source: rtems-graphics-toolkit/libpng-1.5.12/libpng-manual.txt @ 1cb4ff2

Last change on this file since 1cb4ff2 was 1cb4ff2, checked in by Alexandru-Sever Horin <alex.sever.h@…>, on 08/06/12 at 11:43:44

Added libpng-1.5.12 update.

  • Property mode set to 100644
File size: 190.5 KB
Line 
1Libpng-manual.txt - A description on how to use and modify libpng
2
3 libpng version 1.5.12 - July 11, 2012
4 Updated and distributed by Glenn Randers-Pehrson
5 <glennrp at users.sourceforge.net>
6 Copyright (c) 1998-2011 Glenn Randers-Pehrson
7
8 This document is released under the libpng license.
9 For conditions of distribution and use, see the disclaimer
10 and license in png.h
11
12 Based on:
13
14 libpng versions 0.97, January 1998, through 1.5.12 - July 11, 2012
15 Updated and distributed by Glenn Randers-Pehrson
16 Copyright (c) 1998-2011 Glenn Randers-Pehrson
17
18 libpng 1.0 beta 6  version 0.96 May 28, 1997
19 Updated and distributed by Andreas Dilger
20 Copyright (c) 1996, 1997 Andreas Dilger
21
22 libpng 1.0 beta 2 - version 0.88  January 26, 1996
23 For conditions of distribution and use, see copyright
24 notice in png.h. Copyright (c) 1995, 1996 Guy Eric
25 Schalnat, Group 42, Inc.
26
27 Updated/rewritten per request in the libpng FAQ
28 Copyright (c) 1995, 1996 Frank J. T. Wojcik
29 December 18, 1995 & January 20, 1996
30
31I. Introduction
32
33This file describes how to use and modify the PNG reference library
34(known as libpng) for your own use.  There are five sections to this
35file: introduction, structures, reading, writing, and modification and
36configuration notes for various special platforms.  In addition to this
37file, example.c is a good starting point for using the library, as
38it is heavily commented and should include everything most people
39will need.  We assume that libpng is already installed; see the
40INSTALL file for instructions on how to install libpng.
41
42For examples of libpng usage, see the files "example.c", "pngtest.c",
43and the files in the "contrib" directory, all of which are included in
44the libpng distribution.
45
46Libpng was written as a companion to the PNG specification, as a way
47of reducing the amount of time and effort it takes to support the PNG
48file format in application programs.
49
50The PNG specification (second edition), November 2003, is available as
51a W3C Recommendation and as an ISO Standard (ISO/IEC 15948:2003 (E)) at
52<http://www.w3.org/TR/2003/REC-PNG-20031110/
53The W3C and ISO documents have identical technical content.
54
55The PNG-1.2 specification is available at
56<http://www.libpng.org/pub/png/documents/>.  It is technically equivalent
57to the PNG specification (second edition) but has some additional material.
58
59The PNG-1.0 specification is available
60as RFC 2083 <http://www.libpng.org/pub/png/documents/> and as a
61W3C Recommendation <http://www.w3.org/TR/REC.png.html>.
62
63Some additional chunks are described in the special-purpose public chunks
64documents at <http://www.libpng.org/pub/png/documents/>.
65
66Other information
67about PNG, and the latest version of libpng, can be found at the PNG home
68page, <http://www.libpng.org/pub/png/>.
69
70Most users will not have to modify the library significantly; advanced
71users may want to modify it more.  All attempts were made to make it as
72complete as possible, while keeping the code easy to understand.
73Currently, this library only supports C.  Support for other languages
74is being considered.
75
76Libpng has been designed to handle multiple sessions at one time,
77to be easily modifiable, to be portable to the vast majority of
78machines (ANSI, K&R, 16-, 32-, and 64-bit) available, and to be easy
79to use.  The ultimate goal of libpng is to promote the acceptance of
80the PNG file format in whatever way possible.  While there is still
81work to be done (see the TODO file), libpng should cover the
82majority of the needs of its users.
83
84Libpng uses zlib for its compression and decompression of PNG files.
85Further information about zlib, and the latest version of zlib, can
86be found at the zlib home page, <http://www.info-zip.org/pub/infozip/zlib/>.
87The zlib compression utility is a general purpose utility that is
88useful for more than PNG files, and can be used without libpng.
89See the documentation delivered with zlib for more details.
90You can usually find the source files for the zlib utility wherever you
91find the libpng source files.
92
93Libpng is thread safe, provided the threads are using different
94instances of the structures.  Each thread should have its own
95png_struct and png_info instances, and thus its own image.
96Libpng does not protect itself against two threads using the
97same instance of a structure.
98
99II. Structures
100
101There are two main structures that are important to libpng, png_struct
102and png_info.  Both are internal structures that are no longer exposed
103in the libpng interface (as of libpng 1.5.0).
104
105The png_info structure is designed to provide information about the
106PNG file.  At one time, the fields of png_info were intended to be
107directly accessible to the user.  However, this tended to cause problems
108with applications using dynamically loaded libraries, and as a result
109a set of interface functions for png_info (the png_get_*() and png_set_*()
110functions) was developed, and direct access to the png_info fields was
111deprecated..
112
113The png_struct structure is the object used by the library to decode a
114single image.  As of 1.5.0 this structure is also not exposed.
115
116Almost all libpng APIs require a pointer to a png_struct as the first argument.
117Many (in particular the png_set and png_get APIs) also require a pointer
118to png_info as the second argument.  Some application visible macros
119defined in png.h designed for basic data access (reading and writing
120integers in the PNG format) don't take a png_info pointer, but it's almost
121always safe to assume that a (png_struct*) has to be passed to call an API
122function.
123
124You can have more than one png_info structure associated with an image,
125as illustrated in pngtest.c, one for information valid prior to the
126IDAT chunks and another (called "end_info" below) for things after them.
127
128The png.h header file is an invaluable reference for programming with libpng.
129And while I'm on the topic, make sure you include the libpng header file:
130
131#include <png.h>
132
133and also (as of libpng-1.5.0) the zlib header file, if you need it:
134
135#include <zlib.h>
136
137Types
138
139The png.h header file defines a number of integral types used by the
140APIs.  Most of these are fairly obvious; for example types corresponding
141to integers of particular sizes and types for passing color values.
142
143One exception is how non-integral numbers are handled.  For application
144convenience most APIs that take such numbers have C (double) arguments;
145however, internally PNG, and libpng, use 32 bit signed integers and encode
146the value by multiplying by 100,000.  As of libpng 1.5.0 a convenience
147macro PNG_FP_1 is defined in png.h along with a type (png_fixed_point)
148which is simply (png_int_32).
149
150All APIs that take (double) arguments also have a matching API that
151takes the corresponding fixed point integer arguments.  The fixed point
152API has the same name as the floating point one with "_fixed" appended.
153The actual range of values permitted in the APIs is frequently less than
154the full range of (png_fixed_point) (-21474 to +21474).  When APIs require
155a non-negative argument the type is recorded as png_uint_32 above.  Consult
156the header file and the text below for more information.
157
158Special care must be take with sCAL chunk handling because the chunk itself
159uses non-integral values encoded as strings containing decimal floating point
160numbers.  See the comments in the header file.
161
162Configuration
163
164The main header file function declarations are frequently protected by C
165preprocessing directives of the form:
166
167    #ifdef PNG_feature_SUPPORTED
168    declare-function
169    #endif
170    ...
171    #ifdef PNG_feature_SUPPORTED
172    use-function
173    #endif
174
175The library can be built without support for these APIs, although a
176standard build will have all implemented APIs.  Application programs
177should check the feature macros before using an API for maximum
178portability.  From libpng 1.5.0 the feature macros set during the build
179of libpng are recorded in the header file "pnglibconf.h" and this file
180is always included by png.h.
181
182If you don't need to change the library configuration from the default, skip to
183the next section ("Reading").
184
185Notice that some of the makefiles in the 'scripts' directory and (in 1.5.0) all
186of the build project files in the 'projects' directory simply copy
187scripts/pnglibconf.h.prebuilt to pnglibconf.h.  This means that these build
188systems do not permit easy auto-configuration of the library - they only
189support the default configuration.
190
191The easiest way to make minor changes to the libpng configuration when
192auto-configuration is supported is to add definitions to the command line
193using (typically) CPPFLAGS.  For example:
194
195CPPFLAGS=-DPNG_NO_FLOATING_ARITHMETIC
196
197will change the internal libpng math implementation for gamma correction and
198other arithmetic calculations to fixed point, avoiding the need for fast
199floating point support.  The result can be seen in the generated pnglibconf.h -
200make sure it contains the changed feature macro setting.
201
202If you need to make more extensive configuration changes - more than one or two
203feature macro settings - you can either add -DPNG_USER_CONFIG to the build
204command line and put a list of feature macro settings in pngusr.h or you can set
205DFA_XTRA (a makefile variable) to a file containing the same information in the
206form of 'option' settings.
207
208A. Changing pnglibconf.h
209
210A variety of methods exist to build libpng.  Not all of these support
211reconfiguration of pnglibconf.h.  To reconfigure pnglibconf.h it must either be
212rebuilt from scripts/pnglibconf.dfa using awk or it must be edited by hand.
213
214Hand editing is achieved by copying scripts/pnglibconf.h.prebuilt to
215pnglibconf.h and changing the lines defining the supported features, paying
216very close attention to the 'option' information in scripts/pnglibconf.dfa
217that describes those features and their requirements.  This is easy to get
218wrong.
219
220B. Configuration using DFA_XTRA
221
222Rebuilding from pnglibconf.dfa is easy if a functioning 'awk', or a later
223variant such as 'nawk' or 'gawk', is available.  The configure build will
224automatically find an appropriate awk and build pnglibconf.h.
225The scripts/pnglibconf.mak file contains a set of make rules for doing the
226same thing if configure is not used, and many of the makefiles in the scripts
227directory use this approach.
228
229When rebuilding simply write a new file containing changed options and set
230DFA_XTRA to the name of this file.  This causes the build to append the new file
231to the end of scripts/pnglibconf.dfa.  The pngusr.dfa file should contain lines
232of the following forms:
233
234everything = off
235
236This turns all optional features off.  Include it at the start of pngusr.dfa to
237make it easier to build a minimal configuration.  You will need to turn at least
238some features on afterward to enable either reading or writing code, or both.
239
240option feature on
241option feature off
242
243Enable or disable a single feature.  This will automatically enable other
244features required by a feature that is turned on or disable other features that
245require a feature which is turned off.  Conflicting settings will cause an error
246message to be emitted by awk.
247
248setting feature default value
249
250Changes the default value of setting 'feature' to 'value'.  There are a small
251number of settings listed at the top of pnglibconf.h, they are documented in the
252source code.  Most of these values have performance implications for the library
253but most of them have no visible effect on the API.  Some can also be overridden
254from the API.
255
256This method of building a customized pnglibconf.h is illustrated in
257contrib/pngminim/*.  See the "$(PNGCONF):" target in the makefile and
258pngusr.dfa in these directories.
259
260C. Configuration using PNG_USR_CONFIG
261
262If -DPNG_USR_CONFIG is added to the CFLAGS when pnglibconf.h is built the file
263pngusr.h will automatically be included before the options in
264scripts/pnglibconf.dfa are processed.  Your pngusr.h file should contain only
265macro definitions turning features on or off or setting settings.
266
267Apart from the global setting "everything = off" all the options listed above
268can be set using macros in pngusr.h:
269
270#define PNG_feature_SUPPORTED
271
272is equivalent to:
273
274option feature on
275
276#define PNG_NO_feature
277
278is equivalent to:
279
280option feature off
281
282#define PNG_feature value
283
284is equivalent to:
285
286setting feature default value
287
288Notice that in both cases, pngusr.dfa and pngusr.h, the contents of the
289pngusr file you supply override the contents of scripts/pnglibconf.dfa
290
291If confusing or incomprehensible behavior results it is possible to
292examine the intermediate file pnglibconf.dfn to find the full set of
293dependency information for each setting and option.  Simply locate the
294feature in the file and read the C comments that precede it.
295
296This method is also illustrated in the contrib/pngminim/* makefiles and
297pngusr.h.
298
299III. Reading
300
301We'll now walk you through the possible functions to call when reading
302in a PNG file sequentially, briefly explaining the syntax and purpose
303of each one.  See example.c and png.h for more detail.  While
304progressive reading is covered in the next section, you will still
305need some of the functions discussed in this section to read a PNG
306file.
307
308Setup
309
310You will want to do the I/O initialization(*) before you get into libpng,
311so if it doesn't work, you don't have much to undo.  Of course, you
312will also want to insure that you are, in fact, dealing with a PNG
313file.  Libpng provides a simple check to see if a file is a PNG file.
314To use it, pass in the first 1 to 8 bytes of the file to the function
315png_sig_cmp(), and it will return 0 (false) if the bytes match the
316corresponding bytes of the PNG signature, or nonzero (true) otherwise.
317Of course, the more bytes you pass in, the greater the accuracy of the
318prediction.
319
320If you are intending to keep the file pointer open for use in libpng,
321you must ensure you don't read more than 8 bytes from the beginning
322of the file, and you also have to make a call to png_set_sig_bytes_read()
323with the number of bytes you read from the beginning.  Libpng will
324then only check the bytes (if any) that your program didn't read.
325
326(*): If you are not using the standard I/O functions, you will need
327to replace them with custom functions.  See the discussion under
328Customizing libpng.
329
330
331    FILE *fp = fopen(file_name, "rb");
332    if (!fp)
333    {
334       return (ERROR);
335    }
336
337    fread(header, 1, number, fp);
338    is_png = !png_sig_cmp(header, 0, number);
339
340    if (!is_png)
341    {
342       return (NOT_PNG);
343    }
344
345
346Next, png_struct and png_info need to be allocated and initialized.  In
347order to ensure that the size of these structures is correct even with a
348dynamically linked libpng, there are functions to initialize and
349allocate the structures.  We also pass the library version, optional
350pointers to error handling functions, and a pointer to a data struct for
351use by the error functions, if necessary (the pointer and functions can
352be NULL if the default error handlers are to be used).  See the section
353on Changes to Libpng below regarding the old initialization functions.
354The structure allocation functions quietly return NULL if they fail to
355create the structure, so your application should check for that.
356
357    png_structp png_ptr = png_create_read_struct
358        (PNG_LIBPNG_VER_STRING, (png_voidp)user_error_ptr,
359        user_error_fn, user_warning_fn);
360
361    if (!png_ptr)
362       return (ERROR);
363
364    png_infop info_ptr = png_create_info_struct(png_ptr);
365
366    if (!info_ptr)
367    {
368       png_destroy_read_struct(&png_ptr,
369           (png_infopp)NULL, (png_infopp)NULL);
370       return (ERROR);
371    }
372
373If you want to use your own memory allocation routines,
374use a libpng that was built with PNG_USER_MEM_SUPPORTED defined, and use
375png_create_read_struct_2() instead of png_create_read_struct():
376
377    png_structp png_ptr = png_create_read_struct_2
378        (PNG_LIBPNG_VER_STRING, (png_voidp)user_error_ptr,
379        user_error_fn, user_warning_fn, (png_voidp)
380        user_mem_ptr, user_malloc_fn, user_free_fn);
381
382The error handling routines passed to png_create_read_struct()
383and the memory alloc/free routines passed to png_create_struct_2()
384are only necessary if you are not using the libpng supplied error
385handling and memory alloc/free functions.
386
387When libpng encounters an error, it expects to longjmp back
388to your routine.  Therefore, you will need to call setjmp and pass
389your png_jmpbuf(png_ptr).  If you read the file from different
390routines, you will need to update the longjmp buffer every time you enter
391a new routine that will call a png_*() function.
392
393See your documentation of setjmp/longjmp for your compiler for more
394information on setjmp/longjmp.  See the discussion on libpng error
395handling in the Customizing Libpng section below for more information
396on the libpng error handling.  If an error occurs, and libpng longjmp's
397back to your setjmp, you will want to call png_destroy_read_struct() to
398free any memory.
