed a bug in run_cmd with input from filename.
        - create a json coder object when none is given in write direction
          to reduce dependency on JSON.

5.1  Tue Aug 11 03:17:32 CEST 2009
        - tcp_connect now reports the correct error in $!, not an unrelated
          one (was broken since 4.91).
        - AE::Handle did not properly clear rtimeout/wtimeout watchers
          when retrying a tcp_connect.
        - new functions: AnyEvent::Util::run_cmd and close_all_fds_except.
        - fixed wrong documentation in AE::Socket.

5.01 Mon Aug 10 03:16:32 CEST 2009
	- last release broke EV child watchers :/.

5.0  Sun Aug  9 17:32:09 CEST 2009
	- officially introduce and document the AE API.
        - lowering the timeout at runtime did not have immediate effect
          in AnyEvent::Handle.
        - AE::Handle now has separate and independent read and write timeout
          settings.
        - probe for Guard module when AnyEvent::Util is loaded, not at
          runtime (Event doesn't like eval "use" from callbacks, and it avoids
          exporting the probe function).
        - Impl::IOAsync's io watchers did not use the filehandle emulation
          code to work around it's watcher limits.
        - work around IO::Async's problems with overloaded objects.
        - try to disable wearnings inside IO::Async, it's just too buggy.
        - do no longer use Event by default, only EV or Perl, Event is too buggy.
	- hackishly provide fast AE:: interface when EV or Perl is the backend.
        - inherit VERSION from AnyEvent::VERSION in most modules.

4.91 Thu Aug  6 15:42:45 CEST 2009
	- AE::Handle::starttls could get out of sync when the read buffer
          already contains some TLS handshake.
	- AE::Handle did not properly free the TLS session, causing bigger
          memory leaks in Net::SSLeay than expected (reported by toaster).
        - AE::Socket::tcp_connect will now artificially delay invoking
          the callback to avoid returning after invoking the callback.
        - convert many internal watcher uses to AE API.

4.9  Sat Aug  1 11:07:01 CEST 2009
	- Glib silently fails when registering a timeout with
          a negative value, so avoid that.
        - call condvar callback immediately  when it is set
          after the condition is already signalled.
        - check rbuf_max condition only after trying to consume
          data.
	- normalise signal numbers to names when using the ae's
          signal handling, but do not document this (yet).
        - pure perl signal emulation did not properly set nonblocking
          mode on the signal pipe on win32 (this is unlikely to have caused
          any issues).
        - new module: AnyEvent::Debug.
        - AnyEvent::Strict now checks that a signal was specified
          by name (not name or number).
        - reduce memory footprint in typical cases by ~50kb by
          only compiling signal code when necessary.
        - add AnyEvent::Handle->rbuf_max.
        - grab the AE namespace for future new API, implement stubs for
          said future/alternative API.
        - new function: AnyEvent::Socket::format_hostport.

4.881 Tue Jul 28 12:51:53 CEST 2009
	- work around a bug in local in pre-5.10 perls, causing
          AnyEvent::Handle to recurse when it should not
          (analysed by elmex).

4.88 Tue Jul 28 04:04:37 CEST 2009
	- re-bless the handle into a dummy package after calling
          AnyEvent::Handle::destroy, so the user does not need to check
          for errors after every push_write etc.
        - do not attempt to run t/02_signals.t on obviously broken
          platforms.

4.87 Sun Jul 26 02:06:16 CEST 2009
	- do not attempt to linger when there is no longer a valid fh.
        - enforce tls mode to be either accept or connect, do not simply
          segfault in Net::SSLeay.
        - AnyEvent::Handle can now call tcp_connect itself (new parameters
          connect, on_prepare, on_connect and on_connect_error). Updated
          tutorial accordingly.
	- add AnyEvent::Impl::Irssi backend.

4.86 Mon Jul 20 23:52:29 CEST 2009
	- since the verbose warning is not enough, explicitly document
          that versions before 1.33 of Net::SSLeay are not secure.
	- work around signal handling races in Event and (...) Event::Lib.
        - try to align signal-race timer to full-second boundaries.
        - work around Tk not liking negative timeouts.
        - don't complain of different grades of environmental unfriendlyness
          in IO::Async.

