The Perl Toolchain Summit needs more sponsors. If your company depends on Perl, please support this very important event.
=head1 NAME

Imager::API - Imager's C API - introduction.

=head1 SYNOPSIS

  #include "imext.h"
  #include "imperl.h"

  DEFINE_IMAGER_CALLBACKS;

  MODULE = Your::Module  PACKAGE = Your::Module

  ...

  BOOT:
    /* any release with the API */
    PERL_INITIALIZE_IMAGER_CALLBACKS;
    /* preferred from Imager 0.91 */
    PERL_INITIALIZE_IMAGER_CALLBACKS_NAME("My::Module");
  

=head1 DESCRIPTION

=for stopwords XS

The API allows you to access Imager functions at the C level from XS
and from C<Inline::C>.

The intent is to allow users to:

=over

=item *

write C code that does Imager operations the user might do from Perl,
but faster, for example, the L<Imager::CountColor> example.

=item *

write C code that implements an application specific version of some
core Imager object, for example, Imager::SDL.

=item *

write C code that hooks into Imager's existing methods, such as filter
or file format handlers.

=back

See L<Imager::Inline> for information on using Imager's Inline::C
support.

=head1 Beware

=over

=item *

don't return an object you received as a parameter - this will cause
the object to be freed twice.

=back

=head1 Types

The API makes the following types visible:

=over

=item *

L</i_img> - used to represent an image

=item *

L</i_color> - used to represent a color with up
to 8 bits per sample.

=item *

L</i_fcolor> - used to represent
a color with a double per sample.

=item *

L</i_fill_t> - fill objects>> - an abstract fill

=item *

L</im_context_t> - Imager's per-thread state.

=back

At this point there is no consolidated font object type, and hence the
font functions are not visible through Imager's API.

=head2 i_img

This contains the dimensions of the image (C<xsize>, C<ysize>,
C<channels>), image metadata (C<ch_mask>, C<bits>, C<type>,
C<virtual>), potentially image data (C<idata>) and a function table,
with pointers to functions to perform various low level image
operations.

The only time you should directly write to any value in this type is
if you're implementing your own image type.

The typemap includes type names Imager and Imager::ImgRaw as typedefs
for C<i_img *>.

For incoming parameters the typemap will accept either Imager or
Imager::ImgRaw objects.

For return values the typemap will produce a full Imager object for an
Imager return type and a raw image object for an Imager::ImgRaw return
type.

=head2 i_color

Represents an 8-bit per sample color.  This is a union containing
several different structs for access to components of a color:

=over

=item *

C<gray> - single member C<gray_color>.

=item *

C<rgb> - C<r>, C<g>, C<b> members.

=item *

C<rgba> - C<r>, C<g>, C<b>, C<a> members.

=item *

C<channels> - array of channels.

=back

Use C<Imager::Color> for parameter and return value types.

=head2 i_fcolor

Similar to C<i_color> except that each component is a double instead of
an unsigned char.

Use Imager::Color::Float for parameter and return value types.

=head2 i_fill_t

Abstract type containing pointers called to perform low level fill
operations.

Unless you're defining your own fill objects you should treat this as
an opaque type.

Use Imager::FillHandle for parameter and return value types.  At the
Perl level this is stored in the C<fill> member of the Perl level
Imager::Fill object.

=head2 i_io_glue_t

C<i_io_glue_t> is Imager's I/O abstraction.

Historically named C<io_glue>, and this name is available for backward
compatibility.

=head2 im_context_t

This new type is an opaque type that stores Imager's per-thread state,
including the error message stack, the current log file state and
image size file limits.

While Imager's internal typemap provides a C<T_PTROBJ> mapping and a
DESTROY method for this type you B<must> never return objects of this
type back to perl.

See L</Context objects> for more information.

=head1 Create an XS module using the Imager API

=head2 Foo.pm

Load Imager:

  use Imager 0.48;

and bootstrap your XS code - see L<XSLoader> or L<DynaLoader>.

=head2 C<Foo.xs>

You'll need the following in your XS source:

=over

=item *

include the Imager external API header, and the perl interface header:

  #include "imext.h"
  #include "imperl.h"

=item *

create the variables used to hold the callback table:

  DEFINE_IMAGER_CALLBACKS;

=item *

initialize the callback table in your C<BOOT> code:

  BOOT:
    PERL_INITIALIZE_IMAGER_CALLBACKS;

From Imager 0.91 you can supply your module name to improve error
reporting:

  BOOT:
    PERL_INITIALIZE_IMAGER_CALLBACKS_NAME("My::Module");

=back

=head2 foo.c

In any other source files where you want to access the Imager API,
you'll need to:

=over

=item *

include the Imager external API header:

  #include "imext.h"