399
400    if (setjmp(png_jmpbuf(png_ptr)))
401    {
402       png_destroy_read_struct(&png_ptr, &info_ptr,
403           &end_info);
404       fclose(fp);
405       return (ERROR);
406    }
407
408Pass (png_infopp)NULL instead of &end_info if you didn't create
409an end_info structure.
410
411If you would rather avoid the complexity of setjmp/longjmp issues,
412you can compile libpng with PNG_NO_SETJMP, in which case
413errors will result in a call to PNG_ABORT() which defaults to abort().
414
415You can #define PNG_ABORT() to a function that does something
416more useful than abort(), as long as your function does not
417return.
418
419Now you need to set up the input code.  The default for libpng is to
420use the C function fread().  If you use this, you will need to pass a
421valid FILE * in the function png_init_io().  Be sure that the file is
422opened in binary mode.  If you wish to handle reading data in another
423way, you need not call the png_init_io() function, but you must then
424implement the libpng I/O methods discussed in the Customizing Libpng
425section below.
426
427    png_init_io(png_ptr, fp);
428
429If you had previously opened the file and read any of the signature from
430the beginning in order to see if this was a PNG file, you need to let
431libpng know that there are some bytes missing from the start of the file.
432
433    png_set_sig_bytes(png_ptr, number);
434
435You can change the zlib compression buffer size to be used while
436reading compressed data with
437
438    png_set_compression_buffer_size(png_ptr, buffer_size);
439
440where the default size is 8192 bytes.  Note that the buffer size
441is changed immediately and the buffer is reallocated immediately,
442instead of setting a flag to be acted upon later.
443
444If you want CRC errors to be handled in a different manner than
445the default, use
446
447    png_set_crc_action(png_ptr, crit_action, ancil_action);
448
449The values for png_set_crc_action() say how libpng is to handle CRC errors in
450ancillary and critical chunks, and whether to use the data contained
451therein.  Note that it is impossible to "discard" data in a critical
452chunk.
453
454Choices for (int) crit_action are
455   PNG_CRC_DEFAULT      0  error/quit
456   PNG_CRC_ERROR_QUIT   1  error/quit
457   PNG_CRC_WARN_USE     3  warn/use data
458   PNG_CRC_QUIET_USE    4  quiet/use data
459   PNG_CRC_NO_CHANGE    5  use the current value
460
461Choices for (int) ancil_action are
462   PNG_CRC_DEFAULT      0  error/quit
463   PNG_CRC_ERROR_QUIT   1  error/quit
464   PNG_CRC_WARN_DISCARD 2  warn/discard data
465   PNG_CRC_WARN_USE     3  warn/use data
466   PNG_CRC_QUIET_USE    4  quiet/use data
467   PNG_CRC_NO_CHANGE    5  use the current value
468
469Setting up callback code
470
471You can set up a callback function to handle any unknown chunks in the
472input stream. You must supply the function
473
474    read_chunk_callback(png_structp png_ptr,
475         png_unknown_chunkp chunk);
476    {
477       /* The unknown chunk structure contains your
478          chunk data, along with similar data for any other
479          unknown chunks: */
480
481           png_byte name[5];
482           png_byte *data;
483           png_size_t size;
484
485       /* Note that libpng has already taken care of
486          the CRC handling */
487
488       /* put your code here.  Search for your chunk in the
489          unknown chunk structure, process it, and return one
490          of the following: */
491
492       return (-n); /* chunk had an error */
493       return (0); /* did not recognize */
494       return (n); /* success */
495    }
496
497(You can give your function another name that you like instead of
498"read_chunk_callback")
499
500To inform libpng about your function, use
501
502    png_set_read_user_chunk_fn(png_ptr, user_chunk_ptr,
503        read_chunk_callback);
504
505This names not only the callback function, but also a user pointer that
506you can retrieve with
507
508    png_get_user_chunk_ptr(png_ptr);
509
510If you call the png_set_read_user_chunk_fn() function, then all unknown
511chunks will be saved when read, in case your callback function will need
512one or more of them.  This behavior can be changed with the
513png_set_keep_unknown_chunks() function, described below.
514
515At this point, you can set up a callback function that will be
516called after each row has been read, which you can use to control
517a progress meter or the like.  It's demonstrated in pngtest.c.
518You must supply a function
519
520    void read_row_callback(png_structp png_ptr,
521       png_uint_32 row, int pass);
522    {
523      /* put your code here */
524    }
525
526(You can give it another name that you like instead of "read_row_callback")
527
528To inform libpng about your function, use
529
530    png_set_read_status_fn(png_ptr, read_row_callback);
531
532When this function is called the row has already been completely processed and
533the 'row' and 'pass' refer to the next row to be handled.  For the
534non-interlaced case the row that was just handled is simply one less than the
535passed in row number, and pass will always be 0.  For the interlaced case the
536same applies unless the row value is 0, in which case the row just handled was
537the last one from one of the preceding passes.  Because interlacing may skip a
538pass you cannot be sure that the preceding pass is just 'pass-1', if you really
539need to know what the last pass is record (row,pass) from the callback and use
540the last recorded value each time.
541
542As with the user transform you can find the output row using the
543PNG_ROW_FROM_PASS_ROW macro.
544
545Unknown-chunk handling
546
547Now you get to set the way the library processes unknown chunks in the
548input PNG stream. Both known and unknown chunks will be read.  Normal
549behavior is that known chunks will be parsed into information in
550various info_ptr members while unknown chunks will be discarded. This
551behavior can be wasteful if your application will never use some known
552chunk types. To change this, you can call:
553
554    png_set_keep_unknown_chunks(png_ptr, keep,
555        chunk_list, num_chunks);
556    keep       - 0: default unknown chunk handling
557                 1: ignore; do not keep
558                 2: keep only if safe-to-copy
559                 3: keep even if unsafe-to-copy
560
561               You can use these definitions:
562                 PNG_HANDLE_CHUNK_AS_DEFAULT   0
563                 PNG_HANDLE_CHUNK_NEVER        1
564                 PNG_HANDLE_CHUNK_IF_SAFE      2
565                 PNG_HANDLE_CHUNK_ALWAYS       3
566
567    chunk_list - list of chunks affected (a byte string,
568                 five bytes per chunk, NULL or '\0' if
569                 num_chunks is 0)
570
571    num_chunks - number of chunks affected; if 0, all
572                 unknown chunks are affected.  If nonzero,
573                 only the chunks in the list are affected
574
575Unknown chunks declared in this way will be saved as raw data onto a
576list of png_unknown_chunk structures.  If a chunk that is normally
577known to libpng is named in the list, it will be handled as unknown,
578according to the "keep" directive.  If a chunk is named in successive
579instances of png_set_keep_unknown_chunks(), the final instance will
580take precedence.  The IHDR and IEND chunks should not be named in
581chunk_list; if they are, libpng will process them normally anyway.
582If you know that your application will never make use of some particular
583chunks, use PNG_HANDLE_CHUNK_NEVER (or 1) as demonstrated below.
584
585Here is an example of the usage of png_set_keep_unknown_chunks(),
586where the private "vpAg" chunk will later be processed by a user chunk
587callback function:
588
589    png_byte vpAg[5]={118, 112,  65, 103, (png_byte) '\0'};
590
591    #if defined(PNG_UNKNOWN_CHUNKS_SUPPORTED)
592      png_byte unused_chunks[]=
593      {
594        104,  73,  83,  84, (png_byte) '\0',   /* hIST */
595        105,  84,  88, 116, (png_byte) '\0',   /* iTXt */
596        112,  67,  65,  76, (png_byte) '\0',   /* pCAL */
597        115,  67,  65,  76, (png_byte) '\0',   /* sCAL */
598        115,  80,  76,  84, (png_byte) '\0',   /* sPLT */
599        116,  73,  77,  69, (png_byte) '\0',   /* tIME */
600      };
601    #endif
602
603    ...
604
605    #if defined(PNG_UNKNOWN_CHUNKS_SUPPORTED)
606      /* ignore all unknown chunks: */
607      png_set_keep_unknown_chunks(read_ptr, 1, NULL, 0);
608
609      /* except for vpAg: */
610      png_set_keep_unknown_chunks(read_ptr, 2, vpAg, 1);
611
612      /* also ignore unused known chunks: */
613      png_set_keep_unknown_chunks(read_ptr, 1, unused_chunks,
614         (int)sizeof(unused_chunks)/5);
615    #endif
616
617User limits
618
619The PNG specification allows the width and height of an image to be as
620large as 2^31-1 (0x7fffffff), or about 2.147 billion rows and columns.
621Since very few applications really need to process such large images,
622we have imposed an arbitrary 1-million limit on rows and columns.
623Larger images will be rejected immediately with a png_error() call. If
624you wish to change this limit, you can use
625
626   png_set_user_limits(png_ptr, width_max, height_max);
627
628to set your own limits, or use width_max = height_max = 0x7fffffffL
629to allow all valid dimensions (libpng may reject some very large images
630anyway because of potential buffer overflow conditions).
631
632You should put this statement after you create the PNG structure and
633before calling png_read_info(), png_read_png(), or png_process_data().
634
635When writing a PNG datastream, put this statement before calling
636png_write_info() or png_write_png().
637
638If you need to retrieve the limits that are being applied, use
639
640   width_max = png_get_user_width_max(png_ptr);
641   height_max = png_get_user_height_max(png_ptr);
642
643The PNG specification sets no limit on the number of ancillary chunks
644allowed in a PNG datastream.  You can impose a limit on the total number
645of sPLT, tEXt, iTXt, zTXt, and unknown chunks that will be stored, with
646
647   png_set_chunk_cache_max(png_ptr, user_chunk_cache_max);
648
649where 0x7fffffffL means unlimited.  You can retrieve this limit with
650
651   chunk_cache_max = png_get_chunk_cache_max(png_ptr);
652
653This limit also applies to the number of buffers that can be allocated
654by png_decompress_chunk() while decompressing iTXt, zTXt, and iCCP chunks.
655
656You can also set a limit on the amount of memory that a compressed chunk
657other than IDAT can occupy, with
658
659   png_set_chunk_malloc_max(png_ptr, user_chunk_malloc_max);
660
661and you can retrieve the limit with
662
663   chunk_malloc_max = png_get_chunk_malloc_max(png_ptr);
664
665Any chunks that would cause either of these limits to be exceeded will
666be ignored.
667
668Information about your system
669
670If you intend to display the PNG or to incorporate it in other image data you
671need to tell libpng information about your display or drawing surface so that
672libpng can convert the values in the image to match the display.
673
674From libpng-1.5.4 this information can be set before reading the PNG file
675header.  In earlier versions png_set_gamma() existed but behaved incorrectly if
676called before the PNG file header had been read and png_set_alpha_mode() did not
677exist.
678
679If you need to support versions prior to libpng-1.5.4 test the version number
680as illustrated below using "PNG_LIBPNG_VER >= 10504" and follow the procedures
681described in the appropriate manual page.
682
683You give libpng the encoding expected by your system expressed as a 'gamma'
684value.  You can also specify a default encoding for the PNG file in
685case the required information is missing from the file.  By default libpng
686assumes that the PNG data matches your system, to keep this default call:
687
688   png_set_gamma(png_ptr, screen_gamma, 1/screen_gamma/*file gamma*/);
689
690or you can use the fixed point equivalent:
691
692   png_set_gamma_fixed(png_ptr, PNG_FP_1*screen_gamma, PNG_FP_1/screen_gamma);
693
694If you don't know the gamma for your system it is probably 2.2 - a good
695approximation to the IEC standard for display systems (sRGB).  If images are
696too contrasty or washed out you got the value wrong - check your system
697documentation!
698
699Many systems permit the system gamma to be changed via a lookup table in the
700display driver, a few systems, including older Macs, change the response by
701default.  As of 1.5.4 three special values are available to handle common
702situations:
703
704   PNG_DEFAULT_sRGB: Indicates that the system conforms to the IEC 61966-2-1
705                     standard.  This matches almost all systems.
706   PNG_GAMMA_MAC_18: Indicates that the system is an older (pre Mac OS 10.6)
707                     Apple Macintosh system with the default settings.
708   PNG_GAMMA_LINEAR: Just the fixed point value for 1.0 - indicates that the
709                     system expects data with no gamma encoding.
710
711You would use the linear (unencoded) value if you need to process the pixel
712values further because this avoids the need to decode and reencode each
713component value whenever arithmetic is performed.  A lot of graphics software
714uses linear values for this reason, often with higher precision component values
715to preserve overall accuracy.
716
717The second thing you may need to tell libpng about is how your system handles
718alpha channel information.  Some, but not all, PNG files contain an alpha
719channel.  To display these files correctly you need to compose the data onto a
720suitable background, as described in the PNG specification.
721
722Libpng only supports composing onto a single color (using png_set_background;
723see below).  Otherwise you must do the composition yourself and, in this case,
724you may need to call png_set_alpha_mode:
725
726#if PNG_LIBPNG_VER >= 10504
727   png_set_alpha_mode(png_ptr, mode, screen_gamma);
728#else
729   png_set_gamma(png_ptr, screen_gamma, 1.0/screen_gamma);
730#endif
731
732The screen_gamma value is the same as the argument to png_set_gamma; however,
733how it affects the output depends on the mode.  png_set_alpha_mode() sets the
734file gamma default to 1/screen_gamma, so normally you don't need to call
735png_set_gamma.  If you need different defaults call png_set_gamma() before
736png_set_alpha_mode() - if you call it after it will override the settings made
737by png_set_alpha_mode().
738
739The mode is as follows:
740
741    PNG_ALPHA_PNG: The data is encoded according to the PNG specification.  Red,
742green and blue, or gray, components are gamma encoded color
743values and are not premultiplied by the alpha value.  The
744alpha value is a linear measure of the contribution of the
745pixel to the corresponding final output pixel.
746
747You should normally use this format if you intend to perform
748color correction on the color values; most, maybe all, color
749correction software has no handling for the alpha channel and,
750anyway, the math to handle pre-multiplied component values is
751unnecessarily complex.
752
753Before you do any arithmetic on the component values you need
754to remove the gamma encoding and multiply out the alpha
755channel.  See the PNG specification for more detail.  It is
756important to note that when an image with an alpha channel is
757scaled, linear encoded, pre-multiplied component values must
758be used!
759
760The remaining modes assume you don't need to do any further color correction or
761that if you do, your color correction software knows all about alpha (it
762probably doesn't!)
763
764    PNG_ALPHA_STANDARD:  The data libpng produces
765is encoded in the standard way
766assumed by most correctly written graphics software.
767The gamma encoding will be removed by libpng and the
768linear component values will be pre-multiplied by the
769alpha channel.
770
771With this format the final image must be re-encoded to
772match the display gamma before the image is displayed.
773If your system doesn't do that, yet still seems to
774perform arithmetic on the pixels without decoding them,
775it is broken - check out the modes below.
776
777With PNG_ALPHA_STANDARD libpng always produces linear
778component values, whatever screen_gamma you supply.  The
779screen_gamma value is, however, used as a default for
780the file gamma if the PNG file has no gamma information.
781
782If you call png_set_gamma() after png_set_alpha_mode() you
783will override the linear encoding.  Instead the
784pre-multiplied pixel values will be gamma encoded but
785the alpha channel will still be linear.  This may
786actually match the requirements of some broken software,
787but it is unlikely.
788
789While linear 8-bit data is often used it has
790insufficient precision for any image with a reasonable
791dynamic range.  To avoid problems, and if your software
792supports it, use png_set_expand_16() to force all
793components to 16 bits.
794
795    PNG_ALPHA_OPTIMIZED: This mode is the same
796as PNG_ALPHA_STANDARD except that
797completely opaque pixels are gamma encoded according to
798the screen_gamma value.  Pixels with alpha less than 1.0
799will still have linear components.
800
801Use this format if you have control over your
802compositing software and so don't do other arithmetic
803(such as scaling) on the data you get from libpng.  Your
804compositing software can simply copy opaque pixels to
805the output but still has linear values for the
806non-opaque pixels.