4.85 Sat Jul 18 06:16:14 CEST 2009
        - nail the signal race problem in perl once and for all
          (see $AnyEvent::MAX_SIGNAL_LATENCY).
        - take advantage of Async::Interrupt if it is available.
        - load Time::HiRes and Guard modules on demand only.
        - add optional/recommended modules section to AnyEvent
          documentation.
        - reduce memory usage considerably (and reduce startup penalty)
          by not using "strict", "warnings" and "overload" modules.
        - work around buggy windows/openbsd perls and provide EBADMSG
          and EPROTO ourselves when missing.
	- improve perl 5.6 compatibility of the core event loop.
        - made Net::SSLeay version 1.33 a soft requirement.

4.83 Fri Jul 17 16:56:26 CEST 2009
	- implement AnyEvent::Socket::getprotobyname.
	- AnyEvent::CondVar's will now detect recursive blocking
          waits and will croak, as too many people fall into
          this trap.
        - AnyEvent::Handle will now call ->destroy on itself after
          executing the on_error callback, instead of doing some
          half-baked internal shutdown, for fatal errors.
        - clarify on_eof behaviour w.r.t. the read queue and
          on_read callbacks.
	- ignore some possible spurious wake-ups in tcp_connect.

4.82 Sat Jul 11 00:34:55 CEST 2009
	- POE and Event backends didn't accept some callable objects as 
          callbacks.
	- use Config module instead of POSIX module to detect signal names
          in AnyEvent::Strict and AnyEvent::Impl::EventLib, as the POSIX
          module doesn't even have all POSIX signals :/.
        - use more workarounds around the many refcnt/corruption bugs in
          Event::Lib.
        - work around a race condition in perl's select, causing t/03_child.t
          to rarely fail.

4.81 Thu Jul  9 10:30:30 CEST 2009
        - AnyEvent::Handle didn't properly diagnose write errors
          (it expected -1 from syswrite, how lame... :).
	- support file descriptors in addition to file handles
          in AnyEvent->io.
	- new env variables: PERL_ANYEVENT_RESOLV_CONF,
          PERL_ANYEVENT_MAX_OUTSTANDING_DNS, PERL_ANYEVENT_CA_FILE
          and PERL_ANYEVENT_CA_PATH.
        - provide a sensible synopsis section for AnyEvent::TLS.
        - add a "supported backends" section to the manpage.
        - added simple io watcher test to testsuite, using a
          portable_socketpair.
        - tried to improve the stability of the Event::Lib backend,
          YMMV.

4.8  Mon Jul  6 23:45:16 CEST 2009
	- AnyEvent::DNS did not properly follow CNAME records with
          uppercase targets.
        - AnyEvent::DNS would errornously return AAAA records
          with v4 mapped addresses (a faulty record) as ipv4 addresses,
          causing AnyEvent::Socket to throw an exception.
        - added new module AnyEvent::TLS for easier SSL/TLS context
          creation, with many options including hostname verification,
          secure default configuration, lots of documentation and,
          predefined diffie-hellman keys for perfect forward security
          and much more. get it while it's still fresh!
        - use AnyEvent::TLS in AnyEvent::Handle for context management.
	- load AnyEvent::Handle only on demand in AnyEvent::DNS,
          so AnyEvent::Socket users have smaller memory footprint
          in the common case.
        - add AnyEvent::Handle->push_shutdown method.
        - add an additional error message parameter to AnyEvent::Handle's
          on_error callback (for TLS, $! is still available).
        - add AnyEvent::Handle on_starttls/on_stoptls callbacks.
        - make AnyEvent::Handle more robust against early connection
          failures (during new), and return C<undef> in that case
          from the constructor.
        - AnyEvent::Handle will now try to load only JSON::XS first,
          then fall back to JSON.
        - format_ipv4/format_ipv6 are now exported by default, for symmetry,
          and because it was documented that way.

4.451 Fri Jul  3 00:28:58 CEST 2009
	- do not clear rbuf when shutting down an AnyEvent::Handle
          object - doing so breaks AnyEvent::HTTP.

4.45  Mon Jun 29 22:59:26 CEST 2009
        - a write error could cause AnyEvent::Handle to create
          an I/O watcher with an undefined $fh.
        - special-case mapped ipv4 addresses in both
          AnyEvent::Socket::format_address and parse_address,
          to treat them just like ipv4 addresses.
        - updated and overhauled the AnyEvent::Intro doc.
        - implement AnyEvent::Socket::format_ipv4/ipv6.
	- slightly speed up portable_pipe/socketpair.
        - expand condvar begin/end documentation.