=back

=head2 C<Makefile.PL>

If you're creating an XS module that depends on Imager's API your
C<Makefile.PL> will need to do the following:

=over

=item *

C<use Imager::ExtUtils;>

=item *

include Imager's include directory in INC:

  INC => Imager::ExtUtils->includes

=item *

use Imager's typemap:

  TYPEMAPS => [ Imager::ExtUtils->typemap ]

=item *

include Imager 0.48 as a PREREQ_PM:

   PREREQ_PM =>
   {
    Imager => 0.48,
   },

=item *

Since you use Imager::ExtUtils in C<Makefile.PL> (or C<Build.PL>) you
should include Imager in your configure_requires:

   META_MERGE =>
   {
     configure_requires => { Imager => "0.48" }
   },

=back

=head1 Context objects

Starting with Imager 0.93, Imager keeps some state per-thread rather
than storing it in global (or static) variables.  The intent is to
improve support for multi-threaded perl programs.

For the typical XS or Inline::C module using Imager's API this won't
matter - the changes are hidden behind macros and rebuilding your
module should require no source code changes.

Some operations will be slightly slower, these include:

=over

=item *

creating an image

=item *

reporting errors

=item *

creating I/O objects

=item *

setting/getting/testing image file limits

=item *

logging

=back

You can avoid this fairly minor overhead by adding a C<#define>:

  #define IMAGER_NO_CONTEXT

before including any Imager header files, but you will need to manage
context objects yourself.

Some functions and macros that are available without
C<IMAGER_NO_CONTEXT> are not available with it defined, these are:

=over

=item *

mm_log() - to avoid using a different context object for the line
header and the line text you need to use im_log() instead, with a
context object visible in scope.

=back

=head2 C<aIMCTX>

With C<IMAGER_NO_CONTEXT> defined, C<aIMCTX> refers to the locally
defined context object, either via one the of the C<dIMCTX> macros or
as a parameter with the C<pIMCTX> macro.

Without C<IMAGER_NO_CONTEXT>, C<aIMCTX> is a call to
C<im_get_context()> which retrieves the context object for the current
thread.

There is no C<aIMCTX_> macro, any Imager function that can accept a
context parameter always accepts it.

=head2 C<pIMCTX>

This macro declares a variable of type L</im_context_t> that's
accessible via the C<aIMCTX> macro.  This is intended for use as a
parameter declaration for functions:

  void f(pIMCTX) {
    ... use aIMCTX here
  }

  void g(...) {
    ...
    f(aIMCTX);
  }

=head2 C<dIMCTX>

Defines a local context variable and initializes it via
L<im_get_context()|Imager::APIRef/im_get_context()>.

=head2 C<dIMCTXim>

Defines a local context variable and initializes it from the context
stored in an L<image object|/i_img>, eg:

  void f(i_img *im) {
    dIMCTXim(im);
    ...
  }

=head2 C<dIMCTXio>

Defines a local context variable and initializes it from the context
stored in an L<< IE<47>O object|/i_io_glue_t >> object.

  void f(i_io_glue_t *io) {
    dIMCTXio(io);
    ...
  }

=head2 C<dIMCTXctx>

Defines a local context variable accessible via C<aIMCTX> in terms of
an expression you supply:

  void f(my_object *p) {
    dIMCTXctx(p->context);
    ...
  }

This can be used to define your own local context macro:

  #define dIMCTXmine(mine) ((mine)->context)

  void f(my_object *p) {
    dIMCTXmine(p);
    ...
  }

=head1 Mutex Functions

Since some libraries are not thread safe, Imager's API includes some
simple mutex functions.

To create a mutex:

  i_mutex_t m = i_mutex_new();

To control or lock the mutex:

  i_mutex_lock(m);

To release or unlock the mutex:

  i_mutex_unlock(m);

To free any resources used by the mutex:

  i_mutex_destroy(m);

I most cases where you'd use these functions, your code would create
the mutex in your BOOT section, then lock and unlock the mutex as
needed to control access to the library.

=head1 Context slots

=for stopwords
TLS APIs

To avoid abstracting the platform TLS and thread clean up handling,
Imager provides simple APIs for storing per-context information.

To allocate a slot:

  im_slot_t slot = im_context_slot_new(callback)

where callback is a (possibly NULL) function pointer called when the
context object is destroyed.

By default, the stored value for a slot is NULL, whether for a new
context or for a cloned context.

To store a value:

  im_context_slot_set(aIMCTX, slot, somevalue);

where C<somevalue> can be represented as a C<void *>.

To retrieve the value:

  value = im_context_slot_get(aIMCTX, slot);

=head1 AUTHOR

Tony Cook <tonyc@cpan.org>

=head1 SEE ALSO

Imager, Imager::ExtUtils, Imager::APIRef, Imager::Inline

=cut