807
808In normal compositing, where the alpha channel encodes
809partial pixel coverage (as opposed to broad area
810translucency), the inaccuracies of the 8-bit
811representation of non-opaque pixels are irrelevant.
812
813You can also try this format if your software is broken;
814it might look better.
815
816    PNG_ALPHA_BROKEN: This is PNG_ALPHA_STANDARD;
817however, all component values,
818including the alpha channel are gamma encoded.  This is
819an appropriate format to try if your software, or more
820likely hardware, is totally broken, i.e., if it performs
821linear arithmetic directly on gamma encoded values.
822
823In most cases of broken software or hardware the bug in the final display
824manifests as a subtle halo around composited parts of the image.  You may not
825even perceive this as a halo; the composited part of the image may simply appear
826separate from the background, as though it had been cut out of paper and pasted
827on afterward.
828
829If you don't have to deal with bugs in software or hardware, or if you can fix
830them, there are three recommended ways of using png_set_alpha_mode():
831
832   png_set_alpha_mode(png_ptr, PNG_ALPHA_PNG,
833       screen_gamma);
834
835You can do color correction on the result (libpng does not currently
836support color correction internally).  When you handle the alpha channel
837you need to undo the gamma encoding and multiply out the alpha.
838
839   png_set_alpha_mode(png_ptr, PNG_ALPHA_STANDARD,
840       screen_gamma);
841   png_set_expand_16(png_ptr);
842
843If you are using the high level interface, don't call png_set_expand_16();
844instead pass PNG_TRANSFORM_EXPAND_16 to the interface.
845
846With this mode you can't do color correction, but you can do arithmetic,
847including composition and scaling, on the data without further processing.
848
849   png_set_alpha_mode(png_ptr, PNG_ALPHA_OPTIMIZED,
850       screen_gamma);
851
852You can avoid the expansion to 16-bit components with this mode, but you
853lose the ability to scale the image or perform other linear arithmetic.
854All you can do is compose the result onto a matching output.  Since this
855mode is libpng-specific you also need to write your own composition
856software.
857
858If you don't need, or can't handle, the alpha channel you can call
859png_set_background() to remove it by compositing against a fixed color.  Don't
860call png_set_strip_alpha() to do this - it will leave spurious pixel values in
861transparent parts of this image.
862
863   png_set_background(png_ptr, &background_color,
864       PNG_BACKGROUND_GAMMA_SCREEN, 0, 1);
865
866The background_color is an RGB or grayscale value according to the data format
867libpng will produce for you.  Because you don't yet know the format of the PNG
868file, if you call png_set_background at this point you must arrange for the
869format produced by libpng to always have 8-bit or 16-bit components and then
870store the color as an 8-bit or 16-bit color as appropriate.  The color contains
871separate gray and RGB component values, so you can let libpng produce gray or
872RGB output according to the input format, but low bit depth grayscale images
873must always be converted to at least 8-bit format.  (Even though low bit depth
874grayscale images can't have an alpha channel they can have a transparent
875color!)
876
877You set the transforms you need later, either as flags to the high level
878interface or libpng API calls for the low level interface.  For reference the
879settings and API calls required are:
880
8818-bit values:
882   PNG_TRANSFORM_SCALE_16 | PNG_EXPAND
883   png_set_expand(png_ptr); png_set_scale_16(png_ptr);
884
885   If you must get exactly the same inaccurate results
886   produced by default in versions prior to libpng-1.5.4,
887   use PNG_TRANSFORM_STRIP_16 and png_set_strip_16(png_ptr)
888   instead.
889
89016-bit values:
891   PNG_TRANSFORM_EXPAND_16
892   png_set_expand_16(png_ptr);
893
894In either case palette image data will be expanded to RGB.  If you just want
895color data you can add PNG_TRANSFORM_GRAY_TO_RGB or png_set_gray_to_rgb(png_ptr)
896to the list.
897
898Calling png_set_background before the PNG file header is read will not work
899prior to libpng-1.5.4.  Because the failure may result in unexpected warnings or
900errors it is therefore much safer to call png_set_background after the head has
901been read.  Unfortunately this means that prior to libpng-1.5.4 it cannot be
902used with the high level interface.
903
904The high-level read interface
905
906At this point there are two ways to proceed; through the high-level
907read interface, or through a sequence of low-level read operations.
908You can use the high-level interface if (a) you are willing to read
909the entire image into memory, and (b) the input transformations
910you want to do are limited to the following set:
911
912    PNG_TRANSFORM_IDENTITY      No transformation
913    PNG_TRANSFORM_SCALE_16      Strip 16-bit samples to
914                                8-bit accurately
915    PNG_TRANSFORM_STRIP_16      Chop 16-bit samples to
916                                8-bit less accurately
917    PNG_TRANSFORM_STRIP_ALPHA   Discard the alpha channel
918    PNG_TRANSFORM_PACKING       Expand 1, 2 and 4-bit
919                                samples to bytes
920    PNG_TRANSFORM_PACKSWAP      Change order of packed
921                                pixels to LSB first
922    PNG_TRANSFORM_EXPAND        Perform set_expand()
923    PNG_TRANSFORM_INVERT_MONO   Invert monochrome images
924    PNG_TRANSFORM_SHIFT         Normalize pixels to the
925                                sBIT depth
926    PNG_TRANSFORM_BGR           Flip RGB to BGR, RGBA
927                                to BGRA
928    PNG_TRANSFORM_SWAP_ALPHA    Flip RGBA to ARGB or GA
929                                to AG
930    PNG_TRANSFORM_INVERT_ALPHA  Change alpha from opacity
931                                to transparency
932    PNG_TRANSFORM_SWAP_ENDIAN   Byte-swap 16-bit samples
933    PNG_TRANSFORM_GRAY_TO_RGB   Expand grayscale samples
934                                to RGB (or GA to RGBA)
935    PNG_TRANSFORM_EXPAND_16     Expand samples to 16 bits
936
937(This excludes setting a background color, doing gamma transformation,
938quantizing, and setting filler.)  If this is the case, simply do this:
939
940    png_read_png(png_ptr, info_ptr, png_transforms, NULL)
941
942where png_transforms is an integer containing the bitwise OR of some
943set of transformation flags.  This call is equivalent to png_read_info(),
944followed the set of transformations indicated by the transform mask,
945then png_read_image(), and finally png_read_end().
946
947(The final parameter of this call is not yet used.  Someday it might point
948to transformation parameters required by some future input transform.)
949
950You must use png_transforms and not call any png_set_transform() functions
951when you use png_read_png().
952
953After you have called png_read_png(), you can retrieve the image data
954with
955
956   row_pointers = png_get_rows(png_ptr, info_ptr);
957
958where row_pointers is an array of pointers to the pixel data for each row:
959
960   png_bytep row_pointers[height];
961
962If you know your image size and pixel size ahead of time, you can allocate
963row_pointers prior to calling png_read_png() with
964
965   if (height > PNG_UINT_32_MAX/png_sizeof(png_byte))
966      png_error (png_ptr,
967          "Image is too tall to process in memory");
968
969   if (width > PNG_UINT_32_MAX/pixel_size)
970      png_error (png_ptr,
971          "Image is too wide to process in memory");
972
973   row_pointers = png_malloc(png_ptr,
974       height*png_sizeof(png_bytep));
975
976   for (int i=0; i<height, i++)
977      row_pointers[i]=NULL;  /* security precaution */
978
979   for (int i=0; i<height, i++)
980      row_pointers[i]=png_malloc(png_ptr,
981          width*pixel_size);
982
983   png_set_rows(png_ptr, info_ptr, &row_pointers);
984
985Alternatively you could allocate your image in one big block and define
986row_pointers[i] to point into the proper places in your block.
987
988If you use png_set_rows(), the application is responsible for freeing
989row_pointers (and row_pointers[i], if they were separately allocated).
990
991If you don't allocate row_pointers ahead of time, png_read_png() will
992do it, and it'll be free'ed by libpng when you call png_destroy_*().
993
994The low-level read interface
995
996If you are going the low-level route, you are now ready to read all
997the file information up to the actual image data.  You do this with a
998call to png_read_info().
999
1000    png_read_info(png_ptr, info_ptr);
1001
1002This will process all chunks up to but not including the image data.
1003
1004This also copies some of the data from the PNG file into the decode structure
1005for use in later transformations.  Important information copied in is:
1006
10071) The PNG file gamma from the gAMA chunk.  This overwrites the default value
1008provided by an earlier call to png_set_gamma or png_set_alpha_mode.
1009
10102) Prior to libpng-1.5.4 the background color from a bKGd chunk.  This
1011damages the information provided by an earlier call to png_set_background
1012resulting in unexpected behavior.  Libpng-1.5.4 no longer does this.
1013
10143) The number of significant bits in each component value.  Libpng uses this to
1015optimize gamma handling by reducing the internal lookup table sizes.
1016
10174) The transparent color information from a tRNS chunk.  This can be modified by
1018a later call to png_set_tRNS.
1019
1020Querying the info structure
1021
1022Functions are used to get the information from the info_ptr once it
1023has been read.  Note that these fields may not be completely filled
1024in until png_read_end() has read the chunk data following the image.
1025
1026    png_get_IHDR(png_ptr, info_ptr, &width, &height,
1027       &bit_depth, &color_type, &interlace_type,
1028       &compression_type, &filter_method);
1029
1030    width          - holds the width of the image
1031                     in pixels (up to 2^31).
1032
1033    height         - holds the height of the image
1034                     in pixels (up to 2^31).
1035
1036    bit_depth      - holds the bit depth of one of the
1037                     image channels.  (valid values are
1038                     1, 2, 4, 8, 16 and depend also on
1039                     the color_type.  See also
1040                     significant bits (sBIT) below).
1041
1042    color_type     - describes which color/alpha channels
1043                         are present.
1044                     PNG_COLOR_TYPE_GRAY
1045                        (bit depths 1, 2, 4, 8, 16)
1046                     PNG_COLOR_TYPE_GRAY_ALPHA
1047                        (bit depths 8, 16)
1048                     PNG_COLOR_TYPE_PALETTE
1049                        (bit depths 1, 2, 4, 8)
1050                     PNG_COLOR_TYPE_RGB
1051                        (bit_depths 8, 16)
1052                     PNG_COLOR_TYPE_RGB_ALPHA
1053                        (bit_depths 8, 16)
1054
1055                     PNG_COLOR_MASK_PALETTE
1056                     PNG_COLOR_MASK_COLOR
1057                     PNG_COLOR_MASK_ALPHA
1058
1059    interlace_type - (PNG_INTERLACE_NONE or
1060                     PNG_INTERLACE_ADAM7)
1061
1062    compression_type - (must be PNG_COMPRESSION_TYPE_BASE
1063                     for PNG 1.0)
1064
1065    filter_method  - (must be PNG_FILTER_TYPE_BASE
1066                     for PNG 1.0, and can also be
1067                     PNG_INTRAPIXEL_DIFFERENCING if
1068                     the PNG datastream is embedded in
1069                     a MNG-1.0 datastream)
1070
1071    Any or all of interlace_type, compression_type, or
1072    filter_method can be NULL if you are
1073    not interested in their values.
1074
1075    Note that png_get_IHDR() returns 32-bit data into
1076    the application's width and height variables.
1077    This is an unsafe situation if these are 16-bit
1078    variables.  In such situations, the
1079    png_get_image_width() and png_get_image_height()
1080    functions described below are safer.
1081
1082    width            = png_get_image_width(png_ptr,
1083                         info_ptr);
1084
1085    height           = png_get_image_height(png_ptr,
1086                         info_ptr);
1087
1088    bit_depth        = png_get_bit_depth(png_ptr,
1089                         info_ptr);
1090
1091    color_type       = png_get_color_type(png_ptr,
1092                         info_ptr);
1093
1094    interlace_type   = png_get_interlace_type(png_ptr,
1095                         info_ptr);
1096
1097    compression_type = png_get_compression_type(png_ptr,
1098                         info_ptr);
1099
1100    filter_method    = png_get_filter_type(png_ptr,
1101                         info_ptr);
1102
1103    channels = png_get_channels(png_ptr, info_ptr);
1104
1105    channels       - number of channels of info for the
1106                     color type (valid values are 1 (GRAY,
1107                     PALETTE), 2 (GRAY_ALPHA), 3 (RGB),
1108                     4 (RGB_ALPHA or RGB + filler byte))
1109
1110    rowbytes = png_get_rowbytes(png_ptr, info_ptr);
1111
1112    rowbytes       - number of bytes needed to hold a row
1113
1114    signature = png_get_signature(png_ptr, info_ptr);
1115
1116    signature      - holds the signature read from the
1117                     file (if any).  The data is kept in
1118                     the same offset it would be if the
1119                     whole signature were read (i.e. if an
1120                     application had already read in 4
1121                     bytes of signature before starting
1122                     libpng, the remaining 4 bytes would
1123                     be in signature[4] through signature[7]
1124                     (see png_set_sig_bytes())).
1125
1126These are also important, but their validity depends on whether the chunk
1127has been read.  The png_get_valid(png_ptr, info_ptr, PNG_INFO_<chunk>) and
1128png_get_<chunk>(png_ptr, info_ptr, ...) functions return non-zero if the
1129data has been read, or zero if it is missing.  The parameters to the
1130png_get_<chunk> are set directly if they are simple data types, or a
1131pointer into the info_ptr is returned for any complex types.
1132
1133The colorspace data from gAMA, cHRM, sRGB, iCCP, and sBIT chunks
1134is simply returned to give the application information about how the
1135image was encoded.  Libpng itself only does transformations using the file
1136gamma when combining semitransparent pixels with the background color.
1137
1138    png_get_PLTE(png_ptr, info_ptr, &palette,
1139                     &num_palette);
1140
1141    palette        - the palette for the file
1142                     (array of png_color)
1143
1144    num_palette    - number of entries in the palette
1145
1146    png_get_gAMA(png_ptr, info_ptr, &file_gamma);
1147    png_get_gAMA_fixed(png_ptr, info_ptr, &int_file_gamma);
1148
1149    file_gamma     - the gamma at which the file was
1150                     written (PNG_INFO_gAMA)
1151
1152    int_file_gamma - 100,000 times the gamma at which the
1153                     file is written
1154
1155    png_get_cHRM(png_ptr, info_ptr,  &white_x, &white_y, &red_x, &red_y,
1156                     &green_x, &green_y, &blue_x, &blue_y)
1157    png_get_cHRM_XYZ(png_ptr, info_ptr, &red_X, &red_Y, &red_Z, &green_X,
1158                     &green_Y, &green_Z, &blue_X, &blue_Y, &blue_Z)
1159    png_get_cHRM_fixed(png_ptr, info_ptr, &int_white_x, &int_white_y,
1160                     &int_red_x, &int_red_y, &int_green_x, &int_green_y,
1161                     &int_blue_x, &int_blue_y)
1162    png_get_cHRM_XYZ_fixed(png_ptr, info_ptr, &int_red_X, &int_red_Y,
1163                     &int_red_Z, &int_green_X, &int_green_Y, &int_green_Z,
1164                     &int_blue_X, &int_blue_Y, &int_blue_Z)
1165
1166    {white,red,green,blue}_{x,y}
1167                     A color space encoding specified using the chromaticities
1168                     of the end points and the white point. (PNG_INFO_cHRM)
1169
1170    {red,green,blue}_{X,Y,Z}
1171                     A color space encoding specified using the encoding end
1172                     points - the CIE tristimulus specification of the intended
1173                     color of the red, green and blue channels in the PNG RGB
1174                     data.  The white point is simply the sum of the three end
1175                     points. (PNG_INFO_cHRM)
1176
1177    png_get_sRGB(png_ptr, info_ptr, &srgb_intent);
1178
1179    file_srgb_intent - the rendering intent (PNG_INFO_sRGB)
1180                     The presence of the sRGB chunk
1181                     means that the pixel data is in the
1182                     sRGB color space.  This chunk also
1183                     implies specific values of gAMA and
1184                     cHRM.