4.42  Fri Jun 26 08:32:18 CEST 2009
	- preliminary and neccesarily incomplete support for IO::Async.
	- reset SIGCHLD to DEFAULT when AnyEvent is loaded, in
          case it was set to IGNORE, to ensure we can catch child statuses
          even when the calling env acts stupidly.
        - updated benchmarks with IO::Async, which performs very well.

4.412 Wed Jun 24 01:35:57 CEST 2009
	- support an "untaint" attribute for AnyEvent::DNS
          and set it on the default resolver.
	- implement and document AnyEvent::Impl::Perl::loop.
        - remove all anyevent-env variables from %ENV
          when running in tainted mode.
        - mention and extend the IO::Lambda benchmark.

4.411 Sun Jun  7 18:48:02 CEST 2009
	- do not try to use F_SETFD on windows.

4.41  Thu May 14 06:40:11 CEST 2009
        - work around issues in older perls (5.8.0?) when
          a signal handler is deleted from the %SIG hash.
        - use POSIX::_exit in child test, to avoid running
          destructors.
        - speed up CHLD handling by relying on SIGCHLD being
          synchronously delivered, even when we roll our own
          implementation.
	- AnyEvent::DNS: add the "dname" resource record name
          for cosmetic reasons.

4.4   Sun Apr 26 20:12:33 CEST 2009
	- implemented idle watchers, where applicable.
	- AnyEvent->time died when Event backend was in use.
        - fix a memleak in the Tk backend.
        - sped up Tk timer handling.
        - clip negative "after" values to 0 in AnyEvent::Impl::Event
          to avoid spamming out warning messages.
        - fix Qt timers without interval.
        - avoid Qt zero-timeout special case, in old libqt's.

4.352 Mon Apr 20 16:31:11 CEST 2009
	- fix AnyEvent::Strict error messages for child watchers.
        - fix/update Makefile.PL "recommends" versions.
        - add AnyEvent->now_update.

4.351 Sat Apr 11 07:56:14 CEST 2009
        - actually make the signal pipe work under win32.
        - localise $! in signal handler to avoid changing $!, although
          perl itself does not seem to save/restore errno either.
	- set the cloexec flag on the signal pipe (normally set by perl too).

4.35  Fri Mar 27 11:48:20 CET 2009
        - event models relying on AnyEvent's signal watcher emulation did
          invoke the callback asynchronously, contradicting documentation
          and causing signals to get lost (this includes AnyEvent's own
          event loop). AnyEvent now uses the standard pipe trick to make
          callback execution synchronous to the event loop.
        - AnyEvent::Handle didn't free TLS context data on DESTROY
          (patch by Pavel Shaydo).
        - work around the ever-incompatibly-changing API of MakeMaker.
        - document that changing global variables without restoring them
          is a bad idea in Perl (noted by Adam Rosenstein).
        - AnyEvent::Strict now barfs if ->io is passed a file.

4.34  Thu Feb 12 18:32:45 CET 2009
	- separately buffer TLS read data, as otherwise the read queue could
          deadlock as receiving data is not expected while draining the
          read queue (which cna only happen with TLS).
	- raise EBADMSG error on JSON decoding errors.
        - fix some minor manpage bugs (reported by Maximilian Gaß).
        - speed up select bitmask parsing quite a bit in the pure perl
          backend.
        - use CORE::select instead of just select in AnyEvent::Impl::Perl.

4.331 Tue Jan  6 21:07:25 CET 2009
	- socketpair fails on many vista machines because vista has
          completely broken accept/getpeername and getsockname functions,
          so we provide our own socketpair emulation that kind of works
          (AnyEvent::Util::portable_pipe).
        - new function: AnyEvent::Util::portable_socketpair.
	- take advantage of the Guard module if it exists.

4.33 Fri Nov 21 02:35:40 CET 2008
	- AnyEvent::Strict did errornously flag a fileno of 0 as
          illegal.
	- reduce memory usage and slightly speed up the pure perl backend
          by only storing the file descriptor, not the file handle.
        - add missing autocork method to AnyEvent::Handle (reported
          by Adam Rosenstein).
        - AnyEvent::DNS->resolve errornously documented an $rcode result
          argument, but there isn't (spotted by Henrik Krohns).
        - the naked truth about Tk - it's basically unmaintained.