1185
1186    png_get_iCCP(png_ptr, info_ptr, &name,
1187       &compression_type, &profile, &proflen);
1188
1189    name             - The profile name.
1190
1191    compression_type - The compression type; always
1192                       PNG_COMPRESSION_TYPE_BASE for PNG 1.0.
1193                       You may give NULL to this argument to
1194                       ignore it.
1195
1196    profile          - International Color Consortium color
1197                       profile data. May contain NULs.
1198
1199    proflen          - length of profile data in bytes.
1200
1201    png_get_sBIT(png_ptr, info_ptr, &sig_bit);
1202
1203    sig_bit        - the number of significant bits for
1204                     (PNG_INFO_sBIT) each of the gray,
1205                     red, green, and blue channels,
1206                     whichever are appropriate for the
1207                     given color type (png_color_16)
1208
1209    png_get_tRNS(png_ptr, info_ptr, &trans_alpha,
1210                     &num_trans, &trans_color);
1211
1212    trans_alpha    - array of alpha (transparency)
1213                     entries for palette (PNG_INFO_tRNS)
1214
1215    num_trans      - number of transparent entries
1216                     (PNG_INFO_tRNS)
1217
1218    trans_color    - graylevel or color sample values of
1219                     the single transparent color for
1220                     non-paletted images (PNG_INFO_tRNS)
1221
1222    png_get_hIST(png_ptr, info_ptr, &hist);
1223                     (PNG_INFO_hIST)
1224
1225    hist           - histogram of palette (array of
1226                     png_uint_16)
1227
1228    png_get_tIME(png_ptr, info_ptr, &mod_time);
1229
1230    mod_time       - time image was last modified
1231                    (PNG_VALID_tIME)
1232
1233    png_get_bKGD(png_ptr, info_ptr, &background);
1234
1235    background     - background color (of type
1236                     png_color_16p) (PNG_VALID_bKGD)
1237                     valid 16-bit red, green and blue
1238                     values, regardless of color_type
1239
1240    num_comments   = png_get_text(png_ptr, info_ptr,
1241                     &text_ptr, &num_text);
1242
1243    num_comments   - number of comments
1244
1245    text_ptr       - array of png_text holding image
1246                     comments
1247
1248    text_ptr[i].compression - type of compression used
1249                 on "text" PNG_TEXT_COMPRESSION_NONE
1250                           PNG_TEXT_COMPRESSION_zTXt
1251                           PNG_ITXT_COMPRESSION_NONE
1252                           PNG_ITXT_COMPRESSION_zTXt
1253
1254    text_ptr[i].key   - keyword for comment.  Must contain
1255                         1-79 characters.
1256
1257    text_ptr[i].text  - text comments for current
1258                         keyword.  Can be empty.
1259
1260    text_ptr[i].text_length - length of text string,
1261                 after decompression, 0 for iTXt
1262
1263    text_ptr[i].itxt_length - length of itxt string,
1264                 after decompression, 0 for tEXt/zTXt
1265
1266    text_ptr[i].lang  - language of comment (empty
1267                         string for unknown).
1268
1269    text_ptr[i].lang_key  - keyword in UTF-8
1270                         (empty string for unknown).
1271
1272    Note that the itxt_length, lang, and lang_key
1273    members of the text_ptr structure only exist when the
1274    library is built with iTXt chunk support.  Prior to
1275    libpng-1.4.0 the library was built by default without
1276    iTXt support. Also note that when iTXt is supported,
1277    they contain NULL pointers when the "compression"
1278    field contains PNG_TEXT_COMPRESSION_NONE or
1279    PNG_TEXT_COMPRESSION_zTXt.
1280
1281    num_text       - number of comments (same as
1282                     num_comments; you can put NULL here
1283                     to avoid the duplication)
1284
1285    Note while png_set_text() will accept text, language,
1286    and translated keywords that can be NULL pointers, the
1287    structure returned by png_get_text will always contain
1288    regular zero-terminated C strings.  They might be
1289    empty strings but they will never be NULL pointers.
1290
1291    num_spalettes = png_get_sPLT(png_ptr, info_ptr,
1292       &palette_ptr);
1293
1294    num_spalettes  - number of sPLT chunks read.
1295
1296    palette_ptr    - array of palette structures holding
1297                     contents of one or more sPLT chunks
1298                     read.
1299
1300    png_get_oFFs(png_ptr, info_ptr, &offset_x, &offset_y,
1301       &unit_type);
1302
1303    offset_x       - positive offset from the left edge
1304                     of the screen (can be negative)
1305
1306    offset_y       - positive offset from the top edge
1307                     of the screen (can be negative)
1308
1309    unit_type      - PNG_OFFSET_PIXEL, PNG_OFFSET_MICROMETER
1310
1311    png_get_pHYs(png_ptr, info_ptr, &res_x, &res_y,
1312       &unit_type);
1313
1314    res_x          - pixels/unit physical resolution in
1315                     x direction
1316
1317    res_y          - pixels/unit physical resolution in
1318                     x direction
1319
1320    unit_type      - PNG_RESOLUTION_UNKNOWN,
1321                     PNG_RESOLUTION_METER
1322
1323    png_get_sCAL(png_ptr, info_ptr, &unit, &width,
1324       &height)
1325
1326    unit        - physical scale units (an integer)
1327
1328    width       - width of a pixel in physical scale units
1329
1330    height      - height of a pixel in physical scale units
1331                 (width and height are doubles)
1332
1333    png_get_sCAL_s(png_ptr, info_ptr, &unit, &width,
1334       &height)
1335
1336    unit        - physical scale units (an integer)
1337
1338    width       - width of a pixel in physical scale units
1339                  (expressed as a string)
1340
1341    height      - height of a pixel in physical scale units
1342                 (width and height are strings like "2.54")
1343
1344    num_unknown_chunks = png_get_unknown_chunks(png_ptr,
1345       info_ptr, &unknowns)
1346
1347    unknowns          - array of png_unknown_chunk
1348                        structures holding unknown chunks
1349
1350    unknowns[i].name  - name of unknown chunk
1351
1352    unknowns[i].data  - data of unknown chunk
1353
1354    unknowns[i].size  - size of unknown chunk's data
1355
1356    unknowns[i].location - position of chunk in file
1357
1358    The value of "i" corresponds to the order in which the
1359    chunks were read from the PNG file or inserted with the
1360    png_set_unknown_chunks() function.
1361
1362    The value of "location" is a bitwise "or" of
1363
1364         PNG_HAVE_IHDR  (0x01)
1365         PNG_HAVE_PLTE  (0x02)
1366         PNG_AFTER_IDAT (0x08)
1367
1368The data from the pHYs chunk can be retrieved in several convenient
1369forms:
1370
1371    res_x = png_get_x_pixels_per_meter(png_ptr,
1372       info_ptr)
1373
1374    res_y = png_get_y_pixels_per_meter(png_ptr,
1375       info_ptr)
1376
1377    res_x_and_y = png_get_pixels_per_meter(png_ptr,
1378       info_ptr)
1379
1380    res_x = png_get_x_pixels_per_inch(png_ptr,
1381       info_ptr)
1382
1383    res_y = png_get_y_pixels_per_inch(png_ptr,
1384       info_ptr)
1385
1386    res_x_and_y = png_get_pixels_per_inch(png_ptr,
1387       info_ptr)
1388
1389    aspect_ratio = png_get_pixel_aspect_ratio(png_ptr,
1390       info_ptr)
1391
1392    Each of these returns 0 [signifying "unknown"] if
1393       the data is not present or if res_x is 0;
1394       res_x_and_y is 0 if res_x != res_y
1395
1396    Note that because of the way the resolutions are
1397       stored internally, the inch conversions won't
1398       come out to exactly even number.  For example,
1399       72 dpi is stored as 0.28346 pixels/meter, and
1400       when this is retrieved it is 71.9988 dpi, so
1401       be sure to round the returned value appropriately
1402       if you want to display a reasonable-looking result.
1403
1404The data from the oFFs chunk can be retrieved in several convenient
1405forms:
1406
1407    x_offset = png_get_x_offset_microns(png_ptr, info_ptr);
1408
1409    y_offset = png_get_y_offset_microns(png_ptr, info_ptr);
1410
1411    x_offset = png_get_x_offset_inches(png_ptr, info_ptr);
1412
1413    y_offset = png_get_y_offset_inches(png_ptr, info_ptr);
1414
1415    Each of these returns 0 [signifying "unknown" if both
1416       x and y are 0] if the data is not present or if the
1417       chunk is present but the unit is the pixel.  The
1418       remark about inexact inch conversions applies here
1419       as well, because a value in inches can't always be
1420       converted to microns and back without some loss
1421       of precision.
1422
1423For more information, see the
1424PNG specification for chunk contents.  Be careful with trusting
1425rowbytes, as some of the transformations could increase the space
1426needed to hold a row (expand, filler, gray_to_rgb, etc.).
1427See png_read_update_info(), below.
1428
1429A quick word about text_ptr and num_text.  PNG stores comments in
1430keyword/text pairs, one pair per chunk, with no limit on the number
1431of text chunks, and a 2^31 byte limit on their size.  While there are
1432suggested keywords, there is no requirement to restrict the use to these
1433strings.  It is strongly suggested that keywords and text be sensible
1434to humans (that's the point), so don't use abbreviations.  Non-printing
1435symbols are not allowed.  See the PNG specification for more details.
1436There is also no requirement to have text after the keyword.
1437
1438Keywords should be limited to 79 Latin-1 characters without leading or
1439trailing spaces, but non-consecutive spaces are allowed within the
1440keyword.  It is possible to have the same keyword any number of times.
1441The text_ptr is an array of png_text structures, each holding a
1442pointer to a language string, a pointer to a keyword and a pointer to
1443a text string.  The text string, language code, and translated
1444keyword may be empty or NULL pointers.  The keyword/text
1445pairs are put into the array in the order that they are received.
1446However, some or all of the text chunks may be after the image, so, to
1447make sure you have read all the text chunks, don't mess with these
1448until after you read the stuff after the image.  This will be
1449mentioned again below in the discussion that goes with png_read_end().
1450
1451Input transformations
1452
1453After you've read the header information, you can set up the library
1454to handle any special transformations of the image data.  The various
1455ways to transform the data will be described in the order that they
1456should occur.  This is important, as some of these change the color
1457type and/or bit depth of the data, and some others only work on
1458certain color types and bit depths.
1459
1460Transformations you request are ignored if they don't have any meaning for a
1461particular input data format.  However some transformations can have an effect
1462as a result of a previous transformation.  If you specify a contradictory set of
1463transformations, for example both adding and removing the alpha channel, you
1464cannot predict the final result.
1465
1466The color used for the transparency values should be supplied in the same
1467format/depth as the current image data.  It is stored in the same format/depth
1468as the image data in a tRNS chunk, so this is what libpng expects for this data.
1469
1470The color used for the background value depends on the need_expand argument as
1471described below.
1472
1473Data will be decoded into the supplied row buffers packed into bytes
1474unless the library has been told to transform it into another format.
1475For example, 4 bit/pixel paletted or grayscale data will be returned
14762 pixels/byte with the leftmost pixel in the high-order bits of the
1477byte, unless png_set_packing() is called.  8-bit RGB data will be stored
1478in RGB RGB RGB format unless png_set_filler() or png_set_add_alpha()
1479is called to insert filler bytes, either before or after each RGB triplet.
148016-bit RGB data will be returned RRGGBB RRGGBB, with the most significant
1481byte of the color value first, unless png_set_scale_16() is called to
1482transform it to regular RGB RGB triplets, or png_set_filler() or
1483png_set_add alpha() is called to insert filler bytes, either before or
1484after each RRGGBB triplet.  Similarly, 8-bit or 16-bit grayscale data can
1485be modified with png_set_filler(), png_set_add_alpha(), png_set_strip_16(),
1486or png_set_scale_16().
1487
1488The following code transforms grayscale images of less than 8 to 8 bits,
1489changes paletted images to RGB, and adds a full alpha channel if there is
1490transparency information in a tRNS chunk.  This is most useful on
1491grayscale images with bit depths of 2 or 4 or if there is a multiple-image
1492viewing application that wishes to treat all images in the same way.
1493
1494    if (color_type == PNG_COLOR_TYPE_PALETTE)
1495        png_set_palette_to_rgb(png_ptr);
1496
1497    if (png_get_valid(png_ptr, info_ptr,
1498        PNG_INFO_tRNS)) png_set_tRNS_to_alpha(png_ptr);
1499
1500    if (color_type == PNG_COLOR_TYPE_GRAY &&
1501        bit_depth < 8) png_set_expand_gray_1_2_4_to_8(png_ptr);
1502
1503The first two functions are actually aliases for png_set_expand(), added
1504in libpng version 1.0.4, with the function names expanded to improve code
1505readability.  In some future version they may actually do different
1506things.
1507
1508As of libpng version 1.2.9, png_set_expand_gray_1_2_4_to_8() was
1509added.  It expands the sample depth without changing tRNS to alpha.
1510
1511As of libpng version 1.5.2, png_set_expand_16() was added.  It behaves as
1512png_set_expand(); however, the resultant channels have 16 bits rather than 8.
1513Use this when the output color or gray channels are made linear to avoid fairly
1514severe accuracy loss.
1515
1516   if (bit_depth < 16)
1517      png_set_expand_16(png_ptr);
1518
1519PNG can have files with 16 bits per channel.  If you only can handle
15208 bits per channel, this will strip the pixels down to 8-bit.
1521
1522    if (bit_depth == 16)
1523#if PNG_LIBPNG_VER >= 10504
1524       png_set_scale_16(png_ptr);
1525#else
1526       png_set_strip_16(png_ptr);
1527#endif
1528
1529(The more accurate "png_set_scale_16()" API became available in libpng version
15301.5.4).
1531
1532If you need to process the alpha channel on the image separately from the image
1533data (for example if you convert it to a bitmap mask) it is possible to have
1534libpng strip the channel leaving just RGB or gray data:
1535
1536    if (color_type & PNG_COLOR_MASK_ALPHA)
1537       png_set_strip_alpha(png_ptr);
1538
1539If you strip the alpha channel you need to find some other way of dealing with
1540the information.  If, instead, you want to convert the image to an opaque
1541version with no alpha channel use png_set_background; see below.
1542
1543As of libpng version 1.5.2, almost all useful expansions are supported, the
1544major ommissions are conversion of grayscale to indexed images (which can be
1545done trivially in the application) and conversion of indexed to grayscale (which
1546can be done by a trivial manipulation of the palette.)
1547
1548In the following table, the 01 means grayscale with depth<8, 31 means
1549indexed with depth<8, other numerals represent the color type, "T" means
1550the tRNS chunk is present, A means an alpha channel is present, and O
1551means tRNS or alpha is present but all pixels in the image are opaque.
1552
1553  FROM  01  31   0  0T  0O   2  2T  2O   3  3T  3O  4A  4O  6A  6O
1554   TO
1555   01    -  [G]  -   -   -   -   -   -   -   -   -   -   -   -   -
1556   31   [Q]  Q  [Q] [Q] [Q]  Q   Q   Q   Q   Q   Q  [Q] [Q]  Q   Q
1557    0    1   G   +   .   .   G   G   G   G   G   G   B   B  GB  GB
1558   0T    lt  Gt  t   +   .   Gt  G   G   Gt  G   G   Bt  Bt GBt GBt
1559   0O    lt  Gt  t   .   +   Gt  Gt  G   Gt  Gt  G   Bt  Bt GBt GBt
1560    2    C   P   C   C   C   +   .   .   C   -   -  CB  CB   B   B
1561   2T    Ct  -   Ct  C   C   t   +   t   -   -   -  CBt CBt  Bt  Bt
1562   2O    Ct  -   Ct  C   C   t   t   +   -   -   -  CBt CBt  Bt  Bt
1563    3   [Q]  p  [Q] [Q] [Q]  Q   Q   Q   +   .   .  [Q] [Q]  Q   Q
1564   3T   [Qt] p  [Qt][Q] [Q]  Qt  Qt  Qt  t   +   t  [Qt][Qt] Qt  Qt
1565   3O   [Qt] p  [Qt][Q] [Q]  Qt  Qt  Qt  t   t   +  [Qt][Qt] Qt  Qt
1566   4A    lA  G   A   T   T   GA  GT  GT  GA  GT  GT  +   BA  G  GBA
1567   4O    lA GBA  A   T   T   GA  GT  GT  GA  GT  GT  BA  +  GBA  G
1568   6A    CA  PA  CA  C   C   A   T  tT   PA  P   P   C  CBA  +   BA
1569   6O    CA PBA  CA  C   C   A  tT   T   PA  P   P  CBA  C   BA  +
1570
1571Within the matrix,
1572     "+" identifies entries where 'from' and 'to' are the same.
1573     "-" means the transformation is not supported.
1574     "." means nothing is necessary (a tRNS chunk can just be ignored).
1575     "t" means the transformation is obtained by png_set_tRNS.
1576     "A" means the transformation is obtained by png_set_add_alpha().
1577     "X" means the transformation is obtained by png_set_expand().
1578     "1" means the transformation is obtained by
1579         png_set_expand_gray_1_2_4_to_8() (and by png_set_expand() if there
1580         is no transparency in the original or the final format).
1581     "C" means the transformation is obtained by png_set_gray_to_rgb().
1582     "G" means the transformation is obtained by png_set_rgb_to_gray().
1583     "P" means the transformation is obtained by
1584         png_set_expand_palette_to_rgb().
1585     "p" means the transformation is obtained by png_set_packing().
1586     "Q" means the transformation is obtained by png_set_quantize().
1587     "T" means the transformation is obtained by png_set_tRNS_to_alpha().
1588     "B" means the transformation is obtained by png_set_background(), or
1589         png_strip_alpha().
1590
1591When an entry has multiple transforms listed all are required to cause the
1592right overall transformation.  When two transforms are separated by a comma
1593either will do the job.  When transforms are enclosed in [] the transform should
1594do the job but this is currently unimplemented - a different format will result
1595if the suggested transformations are used.
1596
1597In PNG files, the alpha channel in an image
1598is the level of opacity.  If you need the alpha channel in an image to
1599be the level of transparency instead of opacity, you can invert the
1600alpha channel (or the tRNS chunk data) after it's read, so that 0 is
1601fully opaque and 255 (in 8-bit or paletted images) or 65535 (in 16-bit
1602images) is fully transparent, with
1603
1604    png_set_invert_alpha(png_ptr);
1605
1606PNG files pack pixels of bit depths 1, 2, and 4 into bytes as small as
1607they can, resulting in, for example, 8 pixels per byte for 1 bit
1608files.  This code expands to 1 pixel per byte without changing the
1609values of the pixels:
1610
1611    if (bit_depth < 8)
1612       png_set_packing(png_ptr);
1613
1614PNG files have possible bit depths of 1, 2, 4, 8, and 16.  All pixels
1615stored in a PNG image have been "scaled" or "shifted" up to the next
1616higher possible bit depth (e.g. from 5 bits/sample in the range [0,31]
1617to 8 bits/sample in the range [0, 255]).  However, it is also possible
1618to convert the PNG pixel data back to the original bit depth of the
1619image.  This call reduces the pixels back down to the original bit depth:
1620
1621    png_color_8p sig_bit;
1622
1623    if (png_get_sBIT(png_ptr, info_ptr, &sig_bit))
1624       png_set_shift(png_ptr, sig_bit);
1625
1626PNG files store 3-color pixels in red, green, blue order.  This code
1627changes the storage of the pixels to blue, green, red:
1628
1629    if (color_type == PNG_COLOR_TYPE_RGB ||
1630        color_type == PNG_COLOR_TYPE_RGB_ALPHA)
1631       png_set_bgr(png_ptr);
1632
1633PNG files store RGB pixels packed into 3 or 6 bytes. This code expands them
1634into 4 or 8 bytes for windowing systems that need them in this format:
1635
1636    if (color_type == PNG_COLOR_TYPE_RGB)
1637       png_set_filler(png_ptr, filler, PNG_FILLER_BEFORE);
1638
1639where "filler" is the 8 or 16-bit number to fill with, and the location is
1640either PNG_FILLER_BEFORE or PNG_FILLER_AFTER, depending upon whether
1641you want the filler before the RGB or after.  This transformation
1642does not affect images that already have full alpha channels.  To add an
1643opaque alpha channel, use filler=0xff or 0xffff and PNG_FILLER_AFTER which
1644will generate RGBA pixels.
1645
1646Note that png_set_filler() does not change the color type.  If you want
1647to do that, you can add a true alpha channel with
1648
1649    if (color_type == PNG_COLOR_TYPE_RGB ||
1650       color_type == PNG_COLOR_TYPE_GRAY)
1651       png_set_add_alpha(png_ptr, filler, PNG_FILLER_AFTER);
1652
1653where "filler" contains the alpha value to assign to each pixel.
1654This function was added in libpng-1.2.7.
1655
1656If you are reading an image with an alpha channel, and you need the
1657data as ARGB instead of the normal PNG format RGBA:
1658
1659    if (color_type == PNG_COLOR_TYPE_RGB_ALPHA)
1660       png_set_swap_alpha(png_ptr);
1661
1662For some uses, you may want a grayscale image to be represented as
1663RGB.  This code will do that conversion:
1664
1665    if (color_type == PNG_COLOR_TYPE_GRAY ||
1666        color_type == PNG_COLOR_TYPE_GRAY_ALPHA)
1667       png_set_gray_to_rgb(png_ptr);
1668
1669Conversely, you can convert an RGB or RGBA image to grayscale or grayscale
1670with alpha.
1671
1672    if (color_type == PNG_COLOR_TYPE_RGB ||
1673        color_type == PNG_COLOR_TYPE_RGB_ALPHA)
1674       png_set_rgb_to_gray(png_ptr, error_action, double red_weight,
1675          double green_weight);
1676
1677    error_action = 1: silently do the conversion
1678
1679    error_action = 2: issue a warning if the original
1680                      image has any pixel where
1681                      red != green or red != blue
1682
1683    error_action = 3: issue an error and abort the
1684                      conversion if the original
1685                      image has any pixel where
1686                      red != green or red != blue
1687
1688    red_weight:       weight of red component
1689
1690    green_weight:     weight of green component
1691                      If either weight is negative, default
1692                      weights are used.
1693
1694In the corresponding fixed point API the red_weight and green_weight values are
1695simply scaled by 100,000:
1696
1697    png_set_rgb_to_gray(png_ptr, error_action, png_fixed_point red_weight,
1698       png_fixed_point green_weight);
1699
1700If you have set error_action = 1 or 2, you can
1701later check whether the image really was gray, after processing
1702the image rows, with the png_get_rgb_to_gray_status(png_ptr) function.
1703It will return a png_byte that is zero if the image was gray or
17041 if there were any non-gray pixels.  Background and sBIT data
1705will be silently converted to grayscale, using the green channel
1706data for sBIT, regardless of the error_action setting.
1707
1708The default values come from the PNG file cHRM chunk if present; otherwise, the
1709defaults correspond to the ITU-R recommendation 709, and also the sRGB color
1710space, as recommended in the Charles Poynton's Colour FAQ,
1711<http://www.poynton.com/>, in section 9:
1712
1713   <http://www.poynton.com/notes/colour_and_gamma/ColorFAQ.html#RTFToC9>
1714
1715    Y = 0.2126 * R + 0.7152 * G + 0.0722 * B
1716
1717Previous versions of this document, 1998 through 2002, recommended a slightly
1718different formula:
1719
1720    Y = 0.212671 * R + 0.715160 * G + 0.072169 * B
1721
1722Libpng uses an integer approximation:
1723
1724    Y = (6968 * R + 23434 * G + 2366 * B)/32768
1725
1726The calculation is done in a linear colorspace, if the image gamma
1727can be determined.
1728
1729The png_set_background() function has been described already; it tells libpng to
1730composite images with alpha or simple transparency against the supplied
1731background color.  For compatibility with versions of libpng earlier than
1732libpng-1.5.4 it is recommended that you call the function after reading the file
1733header, even if you don't want to use the color in a bKGD chunk, if one exists.
1734
1735If the PNG file contains a bKGD chunk (PNG_INFO_bKGD valid),
1736you may use this color, or supply another color more suitable for
1737the current display (e.g., the background color from a web page).  You
1738need to tell libpng how the color is represented, both the format of the
1739component values in the color (the number of bits) and the gamma encoding of the
1740color.  The function takes two arguments, background_gamma_mode and need_expand
1741to convey this information; however, only two combinations are likely to be
1742useful:
1743
1744    png_color_16 my_background;
1745    png_color_16p image_background;
1746
1747    if (png_get_bKGD(png_ptr, info_ptr, &image_background))
1748       png_set_background(png_ptr, image_background,
1749           PNG_BACKGROUND_GAMMA_FILE, 1/*needs to be expanded*/, 1);
1750    else
1751       png_set_background(png_ptr, &my_background,
1752           PNG_BACKGROUND_GAMMA_SCREEN, 0/*do not expand*/, 1);
1753
1754The second call was described above - my_background is in the format of the
1755final, display, output produced by libpng.  Because you now know the format of
1756the PNG it is possible to avoid the need to choose either 8-bit or 16-bit
1757output and to retain palette images (the palette colors will be modified
1758appropriately and the tRNS chunk removed.)  However, if you are doing this,
1759take great care not to ask for transformations without checking first that
1760they apply!
1761
1762In the first call the background color has the original bit depth and color type
1763of the PNG file.  So, for palette images the color is supplied as a palette
1764index and for low bit greyscale images the color is a reduced bit value in
1765image_background->gray.
1766
1767If you didn't call png_set_gamma() before reading the file header, for example
1768if you need your code to remain compatible with older versions of libpng prior
1769to libpng-1.5.4, this is the place to call it.
1770
1771Do not call it if you called png_set_alpha_mode(); doing so will damage the
1772settings put in place by png_set_alpha_mode().  (If png_set_alpha_mode() is
1773supported then you can certainly do png_set_gamma() before reading the PNG
1774header.)
1775
1776This API unconditionally sets the screen and file gamma values, so it will
1777override the value in the PNG file unless it is called before the PNG file
1778reading starts.  For this reason you must always call it with the PNG file
1779value when you call it in this position:
1780
1781   if (png_get_gAMA(png_ptr, info_ptr, &file_gamma))
1782      png_set_gamma(png_ptr, screen_gamma, file_gamma);
1783
1784   else
1785      png_set_gamma(png_ptr, screen_gamma, 0.45455);
1786
1787If you need to reduce an RGB file to a paletted file, or if a paletted
1788file has more entries then will fit on your screen, png_set_quantize()
1789will do that.  Note that this is a simple match quantization that merely
1790finds the closest color available.  This should work fairly well with
1791optimized palettes, but fairly badly with linear color cubes.  If you
1792pass a palette that is larger than maximum_colors, the file will
1793reduce the number of colors in the palette so it will fit into
1794maximum_colors.  If there is a histogram, libpng will use it to make
1795more intelligent choices when reducing the palette.  If there is no
1796histogram, it may not do as good a job.
1797
1798   if (color_type & PNG_COLOR_MASK_COLOR)
1799   {
1800      if (png_get_valid(png_ptr, info_ptr,
1801          PNG_INFO_PLTE))
1802      {
1803         png_uint_16p histogram = NULL;
1804
1805         png_get_hIST(png_ptr, info_ptr,
1806             &histogram);
1807         png_set_quantize(png_ptr, palette, num_palette,
1808            max_screen_colors, histogram, 1);
1809      }
1810
1811      else
1812      {
1813         png_color std_color_cube[MAX_SCREEN_COLORS] =
1814            { ... colors ... };
1815
1816         png_set_quantize(png_ptr, std_color_cube,
1817            MAX_SCREEN_COLORS, MAX_SCREEN_COLORS,
1818            NULL,0);
1819      }
1820   }
1821
1822PNG files describe monochrome as black being zero and white being one.
1823The following code will reverse this (make black be one and white be
1824zero):
1825
1826   if (bit_depth == 1 && color_type == PNG_COLOR_TYPE_GRAY)
1827      png_set_invert_mono(png_ptr);
1828
1829This function can also be used to invert grayscale and gray-alpha images:
1830
1831   if (color_type == PNG_COLOR_TYPE_GRAY ||
1832       color_type == PNG_COLOR_TYPE_GRAY_ALPHA)
1833      png_set_invert_mono(png_ptr);
1834
1835PNG files store 16-bit pixels in network byte order (big-endian,
1836ie. most significant bits first).  This code changes the storage to the
1837other way (little-endian, i.e. least significant bits first, the
1838way PCs store them):
1839
1840    if (bit_depth == 16)
1841       png_set_swap(png_ptr);
1842
1843If you are using packed-pixel images (1, 2, or 4 bits/pixel), and you
1844need to change the order the pixels are packed into bytes, you can use:
1845
1846    if (bit_depth < 8)
1847       png_set_packswap(png_ptr);
1848
1849Finally, you can write your own transformation function if none of
1850the existing ones meets your needs.  This is done by setting a callback
1851with
1852
1853    png_set_read_user_transform_fn(png_ptr,
1854        read_transform_fn);
1855
1856You must supply the function
1857
1858    void read_transform_fn(png_structp png_ptr, png_row_infop
1859        row_info, png_bytep data)
1860
1861See pngtest.c for a working example.  Your function will be called
1862after all of the other transformations have been processed.  Take care with
1863interlaced images if you do the interlace yourself - the width of the row is the
1864width in 'row_info', not the overall image width.
1865
1866If supported, libpng provides two information routines that you can use to find
1867where you are in processing the image:
1868
1869   png_get_current_pass_number(png_structp png_ptr);
1870   png_get_current_row_number(png_structp png_ptr);
1871
1872Don't try using these outside a transform callback - firstly they are only
1873supported if user transforms are supported, secondly they may well return
1874unexpected results unless the row is actually being processed at the moment they
1875are called.
1876
1877With interlaced
1878images the value returned is the row in the input sub-image image.  Use
1879PNG_ROW_FROM_PASS_ROW(row, pass) and PNG_COL_FROM_PASS_COL(col, pass) to
1880find the output pixel (x,y) given an interlaced sub-image pixel (row,col,pass).
1881
1882The discussion of interlace handling above contains more information on how to
1883use these values.
1884
1885You can also set up a pointer to a user structure for use by your
1886callback function, and you can inform libpng that your transform
1887function will change the number of channels or bit depth with the
1888function
1889
1890    png_set_user_transform_info(png_ptr, user_ptr,
1891        user_depth, user_channels);
1892
1893The user's application, not libpng, is responsible for allocating and
1894freeing any memory required for the user structure.
1895
1896You can retrieve the pointer via the function
1897png_get_user_transform_ptr().  For example:
1898
1899    voidp read_user_transform_ptr =
1900        png_get_user_transform_ptr(png_ptr);
1901
1902The last thing to handle is interlacing; this is covered in detail below,
1903but you must call the function here if you want libpng to handle expansion
1904of the interlaced image.
1905
1906    number_of_passes = png_set_interlace_handling(png_ptr);
1907
1908After setting the transformations, libpng can update your png_info
1909structure to reflect any transformations you've requested with this
1910call.
1911
1912    png_read_update_info(png_ptr, info_ptr);
1913
1914This is most useful to update the info structure's rowbytes
1915field so you can use it to allocate your image memory.  This function
1916will also update your palette with the correct screen_gamma and
1917background if these have been given with the calls above.  You may
1918only call png_read_update_info() once with a particular info_ptr.
1919
1920After you call png_read_update_info(), you can allocate any
1921memory you need to hold the image.  The row data is simply
1922raw byte data for all forms of images.  As the actual allocation
1923varies among applications, no example will be given.  If you
1924are allocating one large chunk, you will need to build an
1925array of pointers to each row, as it will be needed for some
1926of the functions below.
1927
1928Remember: Before you call png_read_update_info(), the png_get_*()
1929functions return the values corresponding to the original PNG image.