4.32 Mon Nov  3 22:46:32 CET 2008
	- fix AnyEvent::Socket::resolve_sockaddr to properly support
          unix sockets again and choose a proper default for
          the socket type.
        - fix call to resolve_sockaddr in tcp_connect.

4.31 Thu Oct 30 04:41:48 CET 2008
	- implemented AnyEvent::Handle->destroy method.
	- hint about unexpected effects in TLS mode.
        - speed up AnyEvent::Impl::Perl by using more arrays
          instead of hashes without the slightest loss of readability :)
        - work around a perl argument refcounting bug.
        - with some perl patches applied, AnyEvent now seems leak-free.

4.3  Fri Oct  3 09:18:43 CEST 2008
	- AnyEvent will now install a no-op signal handler for SIGPIPE,
          unless one has been installed already.
        - warn about not loaidng AnyEvent::Impl::POE early enough
          (patch by Adam Rosenstein).
	- fixed a great number of bugs and corner cases in AnyEvent::Handle.
        - fix a bug in where in SSL connect mode, the
          client would first wait for some data by the server and
          otherwise hang. (reported and analysed in an absolutely
          exemplary manner by Adam Rosenstein).
        - fix a bug in where SSL EOF would not be treated
          as stream EOF, putting the connection into a hung state.
          (reported and analysed in an absolutely exemplary manner by Adam
          Rosenstein).
        - fix a potential segfault when the TLS context
          would go missing in a read callback (e.g. due to stoptls) -
          Net::SSLeay of course makes no type checking whatsoever.
        - AnyEvent::Handle will not stop the read watcher in TLS mode.
        - AnyEvent::Handle->stoptls will send a TLS close notify
          instead of simply destroying the stream now.
        - fix a bug where large blocks of data written in TLS mode
          would not be sent unless triggered by receives.
        - on_drain will now take the tls write buffer into account.
        - SSL operations have been streamlined a bit and should be faster
          now, more to come, though.
        - added a FAQ section, will document nontrivial issues in there.

4.234 Mon Sep 29 04:08:13 CEST 2008
	- fix child watcher documentation: only child exits will be reported,
          no trace events.
        - mention SIGPIPE in AnyEvent::Handle.
	- perl backend will now use POSIX::times as monotonic clock
          when available and the monotonic clock option is not.
        - run even without Time::HiRes in some cases now, but do not
          provide sub-second accuracy (all *supported* perl versions have
          Time::HiRes).

4.233 Fri Aug 22 01:48:05 CEST 2008
	- fix a bug in the testsuite, causing a hang.
        - clarified lots of AnyEvent::Handle constructor arguments.

4.232 Thu Aug 21 20:44:25 CEST 2008
	- fix a bug in AnyEvent::Handle that could cause two fatal error
          callbacks on EOF, which caused AnyEvent::HTTP to signal success twice
          (testcase by Leon Brocard).
        - clarified on_eof/on_error documentation a bit.

4.231 Tue Jul 29 13:12:15 CEST 2008
        - remove some debugging code left in AnyEvent::Util::fork_call
          (and no, it's impossible to implement with the broken windows
          perls without resource leaks or worse).

4.23 Tue Jul 29 12:19:59 CEST 2008
	- document the first parameter passed to condvar callbacks to be
          the callback.
        - add AnyEvent::Socket::{ntoa,aton} aliases.
        - optimize the AE::Handle->push_read (line) for the default
          eol marker.
        - optimize push_read (packstring|storable) for small packets.
        - invoke on_error callback when no on_eof callback is set.
        - fix a bug in push_read (storable) of unknown impact.

4.22 Sun Jul 20 16:34:13 CEST 2008
	- new function AnyEvent::Socket::parse_hostport.
        - as the bulkheads at microsoft can't even get getprotobyname reliably
          working on their shitty fucking broken os we need to hardcode
          some common protocol numbers in AnyEvent::Socket. How can
          people even bother with such a pile of shit as windows.

4.21 Thu Jul 17 14:40:05 CEST 2008
	- INCOMPATIBLE CHANGE: AnyEvent::Util::fork_call now has a prototype
          which is incompatible to previous invocation syntax.
        - work around more windows perl bugs in fork_call. I cannot imagine
          why anybody would bother implementing fork in such an obviously
          unusable way.
	- avoid calling $do_retry if the request has already finished in
          AnyEvent::DNS, thus avoiding an exception (reported by Anatoly
          K. Sharifulin).

4.2  Sat Jul 12 22:42:11 CEST 2008
	- support an "interval" argument for repeating timers.
        - fix a bug in the pure perl event loop that caused timers to
          be delayed under some circumstances.
        - implement optional argument checking via PERL_ANYEVENT_STRICT.
	- not importing anything from Event causes it to malfunction
          and not use Time::HiRes, work around this bug.
        - more bug workarounds against the endlessly broken Tk module,
          seems to work now on 32 bit machines, but less so on 64 bit.
        - minor optimisations applied to most backends.

4.161 Fri Jul  4 14:24:48 CEST 2008
        - fixed the t/handle/02_write.t, it read after write, with the
          assumption that all data could be written before read
          (reported by Jim Razmus).

4.160 Thu Jul  3 04:02:21 CEST 2008
	- re-registering signal watchers after unregistering one did
          not work if a backend falls back on AnyEvent's default
          implementation (also affected child watchers).
	- new AnyEvent::Handle options: autocork and no_delay.

4.152 Sun Jun 22 14:15:44 CEST 2008
	- allow for 32-bit perls that implement shifts differently
          on different architectures in parse_ipv4 (reported and
          analysed by Keiichi DAIBA).

4.151 Fri Jun  6 17:34:24 CEST 2008
	- make sure specifying _only_ on_read and never pushing reads
          works.

4.15 Fri Jun  6 13:00:46 CEST 2008
	- the pure perl backend would keep some watchers alive when more than
          one watcher was registered for the same fd.
	- new "packstring" and "storable" read and write types
          for AnyEvent::Handle.
        - allow on_eof handler to be called after on_error with EPIPE returns.
        - do not immediately call on_read callback in handle constructor.

4.14 Thu Jun  5 20:29:31 CEST 2008
        - Fixed a bug in DNS SRV priority sorting.
        - AnyEvent::Util::guard now reports runtime errors while
          executing the guard block as warnings.
        - handle 0-byte-reads just before EOF correctly in AnyEvent::Handle.

4.13 Thu Jun  5 00:47:59 CEST 2008
	- AnyEvent::DNS only followed cname chains with length 2,
          contrary to documentation. bumped it up to 10, thanks to
          microsoft, the current broken-dns-config-king.
        - AnyEvent::DNS didn't check for socket return status
          but instead relied on perl not creating filehandles in that
          case - too bad it gives you a bogus file handle (reported
          and analysed by Vladimir Timofeev).
	- fix queue management logic in AnyEvent::Handle:
          when on_read was registered, the queue was empty
          and no progress could be made AnyEvent::Handle would
          enter an endless loop.
        - correctly start reading again when the handle became
          busy again after idling.
        - correctly treat tls shut-downs as EOF condition
          (for the time being, would be nice to have a callback for that).
        - correctly call eof callback under all conditions
          (hopefully).
        - the timeout callback did not expect that $self can go
          away any time.

4.12 Tue Jun  3 10:58:04 CEST 2008
	- include AnyEvent::Intro, a tutorial for anyevent,
          anyevent::socket and anyevent::handle.
        - allow more options in on_error.

4.11 Fri May 30 23:42:25 CEST 2008
        - INCOMPATIBLE CHANGE: replace ptr by real PTR lookup, provide
          reverse_lookup and reverse_verify to replace it, support
          v4mapped and v4compat addresses.
	- provide more documentation for the resolver class.
        - really replace longest run of :0: by :: in format_address,
          also properly convert :: and ::1 again.
        - support NAPTR record name and decode it.
        - implement random weight sampling for SRV records, as per
          rfc 2782.
        - correctly abort on srv-record targets of ".".
        - added AnyEvent::DNS::wait_for_slot.
        - in the unlikely event of a virtual circuit connection
          being invalidated by a delayed udp reply, AnyEvent::DNS
          could die. this has been fixed.
        - plug a probable memleak in the DNS vc code.
        - use configured timeout also for VC request phase.
        - implement timeout and max_outstanding methods
          for resolver class.
        - update version numbers in all modules.