1930After you call png_read_update_info the values refer to the image
1931that libpng will output.  Consequently you must call all the png_set_
1932functions before you call png_read_update_info().  This is particularly
1933important for png_set_interlace_handling() - if you are going to call
1934png_read_update_info() you must call png_set_interlace_handling() before
1935it unless you want to receive interlaced output.
1936
1937Reading image data
1938
1939After you've allocated memory, you can read the image data.
1940The simplest way to do this is in one function call.  If you are
1941allocating enough memory to hold the whole image, you can just
1942call png_read_image() and libpng will read in all the image data
1943and put it in the memory area supplied.  You will need to pass in
1944an array of pointers to each row.
1945
1946This function automatically handles interlacing, so you don't
1947need to call png_set_interlace_handling() (unless you call
1948png_read_update_info()) or call this function multiple times, or any
1949of that other stuff necessary with png_read_rows().
1950
1951   png_read_image(png_ptr, row_pointers);
1952
1953where row_pointers is:
1954
1955   png_bytep row_pointers[height];
1956
1957You can point to void or char or whatever you use for pixels.
1958
1959If you don't want to read in the whole image at once, you can
1960use png_read_rows() instead.  If there is no interlacing (check
1961interlace_type == PNG_INTERLACE_NONE), this is simple:
1962
1963    png_read_rows(png_ptr, row_pointers, NULL,
1964        number_of_rows);
1965
1966where row_pointers is the same as in the png_read_image() call.
1967
1968If you are doing this just one row at a time, you can do this with
1969a single row_pointer instead of an array of row_pointers:
1970
1971    png_bytep row_pointer = row;
1972    png_read_row(png_ptr, row_pointer, NULL);
1973
1974If the file is interlaced (interlace_type != 0 in the IHDR chunk), things
1975get somewhat harder.  The only current (PNG Specification version 1.2)
1976interlacing type for PNG is (interlace_type == PNG_INTERLACE_ADAM7);
1977a somewhat complicated 2D interlace scheme, known as Adam7, that
1978breaks down an image into seven smaller images of varying size, based
1979on an 8x8 grid.  This number is defined (from libpng 1.5) as
1980PNG_INTERLACE_ADAM7_PASSES in png.h
1981
1982libpng can fill out those images or it can give them to you "as is".
1983It is almost always better to have libpng handle the interlacing for you.
1984If you want the images filled out, there are two ways to do that.  The one
1985mentioned in the PNG specification is to expand each pixel to cover
1986those pixels that have not been read yet (the "rectangle" method).
1987This results in a blocky image for the first pass, which gradually
1988smooths out as more pixels are read.  The other method is the "sparkle"
1989method, where pixels are drawn only in their final locations, with the
1990rest of the image remaining whatever colors they were initialized to
1991before the start of the read.  The first method usually looks better,
1992but tends to be slower, as there are more pixels to put in the rows.
1993
1994If, as is likely, you want libpng to expand the images, call this before
1995calling png_start_read_image() or png_read_update_info():
1996
1997    if (interlace_type == PNG_INTERLACE_ADAM7)
1998       number_of_passes
1999           = png_set_interlace_handling(png_ptr);
2000
2001This will return the number of passes needed.  Currently, this is seven,
2002but may change if another interlace type is added.  This function can be
2003called even if the file is not interlaced, where it will return one pass.
2004You then need to read the whole image 'number_of_passes' times.  Each time
2005will distribute the pixels from the current pass to the correct place in
2006the output image, so you need to supply the same rows to png_read_rows in
2007each pass.
2008
2009If you are not going to display the image after each pass, but are
2010going to wait until the entire image is read in, use the sparkle
2011effect.  This effect is faster and the end result of either method
2012is exactly the same.  If you are planning on displaying the image
2013after each pass, the "rectangle" effect is generally considered the
2014better looking one.
2015
2016If you only want the "sparkle" effect, just call png_read_rows() as
2017normal, with the third parameter NULL.  Make sure you make pass over
2018the image number_of_passes times, and you don't change the data in the
2019rows between calls.  You can change the locations of the data, just
2020not the data.  Each pass only writes the pixels appropriate for that
2021pass, and assumes the data from previous passes is still valid.
2022
2023    png_read_rows(png_ptr, row_pointers, NULL,
2024        number_of_rows);
2025
2026If you only want the first effect (the rectangles), do the same as
2027before except pass the row buffer in the third parameter, and leave
2028the second parameter NULL.
2029
2030    png_read_rows(png_ptr, NULL, row_pointers,
2031        number_of_rows);
2032
2033If you don't want libpng to handle the interlacing details, just call
2034png_read_rows() PNG_INTERLACE_ADAM7_PASSES times to read in all the images.
2035Each of the images is a valid image by itself; however, you will almost
2036certainly need to distribute the pixels from each sub-image to the
2037correct place.  This is where everything gets very tricky.
2038
2039If you want to retrieve the separate images you must pass the correct
2040number of rows to each successive call of png_read_rows().  The calculation
2041gets pretty complicated for small images, where some sub-images may
2042not even exist because either their width or height ends up zero.
2043libpng provides two macros to help you in 1.5 and later versions:
2044
2045   png_uint_32 width = PNG_PASS_COLS(image_width, pass_number);
2046   png_uint_32 height = PNG_PASS_ROWS(image_height, pass_number);
2047
2048Respectively these tell you the width and height of the sub-image
2049corresponding to the numbered pass.  'pass' is in in the range 0 to 6 -
2050this can be confusing because the specification refers to the same passes
2051as 1 to 7!  Be careful, you must check both the width and height before
2052calling png_read_rows() and not call it for that pass if either is zero.
2053
2054You can, of course, read each sub-image row by row.  If you want to
2055produce optimal code to make a pixel-by-pixel transformation of an
2056interlaced image this is the best approach; read each row of each pass,
2057transform it, and write it out to a new interlaced image.
2058
2059If you want to de-interlace the image yourself libpng provides further
2060macros to help that tell you where to place the pixels in the output image.
2061Because the interlacing scheme is rectangular - sub-image pixels are always
2062arranged on a rectangular grid - all you need to know for each pass is the
2063starting column and row in the output image of the first pixel plus the
2064spacing between each pixel.  As of libpng 1.5 there are four macros to
2065retrieve this information:
2066
2067   png_uint_32 x = PNG_PASS_START_COL(pass);
2068   png_uint_32 y = PNG_PASS_START_ROW(pass);
2069   png_uint_32 xStep = 1U << PNG_PASS_COL_SHIFT(pass);
2070   png_uint_32 yStep = 1U << PNG_PASS_ROW_SHIFT(pass);
2071
2072These allow you to write the obvious loop:
2073
2074   png_uint_32 input_y = 0;
2075   png_uint_32 output_y = PNG_PASS_START_ROW(pass);
2076
2077   while (output_y < output_image_height)
2078   {
2079      png_uint_32 input_x = 0;
2080      png_uint_32 output_x = PNG_PASS_START_COL(pass);
2081
2082      while (output_x < output_image_width)
2083      {
2084         image[output_y][output_x] =
2085             subimage[pass][input_y][input_x++];
2086
2087         output_x += xStep;
2088      }
2089
2090      ++input_y;
2091      output_y += yStep;
2092   }
2093
2094Notice that the steps between successive output rows and columns are
2095returned as shifts.  This is possible because the pixels in the subimages
2096are always a power of 2 apart - 1, 2, 4 or 8 pixels - in the original
2097image.  In practice you may need to directly calculate the output coordinate
2098given an input coordinate.  libpng provides two further macros for this
2099purpose:
2100
2101   png_uint_32 output_x = PNG_COL_FROM_PASS_COL(input_x, pass);
2102   png_uint_32 output_y = PNG_ROW_FROM_PASS_ROW(input_y, pass);
2103
2104Finally a pair of macros are provided to tell you if a particular image
2105row or column appears in a given pass:
2106
2107   int col_in_pass = PNG_COL_IN_INTERLACE_PASS(output_x, pass);
2108   int row_in_pass = PNG_ROW_IN_INTERLACE_PASS(output_y, pass);
2109
2110Bear in mind that you will probably also need to check the width and height
2111of the pass in addition to the above to be sure the pass even exists!
2112
2113With any luck you are convinced by now that you don't want to do your own
2114interlace handling.  In reality normally the only good reason for doing this
2115is if you are processing PNG files on a pixel-by-pixel basis and don't want
2116to load the whole file into memory when it is interlaced.
2117
2118libpng includes a test program, pngvalid, that illustrates reading and
2119writing of interlaced images.  If you can't get interlacing to work in your
2120code and don't want to leave it to libpng (the recommended approach), see
2121how pngvalid.c does it.
2122
2123Finishing a sequential read
2124
2125After you are finished reading the image through the
2126low-level interface, you can finish reading the file.  If you are
2127interested in comments or time, which may be stored either before or
2128after the image data, you should pass the separate png_info struct if
2129you want to keep the comments from before and after the image
2130separate.
2131
2132    png_infop end_info = png_create_info_struct(png_ptr);
2133
2134    if (!end_info)
2135    {
2136       png_destroy_read_struct(&png_ptr, &info_ptr,
2137           (png_infopp)NULL);
2138       return (ERROR);
2139    }
2140
2141   png_read_end(png_ptr, end_info);
2142
2143If you are not interested, you should still call png_read_end()
2144but you can pass NULL, avoiding the need to create an end_info structure.
2145
2146   png_read_end(png_ptr, (png_infop)NULL);
2147
2148If you don't call png_read_end(), then your file pointer will be
2149left pointing to the first chunk after the last IDAT, which is probably
2150not what you want if you expect to read something beyond the end of
2151the PNG datastream.
2152
2153When you are done, you can free all memory allocated by libpng like this:
2154
2155   png_destroy_read_struct(&png_ptr, &info_ptr,
2156       &end_info);
2157
2158or, if you didn't create an end_info structure,
2159
2160   png_destroy_read_struct(&png_ptr, &info_ptr,
2161       (png_infopp)NULL);
2162
2163It is also possible to individually free the info_ptr members that
2164point to libpng-allocated storage with the following function:
2165
2166    png_free_data(png_ptr, info_ptr, mask, seq)
2167
2168    mask - identifies data to be freed, a mask
2169           containing the bitwise OR of one or
2170           more of
2171             PNG_FREE_PLTE, PNG_FREE_TRNS,
2172             PNG_FREE_HIST, PNG_FREE_ICCP,
2173             PNG_FREE_PCAL, PNG_FREE_ROWS,
2174             PNG_FREE_SCAL, PNG_FREE_SPLT,
2175             PNG_FREE_TEXT, PNG_FREE_UNKN,
2176           or simply PNG_FREE_ALL
2177
2178    seq  - sequence number of item to be freed
2179           (-1 for all items)
2180
2181This function may be safely called when the relevant storage has
2182already been freed, or has not yet been allocated, or was allocated
2183by the user and not by libpng,  and will in those cases do nothing.
2184The "seq" parameter is ignored if only one item of the selected data
2185type, such as PLTE, is allowed.  If "seq" is not -1, and multiple items
2186are allowed for the data type identified in the mask, such as text or
2187sPLT, only the n'th item in the structure is freed, where n is "seq".
2188
2189The default behavior is only to free data that was allocated internally
2190by libpng.  This can be changed, so that libpng will not free the data,
2191or so that it will free data that was allocated by the user with png_malloc()
2192or png_zalloc() and passed in via a png_set_*() function, with
2193
2194    png_data_freer(png_ptr, info_ptr, freer, mask)
2195
2196    freer  - one of
2197               PNG_DESTROY_WILL_FREE_DATA
2198               PNG_SET_WILL_FREE_DATA
2199               PNG_USER_WILL_FREE_DATA
2200
2201    mask   - which data elements are affected
2202             same choices as in png_free_data()
2203
2204This function only affects data that has already been allocated.
2205You can call this function after reading the PNG data but before calling
2206any png_set_*() functions, to control whether the user or the png_set_*()
2207function is responsible for freeing any existing data that might be present,
2208and again after the png_set_*() functions to control whether the user
2209or png_destroy_*() is supposed to free the data.  When the user assumes
2210responsibility for libpng-allocated data, the application must use
2211png_free() to free it, and when the user transfers responsibility to libpng
2212for data that the user has allocated, the user must have used png_malloc()
2213or png_zalloc() to allocate it.
2214
2215If you allocated your row_pointers in a single block, as suggested above in
2216the description of the high level read interface, you must not transfer
2217responsibility for freeing it to the png_set_rows or png_read_destroy function,
2218because they would also try to free the individual row_pointers[i].
2219
2220If you allocated text_ptr.text, text_ptr.lang, and text_ptr.translated_keyword
2221separately, do not transfer responsibility for freeing text_ptr to libpng,
2222because when libpng fills a png_text structure it combines these members with
2223the key member, and png_free_data() will free only text_ptr.key.  Similarly,
2224if you transfer responsibility for free'ing text_ptr from libpng to your
2225application, your application must not separately free those members.
2226
2227The png_free_data() function will turn off the "valid" flag for anything
2228it frees.  If you need to turn the flag off for a chunk that was freed by
2229your application instead of by libpng, you can use
2230
2231    png_set_invalid(png_ptr, info_ptr, mask);
2232
2233    mask - identifies the chunks to be made invalid,
2234           containing the bitwise OR of one or
2235           more of
2236             PNG_INFO_gAMA, PNG_INFO_sBIT,
2237             PNG_INFO_cHRM, PNG_INFO_PLTE,
2238             PNG_INFO_tRNS, PNG_INFO_bKGD,
2239             PNG_INFO_hIST, PNG_INFO_pHYs,
2240             PNG_INFO_oFFs, PNG_INFO_tIME,
2241             PNG_INFO_pCAL, PNG_INFO_sRGB,
2242             PNG_INFO_iCCP, PNG_INFO_sPLT,
2243             PNG_INFO_sCAL, PNG_INFO_IDAT
2244
2245For a more compact example of reading a PNG image, see the file example.c.
2246
2247Reading PNG files progressively
2248
2249The progressive reader is slightly different then the non-progressive
2250reader.  Instead of calling png_read_info(), png_read_rows(), and
2251png_read_end(), you make one call to png_process_data(), which calls
2252callbacks when it has the info, a row, or the end of the image.  You
2253set up these callbacks with png_set_progressive_read_fn().  You don't
2254have to worry about the input/output functions of libpng, as you are
2255giving the library the data directly in png_process_data().  I will
2256assume that you have read the section on reading PNG files above,
2257so I will only highlight the differences (although I will show
2258all of the code).
2259
2260png_structp png_ptr;
2261png_infop info_ptr;
2262
2263 /*  An example code fragment of how you would
2264     initialize the progressive reader in your
2265     application. */
2266 int
2267 initialize_png_reader()
2268 {
2269    png_ptr = png_create_read_struct
2270        (PNG_LIBPNG_VER_STRING, (png_voidp)user_error_ptr,
2271         user_error_fn, user_warning_fn);
2272
2273    if (!png_ptr)
2274        return (ERROR);
2275
2276    info_ptr = png_create_info_struct(png_ptr);
2277
2278    if (!info_ptr)
2279    {
2280       png_destroy_read_struct(&png_ptr,
2281          (png_infopp)NULL, (png_infopp)NULL);
2282       return (ERROR);
2283    }
2284
2285    if (setjmp(png_jmpbuf(png_ptr)))
2286    {
2287       png_destroy_read_struct(&png_ptr, &info_ptr,
2288          (png_infopp)NULL);
2289       return (ERROR);
2290    }
2291
2292    /* This one's new.  You can provide functions
2293       to be called when the header info is valid,
2294       when each row is completed, and when the image
2295       is finished.  If you aren't using all functions,
2296       you can specify NULL parameters.  Even when all
2297       three functions are NULL, you need to call
2298       png_set_progressive_read_fn().  You can use
2299       any struct as the user_ptr (cast to a void pointer
2300       for the function call), and retrieve the pointer
2301       from inside the callbacks using the function
2302
2303          png_get_progressive_ptr(png_ptr);
2304
2305       which will return a void pointer, which you have
2306       to cast appropriately.