4.1  Thu May 29 05:45:40 CEST 2008
	- INCOMPATIBLE CHANGE: renamed xxx_ip functions to xxx_address
          which mirrors their purpose better (old names still available
          for a while). Also moved AnyEvent::DNS::addr to
          AnyEvent::Socket::resolve_sockaddr.
	- implement AnyEvent->time and AnyEvent->now.
        - fix IPv6 support in pack_sockaddr.
	- officially un-experimentalise all newly introduced code.
        - support unix domain sockets everywhere by specifying
          a host of "unix/" and the pathname as service.
        - implement an activity timeout in AnyEvent::Handle.
	- added a regex read type to AnyEvent::Handle.
	- added a json read/write type to AnyEvent::Handle.
        - always croak in on_error in AnyEvent::Handle.
        - document how to subclass AnyEvent::Handle.
        - implement AnyEvent::Util::fork_call.
        - add support for IPv6 nameservers and nameserver statements.
        - work around _yet_ _another_ windows perl bug in where empty
          select masks cause errors in the pure perl backend (this
          workaround cannot be implemented for the other event loops,
          but EV also works around this).
        - supply AnyEvent::Socket::address_type and make good use of it.
        - clarify and fix the Handle documentation w.r.t $self vs. $handle
          vs. callback arguments.
	- add some recommends to the META.yml.

4.05 Mon May 26 19:44:06 CEST 2008
        - some platforms ignore the rfc and prepend an extra sa_len member to
          the sockaddr structure, cater for those. also use sockaddr_family,
          didn't know it was there, but comes in handy for the workaround.
	- undo the import WIN32 hack.

4.04 Mon May 26 08:03:31 CEST 2008
	- try to work around yet another windows bug: failed connects are reported
          as if it were out-of-band data. windows users: you suck. Only
          EV, Glib, Event and the pure perl backend can handle this condition.
        - optimize write algorithm in AnyEvent::Handle.
	- properly parse PERL_ANYEVENT_PROTOCOLS and use the info correctly.
        - tcp_connect now properly iterates through all targets.
	- check wether IPv6 sockets can actually be created, otherwise, disable ipv6
          support entirely. also hardcode some AF_INET6 constants for the
          sake of perl 5.8.
        - vastly improved nameserver/suffix detection algorithm for windows
          (still a hack...)
        - try to use a fallback dns server if no dns servers could be found.
        - splatter around some local $SIG{__DIE__}'s to work around bad
          die handlers.

4.03 Sun May 25 05:05:57 CEST 2008
	- work around perl bugs on windows, where perl returns undocumented
          error codes for sysread, syswrite etc.
	- fix AnyEvent::Handle::unshift_read to really unshift and not push.
        - AnyEvent::Handle could sometimes stumble over 0-byte writes.
        - fix unpack_sockaddr, now works for AF_INET6 sockaddr's.
        - updated documentation.
        - lots of workarounds for perl 5.8 quirks.
        - implement a more flexible type system for both reads and writes
          in AnyEvent::Handle.
        - support IPv6 in tcp_server.
        - pass local hostname and port to the prepare callback in tcp_server.
        - make types extendable in AnyEvent::Handle.
        - croak, not die, in AnyEvent::Handle;

4.0  Sat May 24 19:58:08 CEST 2008
	- added (experimental) AnyEvent::DNS module.
        - added (experimental) AnyEvent::Socket module.
        - added (experiemntal) support for TLS/SSL in AnyEvent::Handle.
        - make condvar->end call ->send as documented.
        - make condvar constructor support the documented cb => argument.
        - overload condvars so you can use them instead of code references.
        - support more of the POSIX forms instead of a pure dotted quad for
          verbatim ip addresses.
        - AnyEvent::Socket::inet_aton will use AnyEvent::DNS now and not fork anymore.
	- try very hard not to freeze on broken windows perls.
        - make on_eof optional in AnyEvent::Handle.
	- support service names in tcp_connect and tcp_server.
        - work around netbsd/openbsd bug causing unix sockets not to
          become writable until empty.
	- only make a dummy read for error detection when getpeername
          returns ENOTCONN.
        - defend AnyEvent::Util::inet_aton against Coro::LWP's brutal
          override.
        - new AnyEvent::Guard::cancel.

3.5  Sat May 17 23:17:14 CEST 2008
	- gracefully deal with systems not defining CLOCK_MONOTONIC
          (as opposed to not supporting the clock itself, which worked in 3.41).
	- added AnyEvent::Util::tcp_connect and AnyEvent::Util::tcp_server
          helpers, to replace the ill-designed AnyEvent::Socket class.
	- added some examples (eg/connect, eg/handle).
	- added AnyEvent::Util::guard.
        - AnyEvent::Util now exports most of its functions.
	- fixed a bug in *_read_chunk, where the callback was called with $self as
          first and second argument.
        - simplified and fixed regex handling in AnyEvent::Handle::*_read_line.
	- fix manpage condvars examples to use new syntax.
        - considerably improve inet_aton with EV::ADNS backend to support
          cname chains (common dns configuration bug, e.g. with www.google.de)
          and aaaa records if no a record could be found.

3.41 Sun May 11 19:53:13 CEST 2008
	- work around a bug in perl 5.8.8's local. Reported by Yi Ma Mao.

3.4  Sun May 11 00:29:25 CEST 2008
	- complete re-implementation of condvars with a hopefully much more useful
          API (backward compatible functions exist but are undocumented).
	- AnyEvent::Handle fully rewritten with a hopefully more powerful API
          (still subject to change).
	- take advantage of CLOCK_MONOTONIC, if available, in AnyEvent::Impl::Perl.
        - provide AnyEvent::post_detect and @AnyEvent::post_detect, which
          allows module authors to avoid forcing event loop detection.
          (used by Coro::AIO, Coro::BDB, Coro::AnyEvent for example).
        - remove coro backends: Coro now provides generic support for AnyEvent
          via Coro::AnyEvent.

3.3  Mon Apr 28 09:51:06 CEST 2008
        - added AnyEvent::Handle, AnyEvent::Socket and AnyEvent::Util modules.
        - fix a bug in the pure perl backend that kept watchers alive
          when multiple watchers were registered for the same fh/poll combo.
	- add a benchmark section showing AnyEvent overhead and comparing
          the different event loops with each other.
        - prefer pure perl over tk when autoprobing, as it's about as fast,
          but doesn't crash with many watchers.
	- declare Qt support non-experimental.
        - clarify the confusing section about the file descriptor being
          kept alive.
        - document the race between loading of an event module and
          child processes exiting.
        - support POE as "backend" (with some caveats, POE is not generic
          enough, and darn slow).
        - support Wx and Prima through POE.
        - optimise perl backend to use 20% less memory and take advantage
          of typical timeout behaviour. It can now compete with
          select/poll-based C event loops in most cases (it is usually
          faster than Event and Glib :).
        - roughly cut EV memory use in half and increase its speed by 30%,
          by removing undocumented functionality. Did something similar
          to other event loops where possible.

3.2  Thu Apr 24 10:10:40 CEST 2008
	- do not die when anyevent watchers are destroyed while running callbacks
          in the pureperl backend (could only happen when two watchers are registered
          for the same fh/poll combo).
        - support autoloading for child watchers, was broken in all versions.
	- implement PERL_ANYEVENT_MODEL env variable.
        - (experimental) implement interface to the (very crashy/buggy) Event::Lib module.
        - (experimental) implement interface to the Qt module (cannot be autoprobed).
        - this release is IO:AnyEvent-proof.

3.12 Tue Apr 22 07:11:46 CEST 2008
	- reinstate AUTHOR section that got lost somehow.
        - do not hang in the testsuite with badly broken perls (activestate,
          strawberry...), but instead diagnose the problem and continue.
        - use INT instead of CHLD in an attempt to support broken windows perls
          better (this decreases the test quality, unfortunately).
        - do not send the signal to the process group
          (no problem for CHLD, bad for INT :).

3.11 Sat Apr 19 06:57:31 CEST 2008
	- major documentation rework.
        - document the fact that child watchers only watch for zombies.
        - fix the child watcher example.

3.1  Wed Apr 16 17:09:01 CEST 2008
	- work around recurring bugs in Tk by dup'ing filehandles, the
          only method with good success chances on Tk (the bugs apparently
          don't get fixed anytime soon).
        - lift the restriction of only one watcher per fh direction
          (as the Tk bug workaround also lifts it and only Tk imposed
          such strong limits).
        - changed probe order to prefer coro adaptors.
        - explain why recursion into the event loop is not supported
          unless the backend supports it (only Coro::EV does without
          any restrictions...).
        - add simple manpages for all backend modules.