2307     */
2308    png_set_progressive_read_fn(png_ptr, (void *)user_ptr,
2309        info_callback, row_callback, end_callback);
2310
2311    return 0;
2312 }
2313
2314 /* A code fragment that you call as you receive blocks
2315   of data */
2316 int
2317 process_data(png_bytep buffer, png_uint_32 length)
2318 {
2319    if (setjmp(png_jmpbuf(png_ptr)))
2320    {
2321       png_destroy_read_struct(&png_ptr, &info_ptr,
2322           (png_infopp)NULL);
2323       return (ERROR);
2324    }
2325
2326    /* This one's new also.  Simply give it a chunk
2327       of data from the file stream (in order, of
2328       course).  On machines with segmented memory
2329       models machines, don't give it any more than
2330       64K.  The library seems to run fine with sizes
2331       of 4K. Although you can give it much less if
2332       necessary (I assume you can give it chunks of
2333       1 byte, I haven't tried less then 256 bytes
2334       yet).  When this function returns, you may
2335       want to display any rows that were generated
2336       in the row callback if you don't already do
2337       so there.
2338     */
2339    png_process_data(png_ptr, info_ptr, buffer, length);
2340
2341    /* At this point you can call png_process_data_skip if
2342       you want to handle data the library will skip yourself;
2343       it simply returns the number of bytes to skip (and stops
2344       libpng skipping that number of bytes on the next
2345       png_process_data call).
2346    return 0;
2347 }
2348
2349 /* This function is called (as set by
2350    png_set_progressive_read_fn() above) when enough data
2351    has been supplied so all of the header has been
2352    read.
2353 */
2354 void
2355 info_callback(png_structp png_ptr, png_infop info)
2356 {
2357    /* Do any setup here, including setting any of
2358       the transformations mentioned in the Reading
2359       PNG files section.  For now, you _must_ call
2360       either png_start_read_image() or
2361       png_read_update_info() after all the
2362       transformations are set (even if you don't set
2363       any).  You may start getting rows before
2364       png_process_data() returns, so this is your
2365       last chance to prepare for that.
2366
2367       This is where you turn on interlace handling,
2368       assuming you don't want to do it yourself.
2369
2370       If you need to you can stop the processing of
2371       your original input data at this point by calling
2372       png_process_data_pause.  This returns the number
2373       of unprocessed bytes from the last png_process_data
2374       call - it is up to you to ensure that the next call
2375       sees these bytes again.  If you don't want to bother
2376       with this you can get libpng to cache the unread
2377       bytes by setting the 'save' parameter (see png.h) but
2378       then libpng will have to copy the data internally.
2379     */
2380 }
2381
2382 /* This function is called when each row of image
2383    data is complete */
2384 void
2385 row_callback(png_structp png_ptr, png_bytep new_row,
2386    png_uint_32 row_num, int pass)
2387 {
2388    /* If the image is interlaced, and you turned
2389       on the interlace handler, this function will
2390       be called for every row in every pass.  Some
2391       of these rows will not be changed from the
2392       previous pass.  When the row is not changed,
2393       the new_row variable will be NULL.  The rows
2394       and passes are called in order, so you don't
2395       really need the row_num and pass, but I'm
2396       supplying them because it may make your life
2397       easier.
2398
2399       If you did not turn on interlace handling then
2400       the callback is called for each row of each
2401       sub-image when the image is interlaced.  In this
2402       case 'row_num' is the row in the sub-image, not
2403       the row in the output image as it is in all other
2404       cases.
2405
2406       For the non-NULL rows of interlaced images when
2407       you have switched on libpng interlace handling,
2408       you must call png_progressive_combine_row()
2409       passing in the row and the old row.  You can
2410       call this function for NULL rows (it will just
2411       return) and for non-interlaced images (it just
2412       does the memcpy for you) if it will make the
2413       code easier.  Thus, you can just do this for
2414       all cases if you switch on interlace handling;
2415     */
2416
2417        png_progressive_combine_row(png_ptr, old_row,
2418          new_row);
2419
2420    /* where old_row is what was displayed for
2421       previously for the row.  Note that the first
2422       pass (pass == 0, really) will completely cover
2423       the old row, so the rows do not have to be
2424       initialized.  After the first pass (and only
2425       for interlaced images), you will have to pass
2426       the current row, and the function will combine
2427       the old row and the new row.
2428
2429       You can also call png_process_data_pause in this
2430       callback - see above.
2431    */
2432 }
2433
2434 void
2435 end_callback(png_structp png_ptr, png_infop info)
2436 {
2437    /* This function is called after the whole image
2438       has been read, including any chunks after the
2439       image (up to and including the IEND).  You
2440       will usually have the same info chunk as you
2441       had in the header, although some data may have
2442       been added to the comments and time fields.
2443
2444       Most people won't do much here, perhaps setting
2445       a flag that marks the image as finished.
2446     */
2447 }
2448
2449
2450
2451IV. Writing
2452
2453Much of this is very similar to reading.  However, everything of
2454importance is repeated here, so you won't have to constantly look
2455back up in the reading section to understand writing.
2456
2457Setup
2458
2459You will want to do the I/O initialization before you get into libpng,
2460so if it doesn't work, you don't have anything to undo. If you are not
2461using the standard I/O functions, you will need to replace them with
2462custom writing functions.  See the discussion under Customizing libpng.
2463
2464    FILE *fp = fopen(file_name, "wb");
2465
2466    if (!fp)
2467       return (ERROR);
2468
2469Next, png_struct and png_info need to be allocated and initialized.
2470As these can be both relatively large, you may not want to store these
2471on the stack, unless you have stack space to spare.  Of course, you
2472will want to check if they return NULL.  If you are also reading,
2473you won't want to name your read structure and your write structure
2474both "png_ptr"; you can call them anything you like, such as
2475"read_ptr" and "write_ptr".  Look at pngtest.c, for example.
2476
2477    png_structp png_ptr = png_create_write_struct
2478       (PNG_LIBPNG_VER_STRING, (png_voidp)user_error_ptr,
2479        user_error_fn, user_warning_fn);
2480
2481    if (!png_ptr)
2482       return (ERROR);
2483
2484    png_infop info_ptr = png_create_info_struct(png_ptr);
2485    if (!info_ptr)
2486    {
2487       png_destroy_write_struct(&png_ptr,
2488           (png_infopp)NULL);
2489       return (ERROR);
2490    }
2491
2492If you want to use your own memory allocation routines,
2493define PNG_USER_MEM_SUPPORTED and use
2494png_create_write_struct_2() instead of png_create_write_struct():
2495
2496    png_structp png_ptr = png_create_write_struct_2
2497       (PNG_LIBPNG_VER_STRING, (png_voidp)user_error_ptr,
2498        user_error_fn, user_warning_fn, (png_voidp)
2499        user_mem_ptr, user_malloc_fn, user_free_fn);
2500
2501After you have these structures, you will need to set up the
2502error handling.  When libpng encounters an error, it expects to
2503longjmp() back to your routine.  Therefore, you will need to call
2504setjmp() and pass the png_jmpbuf(png_ptr).  If you
2505write the file from different routines, you will need to update
2506the png_jmpbuf(png_ptr) every time you enter a new routine that will
2507call a png_*() function.  See your documentation of setjmp/longjmp
2508for your compiler for more information on setjmp/longjmp.  See
2509the discussion on libpng error handling in the Customizing Libpng
2510section below for more information on the libpng error handling.
2511
2512    if (setjmp(png_jmpbuf(png_ptr)))
2513    {
2514    png_destroy_write_struct(&png_ptr, &info_ptr);
2515       fclose(fp);
2516       return (ERROR);
2517    }
2518    ...
2519    return;
2520
2521If you would rather avoid the complexity of setjmp/longjmp issues,
2522you can compile libpng with PNG_NO_SETJMP, in which case
2523errors will result in a call to PNG_ABORT() which defaults to abort().
2524
2525You can #define PNG_ABORT() to a function that does something
2526more useful than abort(), as long as your function does not
2527return.
2528
2529Now you need to set up the output code.  The default for libpng is to
2530use the C function fwrite().  If you use this, you will need to pass a
2531valid FILE * in the function png_init_io().  Be sure that the file is
2532opened in binary mode.  Again, if you wish to handle writing data in
2533another way, see the discussion on libpng I/O handling in the Customizing
2534Libpng section below.
2535
2536    png_init_io(png_ptr, fp);
2537
2538If you are embedding your PNG into a datastream such as MNG, and don't
2539want libpng to write the 8-byte signature, or if you have already
2540written the signature in your application, use
2541
2542    png_set_sig_bytes(png_ptr, 8);
2543
2544to inform libpng that it should not write a signature.
2545
2546Write callbacks
2547
2548At this point, you can set up a callback function that will be
2549called after each row has been written, which you can use to control
2550a progress meter or the like.  It's demonstrated in pngtest.c.
2551You must supply a function
2552
2553    void write_row_callback(png_structp png_ptr, png_uint_32 row,
2554       int pass);
2555    {
2556      /* put your code here */
2557    }
2558
2559(You can give it another name that you like instead of "write_row_callback")
2560
2561To inform libpng about your function, use
2562
2563    png_set_write_status_fn(png_ptr, write_row_callback);
2564
2565When this function is called the row has already been completely processed and
2566it has also been written out.  The 'row' and 'pass' refer to the next row to be
2567handled.  For the
2568non-interlaced case the row that was just handled is simply one less than the
2569passed in row number, and pass will always be 0.  For the interlaced case the
2570same applies unless the row value is 0, in which case the row just handled was
2571the last one from one of the preceding passes.  Because interlacing may skip a
2572pass you cannot be sure that the preceding pass is just 'pass-1', if you really
2573need to know what the last pass is record (row,pass) from the callback and use
2574the last recorded value each time.
2575
2576As with the user transform you can find the output row using the
2577PNG_ROW_FROM_PASS_ROW macro.
2578
2579You now have the option of modifying how the compression library will
2580run.  The following functions are mainly for testing, but may be useful
2581in some cases, like if you need to write PNG files extremely fast and
2582are willing to give up some compression, or if you want to get the
2583maximum possible compression at the expense of slower writing.  If you
2584have no special needs in this area, let the library do what it wants by
2585not calling this function at all, as it has been tuned to deliver a good
2586speed/compression ratio. The second parameter to png_set_filter() is
2587the filter method, for which the only valid values are 0 (as of the
2588July 1999 PNG specification, version 1.2) or 64 (if you are writing
2589a PNG datastream that is to be embedded in a MNG datastream).  The third
2590parameter is a flag that indicates which filter type(s) are to be tested
2591for each scanline.  See the PNG specification for details on the specific
2592filter types.
2593
2594
2595    /* turn on or off filtering, and/or choose
2596       specific filters.  You can use either a single
2597       PNG_FILTER_VALUE_NAME or the bitwise OR of one
2598       or more PNG_FILTER_NAME masks.
2599     */
2600    png_set_filter(png_ptr, 0,
2601       PNG_FILTER_NONE  | PNG_FILTER_VALUE_NONE |
2602       PNG_FILTER_SUB   | PNG_FILTER_VALUE_SUB  |
2603       PNG_FILTER_UP    | PNG_FILTER_VALUE_UP   |
2604       PNG_FILTER_AVG   | PNG_FILTER_VALUE_AVG  |
2605       PNG_FILTER_PAETH | PNG_FILTER_VALUE_PAETH|
2606       PNG_ALL_FILTERS);
2607
2608If an application wants to start and stop using particular filters during
2609compression, it should start out with all of the filters (to ensure that
2610the previous row of pixels will be stored in case it's needed later),
2611and then add and remove them after the start of compression.
2612
2613If you are writing a PNG datastream that is to be embedded in a MNG
2614datastream, the second parameter can be either 0 or 64.
2615
2616The png_set_compression_*() functions interface to the zlib compression
2617library, and should mostly be ignored unless you really know what you are
2618doing.  The only generally useful call is png_set_compression_level()
2619which changes how much time zlib spends on trying to compress the image
2620data.  See the Compression Library (zlib.h and algorithm.txt, distributed
2621with zlib) for details on the compression levels.
2622
2623    #include zlib.h
2624
2625    /* Set the zlib compression level */
2626    png_set_compression_level(png_ptr,
2627        Z_BEST_COMPRESSION);
2628
2629    /* Set other zlib parameters for compressing IDAT */
2630    png_set_compression_mem_level(png_ptr, 8);
2631    png_set_compression_strategy(png_ptr,
2632        Z_DEFAULT_STRATEGY);
2633    png_set_compression_window_bits(png_ptr, 15);
2634    png_set_compression_method(png_ptr, 8);
2635    png_set_compression_buffer_size(png_ptr, 8192)
2636
2637    /* Set zlib parameters for text compression
2638     * If you don't call these, the parameters
2639     * fall back on those defined for IDAT chunks
2640     */
2641    png_set_text_compression_mem_level(png_ptr, 8);
2642    png_set_text_compression_strategy(png_ptr,
2643        Z_DEFAULT_STRATEGY);
2644    png_set_text_compression_window_bits(png_ptr, 15);
2645    png_set_text_compression_method(png_ptr, 8);
2646
2647Setting the contents of info for output
2648
2649You now need to fill in the png_info structure with all the data you
2650wish to write before the actual image.  Note that the only thing you
2651are allowed to write after the image is the text chunks and the time
2652chunk (as of PNG Specification 1.2, anyway).  See png_write_end() and
2653the latest PNG specification for more information on that.  If you
2654wish to write them before the image, fill them in now, and flag that
2655data as being valid.  If you want to wait until after the data, don't
2656fill them until png_write_end().  For all the fields in png_info and
2657their data types, see png.h.  For explanations of what the fields
2658contain, see the PNG specification.
2659
2660Some of the more important parts of the png_info are:
2661
2662    png_set_IHDR(png_ptr, info_ptr, width, height,
2663       bit_depth, color_type, interlace_type,
2664       compression_type, filter_method)
2665
2666    width          - holds the width of the image
2667                     in pixels (up to 2^31).
2668
2669    height         - holds the height of the image
2670                     in pixels (up to 2^31).
2671
2672    bit_depth      - holds the bit depth of one of the
2673                     image channels.
2674                     (valid values are 1, 2, 4, 8, 16
2675                     and depend also on the
2676                     color_type.  See also significant
2677                     bits (sBIT) below).
2678
2679    color_type     - describes which color/alpha
2680                     channels are present.
2681                     PNG_COLOR_TYPE_GRAY
2682                        (bit depths 1, 2, 4, 8, 16)
2683                     PNG_COLOR_TYPE_GRAY_ALPHA
2684                        (bit depths 8, 16)
2685                     PNG_COLOR_TYPE_PALETTE
2686                        (bit depths 1, 2, 4, 8)
2687                     PNG_COLOR_TYPE_RGB
2688                        (bit_depths 8, 16)
2689                     PNG_COLOR_TYPE_RGB_ALPHA
2690                        (bit_depths 8, 16)
2691
2692                     PNG_COLOR_MASK_PALETTE
2693                     PNG_COLOR_MASK_COLOR
2694                     PNG_COLOR_MASK_ALPHA
2695
2696    interlace_type - PNG_INTERLACE_NONE or
2697                     PNG_INTERLACE_ADAM7
2698
2699    compression_type - (must be
2700                     PNG_COMPRESSION_TYPE_DEFAULT)
2701
2702    filter_method  - (must be PNG_FILTER_TYPE_DEFAULT
2703                     or, if you are writing a PNG to
2704                     be embedded in a MNG datastream,
2705                     can also be
2706                     PNG_INTRAPIXEL_DIFFERENCING)
2707
2708If you call png_set_IHDR(), the call must appear before any of the
2709other png_set_*() functions, because they might require access to some of
2710the IHDR settings.  The remaining png_set_*() functions can be called
2711in any order.
2712
2713If you wish, you can reset the compression_type, interlace_type, or
2714filter_method later by calling png_set_IHDR() again; if you do this, the
2715width, height, bit_depth, and color_type must be the same in each call.