3.0  Mon Apr  7 21:30:23 CEST 2008
	- Coro::Signal changed semantics, roll our own, also cleaning
          up the Coro implementation in general.
        - rename Coro backend to CoroEvent.
        - add some decision helping paragraph to the manpage that should
          help people to decide whether AnyEvent is the right thing for them.

2.9  Mon Jan 28 13:31:54 CET 2008
	- update for EV 3.0 API changes.

2.8  Sun Nov 25 15:06:03 CET 2007
	- waitpid can validly return 0. accept this fact of life
          instead of reporting it to any watchers.

2.7  Fri Nov 23 11:41:14 CET 2007
	- force use of AnyEvent::Impl::Perl in testsuite, there is
          too much breakage outside AnyEvent.
        - deliver signals synchronously in AnyEvent::Impl::Perl
          or any other event loop that relies on AnyEvents child
          watcher emulation. *Could* help with hanging testsuite
          (except when Event or EV are installed).

2.6  Fri Nov  9 20:36:35 CET 2007
	- fix bug in testsuite.
        - move EV adaptor modules to AnyEvent.
        - add Coro+EV adaptor module.

2.55 Tue Nov  6 17:41:32 CET 2007
	- add EV to the list of supported event models.
        - do not auto-reset pid watchers, pass pid and status to them.
        - allow a pid of zero to watch for all children in child watchers.

2.54 Wed Jul 18 17:36:23 CEST 2007
	- work around a perl bug that results in
          BEGIN not safe after errors--compilation aborted without
          any discernible reason or error message by once
          again not use'ing strict.

2.53 Sun Jul  8 10:51:53 CEST 2007
	- make Glib interface work again (spotted by elmex).

2.52 Wed Mar  7 18:36:16 CET 2007
	- child watchers stopped working permanently when
          all current child watchers were destroyed.

2.51 Mon Dec 11 21:33:24 CET 2006
	- work around bugs in perl where eval "require Module" returns
          true even if the module couldn't be loaded :/.

2.5  Mon Dec 11 02:15:28 CET 2006
	- avoid $AUTOLOAD because many perls corrupt it.
        - AnyEvent::detect forces autodetection.
        - implement signal watchers (experimental)
          (Tk does not support async signals (<= 804.027 at least),
          so they might get delayed indefinitely on Tk).
        - implement child watchers (experimental).
        - moved default condvar implementation into base module, simplifying
          most implementation modules.

2.1  Fri Nov 24 15:50:48 CET 2006
	- better docs.
        - simple testsuite.
        - added AnyEvent->one_event method for special purposes.

2.0  Wed Nov  1 02:21:30 CET 2006
	- INCOMPATIBLE CHANGE: poll can now either be "r" or "w" but not both.
	- INCOMPATIBLE CHANGE: io watcher callbacks have no arguments anymore.
        - EXPERIMENTAL:
          new pure-perl model added, AnyEvent now always finds a suitable
          event model.
        - improved documentation.
        - reduced cpu overhead.

1.02 Fri Jan 13 14:15:40 CET 2006
	- add COPYING to clarify license.

1.01 Sun Jan  8 05:48:09 CET 2006
	- correctly cancel Event timers, avoid memory leaks.

1.0  Sun Jan  8 05:40:58 CET 2006
	- allow third-party interfaces to be added at runtime.
        - fix version number.

0.9  Fri Dec 30 02:25:41 CET 2005
	- fixed errors in the documentation.
        - going back to using $VERSION as "load check",
          as suggested by Jörn Reder. Let's see why I stopped using
          this method.
        - add 'hup' to the list of events to receive for glib
          and generate 'rw' events for it, as some versions of
          glib seem to require this, while this breaks other
          (older) versions of glib.
        - implemented PERL_ANYEVENT_VERBOSE env variable.

0.3   Sun Dec  4 10:43:33 CET 2005
        - no changes to the code.
	- fixed documentation.
        - clarified some corner cases.
        - vastly improved documentation with a long example.

0.2   Thu Dec  1 22:18:49 CET 2005
	- check whether a valid method was called, to avoid endless loops.
        - move condvars into their own class (cleaner, safer).
        - don't die when some modules are not available.
        - Tk is too broken to allow for words. Will work once Tk works.

0.1   Thu Dec  1 19:44:57 CET 2005
	- spiced up and cleaned up for release.

0.01  Tue Sep 21 18:12:09 CEST 2004
	- original version; copied from Convert::Scalar.