2716
2717    png_set_PLTE(png_ptr, info_ptr, palette,
2718       num_palette);
2719
2720    palette        - the palette for the file
2721                     (array of png_color)
2722    num_palette    - number of entries in the palette
2723
2724    png_set_gAMA(png_ptr, info_ptr, file_gamma);
2725    png_set_gAMA_fixed(png_ptr, info_ptr, int_file_gamma);
2726
2727    file_gamma     - the gamma at which the image was
2728                     created (PNG_INFO_gAMA)
2729
2730    int_file_gamma - 100,000 times the gamma at which
2731                     the image was created
2732
2733    png_set_cHRM(png_ptr, info_ptr,  white_x, white_y, red_x, red_y,
2734                     green_x, green_y, blue_x, blue_y)
2735    png_set_cHRM_XYZ(png_ptr, info_ptr, red_X, red_Y, red_Z, green_X,
2736                     green_Y, green_Z, blue_X, blue_Y, blue_Z)
2737    png_set_cHRM_fixed(png_ptr, info_ptr, int_white_x, int_white_y,
2738                     int_red_x, int_red_y, int_green_x, int_green_y,
2739                     int_blue_x, int_blue_y)
2740    png_set_cHRM_XYZ_fixed(png_ptr, info_ptr, int_red_X, int_red_Y,
2741                     int_red_Z, int_green_X, int_green_Y, int_green_Z,
2742                     int_blue_X, int_blue_Y, int_blue_Z)
2743
2744    {white,red,green,blue}_{x,y}
2745                     A color space encoding specified using the chromaticities
2746                     of the end points and the white point.
2747
2748    {red,green,blue}_{X,Y,Z}
2749                     A color space encoding specified using the encoding end
2750                     points - the CIE tristimulus specification of the intended
2751                     color of the red, green and blue channels in the PNG RGB
2752                     data.  The white point is simply the sum of the three end
2753                     points.
2754
2755    png_set_sRGB(png_ptr, info_ptr, srgb_intent);
2756
2757    srgb_intent    - the rendering intent
2758                     (PNG_INFO_sRGB) The presence of
2759                     the sRGB chunk means that the pixel
2760                     data is in the sRGB color space.
2761                     This chunk also implies specific
2762                     values of gAMA and cHRM.  Rendering
2763                     intent is the CSS-1 property that
2764                     has been defined by the International
2765                     Color Consortium
2766                     (http://www.color.org).
2767                     It can be one of
2768                     PNG_sRGB_INTENT_SATURATION,
2769                     PNG_sRGB_INTENT_PERCEPTUAL,
2770                     PNG_sRGB_INTENT_ABSOLUTE, or
2771                     PNG_sRGB_INTENT_RELATIVE.
2772
2773
2774    png_set_sRGB_gAMA_and_cHRM(png_ptr, info_ptr,
2775       srgb_intent);
2776
2777    srgb_intent    - the rendering intent
2778                     (PNG_INFO_sRGB) The presence of the
2779                     sRGB chunk means that the pixel
2780                     data is in the sRGB color space.
2781                     This function also causes gAMA and
2782                     cHRM chunks with the specific values
2783                     that are consistent with sRGB to be
2784                     written.
2785
2786    png_set_iCCP(png_ptr, info_ptr, name, compression_type,
2787                       profile, proflen);
2788
2789    name             - The profile name.
2790
2791    compression_type - The compression type; always
2792                       PNG_COMPRESSION_TYPE_BASE for PNG 1.0.
2793                       You may give NULL to this argument to
2794                       ignore it.
2795
2796    profile          - International Color Consortium color
2797                       profile data. May contain NULs.
2798
2799    proflen          - length of profile data in bytes.
2800
2801    png_set_sBIT(png_ptr, info_ptr, sig_bit);
2802
2803    sig_bit        - the number of significant bits for
2804                     (PNG_INFO_sBIT) each of the gray, red,
2805                     green, and blue channels, whichever are
2806                     appropriate for the given color type
2807                     (png_color_16)
2808
2809    png_set_tRNS(png_ptr, info_ptr, trans_alpha,
2810       num_trans, trans_color);
2811
2812    trans_alpha    - array of alpha (transparency)
2813                     entries for palette (PNG_INFO_tRNS)
2814
2815    num_trans      - number of transparent entries
2816                     (PNG_INFO_tRNS)
2817
2818    trans_color    - graylevel or color sample values
2819                     (in order red, green, blue) of the
2820                     single transparent color for
2821                     non-paletted images (PNG_INFO_tRNS)
2822
2823    png_set_hIST(png_ptr, info_ptr, hist);
2824
2825    hist           - histogram of palette (array of
2826                     png_uint_16) (PNG_INFO_hIST)
2827
2828    png_set_tIME(png_ptr, info_ptr, mod_time);
2829
2830    mod_time       - time image was last modified
2831                     (PNG_VALID_tIME)
2832
2833    png_set_bKGD(png_ptr, info_ptr, background);
2834
2835    background     - background color (of type
2836                     png_color_16p) (PNG_VALID_bKGD)
2837
2838    png_set_text(png_ptr, info_ptr, text_ptr, num_text);
2839
2840    text_ptr       - array of png_text holding image
2841                     comments
2842
2843    text_ptr[i].compression - type of compression used
2844                 on "text" PNG_TEXT_COMPRESSION_NONE
2845                           PNG_TEXT_COMPRESSION_zTXt
2846                           PNG_ITXT_COMPRESSION_NONE
2847                           PNG_ITXT_COMPRESSION_zTXt
2848    text_ptr[i].key   - keyword for comment.  Must contain
2849                 1-79 characters.
2850    text_ptr[i].text  - text comments for current
2851                         keyword.  Can be NULL or empty.
2852    text_ptr[i].text_length - length of text string,
2853                 after decompression, 0 for iTXt
2854    text_ptr[i].itxt_length - length of itxt string,
2855                 after decompression, 0 for tEXt/zTXt
2856    text_ptr[i].lang  - language of comment (NULL or
2857                         empty for unknown).
2858    text_ptr[i].translated_keyword  - keyword in UTF-8 (NULL
2859                         or empty for unknown).
2860
2861    Note that the itxt_length, lang, and lang_key
2862    members of the text_ptr structure only exist when the
2863    library is built with iTXt chunk support.  Prior to
2864    libpng-1.4.0 the library was built by default without
2865    iTXt support. Also note that when iTXt is supported,
2866    they contain NULL pointers when the "compression"
2867    field contains PNG_TEXT_COMPRESSION_NONE or
2868    PNG_TEXT_COMPRESSION_zTXt.
2869
2870    num_text       - number of comments
2871
2872    png_set_sPLT(png_ptr, info_ptr, &palette_ptr,
2873       num_spalettes);
2874
2875    palette_ptr    - array of png_sPLT_struct structures
2876                     to be added to the list of palettes
2877                     in the info structure.
2878    num_spalettes  - number of palette structures to be
2879                     added.
2880
2881    png_set_oFFs(png_ptr, info_ptr, offset_x, offset_y,
2882        unit_type);
2883
2884    offset_x  - positive offset from the left
2885                     edge of the screen
2886
2887    offset_y  - positive offset from the top
2888                     edge of the screen
2889
2890    unit_type - PNG_OFFSET_PIXEL, PNG_OFFSET_MICROMETER
2891
2892    png_set_pHYs(png_ptr, info_ptr, res_x, res_y,
2893        unit_type);
2894
2895    res_x       - pixels/unit physical resolution
2896                  in x direction
2897
2898    res_y       - pixels/unit physical resolution
2899                  in y direction
2900
2901    unit_type   - PNG_RESOLUTION_UNKNOWN,
2902                  PNG_RESOLUTION_METER
2903
2904    png_set_sCAL(png_ptr, info_ptr, unit, width, height)
2905
2906    unit        - physical scale units (an integer)
2907
2908    width       - width of a pixel in physical scale units
2909
2910    height      - height of a pixel in physical scale units
2911                  (width and height are doubles)
2912
2913    png_set_sCAL_s(png_ptr, info_ptr, unit, width, height)
2914
2915    unit        - physical scale units (an integer)
2916
2917    width       - width of a pixel in physical scale units
2918                  expressed as a string
2919
2920    height      - height of a pixel in physical scale units
2921                 (width and height are strings like "2.54")
2922
2923    png_set_unknown_chunks(png_ptr, info_ptr, &unknowns,
2924       num_unknowns)
2925
2926    unknowns          - array of png_unknown_chunk
2927                        structures holding unknown chunks
2928    unknowns[i].name  - name of unknown chunk
2929    unknowns[i].data  - data of unknown chunk
2930    unknowns[i].size  - size of unknown chunk's data
2931    unknowns[i].location - position to write chunk in file
2932                           0: do not write chunk
2933                           PNG_HAVE_IHDR: before PLTE
2934                           PNG_HAVE_PLTE: before IDAT
2935                           PNG_AFTER_IDAT: after IDAT
2936
2937The "location" member is set automatically according to
2938what part of the output file has already been written.
2939You can change its value after calling png_set_unknown_chunks()
2940as demonstrated in pngtest.c.  Within each of the "locations",
2941the chunks are sequenced according to their position in the
2942structure (that is, the value of "i", which is the order in which
2943the chunk was either read from the input file or defined with
2944png_set_unknown_chunks).
2945
2946A quick word about text and num_text.  text is an array of png_text
2947structures.  num_text is the number of valid structures in the array.
2948Each png_text structure holds a language code, a keyword, a text value,
2949and a compression type.
2950
2951The compression types have the same valid numbers as the compression
2952types of the image data.  Currently, the only valid number is zero.
2953However, you can store text either compressed or uncompressed, unlike
2954images, which always have to be compressed.  So if you don't want the
2955text compressed, set the compression type to PNG_TEXT_COMPRESSION_NONE.
2956Because tEXt and zTXt chunks don't have a language field, if you
2957specify PNG_TEXT_COMPRESSION_NONE or PNG_TEXT_COMPRESSION_zTXt
2958any language code or translated keyword will not be written out.
2959
2960Until text gets around a few hundred bytes, it is not worth compressing it.
2961After the text has been written out to the file, the compression type
2962is set to PNG_TEXT_COMPRESSION_NONE_WR or PNG_TEXT_COMPRESSION_zTXt_WR,
2963so that it isn't written out again at the end (in case you are calling
2964png_write_end() with the same struct).
2965
2966The keywords that are given in the PNG Specification are:
2967
2968    Title            Short (one line) title or
2969                     caption for image
2970
2971    Author           Name of image's creator
2972
2973    Description      Description of image (possibly long)
2974
2975    Copyright        Copyright notice
2976
2977    Creation Time    Time of original image creation
2978                     (usually RFC 1123 format, see below)
2979
2980    Software         Software used to create the image
2981
2982    Disclaimer       Legal disclaimer
2983
2984    Warning          Warning of nature of content
2985
2986    Source           Device used to create the image
2987
2988    Comment          Miscellaneous comment; conversion
2989                     from other image format
2990
2991The keyword-text pairs work like this.  Keywords should be short
2992simple descriptions of what the comment is about.  Some typical
2993keywords are found in the PNG specification, as is some recommendations
2994on keywords.  You can repeat keywords in a file.  You can even write
2995some text before the image and some after.  For example, you may want
2996to put a description of the image before the image, but leave the
2997disclaimer until after, so viewers working over modem connections
2998don't have to wait for the disclaimer to go over the modem before
2999they start seeing the image.  Finally, keywords should be full
3000words, not abbreviations.  Keywords and text are in the ISO 8859-1
3001(Latin-1) character set (a superset of regular ASCII) and can not
3002contain NUL characters, and should not contain control or other
3003unprintable characters.  To make the comments widely readable, stick
3004with basic ASCII, and avoid machine specific character set extensions
3005like the IBM-PC character set.  The keyword must be present, but
3006you can leave off the text string on non-compressed pairs.
3007Compressed pairs must have a text string, as only the text string
3008is compressed anyway, so the compression would be meaningless.
3009
3010PNG supports modification time via the png_time structure.  Two
3011conversion routines are provided, png_convert_from_time_t() for
3012time_t and png_convert_from_struct_tm() for struct tm.  The
3013time_t routine uses gmtime().  You don't have to use either of
3014these, but if you wish to fill in the png_time structure directly,
3015you should provide the time in universal time (GMT) if possible
3016instead of your local time.  Note that the year number is the full
3017year (e.g. 1998, rather than 98 - PNG is year 2000 compliant!), and
3018that months start with 1.
3019
3020If you want to store the time of the original image creation, you should
3021use a plain tEXt chunk with the "Creation Time" keyword.  This is
3022necessary because the "creation time" of a PNG image is somewhat vague,
3023depending on whether you mean the PNG file, the time the image was
3024created in a non-PNG format, a still photo from which the image was
3025scanned, or possibly the subject matter itself.  In order to facilitate
3026machine-readable dates, it is recommended that the "Creation Time"
3027tEXt chunk use RFC 1123 format dates (e.g. "22 May 1997 18:07:10 GMT"),
3028although this isn't a requirement.  Unlike the tIME chunk, the
3029"Creation Time" tEXt chunk is not expected to be automatically changed
3030by the software.  To facilitate the use of RFC 1123 dates, a function
3031png_convert_to_rfc1123(png_timep) is provided to convert from PNG
3032time to an RFC 1123 format string.
3033
3034Writing unknown chunks
3035
3036You can use the png_set_unknown_chunks function to queue up chunks
3037for writing.  You give it a chunk name, raw data, and a size; that's
3038all there is to it.  The chunks will be written by the next following
3039png_write_info_before_PLTE, png_write_info, or png_write_end function.
3040Any chunks previously read into the info structure's unknown-chunk
3041list will also be written out in a sequence that satisfies the PNG
3042specification's ordering rules.
3043
3044The high-level write interface
3045
3046At this point there are two ways to proceed; through the high-level
3047write interface, or through a sequence of low-level write operations.
3048You can use the high-level interface if your image data is present
3049in the info structure.  All defined output
3050transformations are permitted, enabled by the following masks.
3051
3052    PNG_TRANSFORM_IDENTITY      No transformation
3053    PNG_TRANSFORM_PACKING       Pack 1, 2 and 4-bit samples
3054    PNG_TRANSFORM_PACKSWAP      Change order of packed
3055                                pixels to LSB first
3056    PNG_TRANSFORM_INVERT_MONO   Invert monochrome images
3057    PNG_TRANSFORM_SHIFT         Normalize pixels to the
3058                                sBIT depth
3059    PNG_TRANSFORM_BGR           Flip RGB to BGR, RGBA
3060                                to BGRA
3061    PNG_TRANSFORM_SWAP_ALPHA    Flip RGBA to ARGB or GA
3062                                to AG
3063    PNG_TRANSFORM_INVERT_ALPHA  Change alpha from opacity
3064                                to transparency
3065    PNG_TRANSFORM_SWAP_ENDIAN   Byte-swap 16-bit samples
3066    PNG_TRANSFORM_STRIP_FILLER        Strip out filler
3067                                      bytes (deprecated).
3068    PNG_TRANSFORM_STRIP_FILLER_BEFORE Strip out leading
3069                                      filler bytes
3070    PNG_TRANSFORM_STRIP_FILLER_AFTER  Strip out trailing
3071                                      filler bytes
3072
3073If you have valid image data in the info structure (you can use
3074png_set_rows() to put image data in the info structure), simply do this:
3075
3076    png_write_png(png_ptr, info_ptr, png_transforms, NULL)
3077
3078where png_transforms is an integer containing the bitwise OR of some set of
3079transformation flags.  This call is equivalent to png_write_info(),
3080followed the set of transformations indicated by the transform mask,
3081then png_write_image(), and finally png_write_end().
3082
3083(The final parameter of this call is not yet used.  Someday it might point
3084to transformation parameters required by some future output transform.)
3085
3086You must use png_transforms and not call any png_set_transform() functions
3087when you use png_write_png().
3088
3089The low-level write interface
3090
3091If you are going the low-level route instead, you are now ready to
3092write all the file information up to the actual image data.  You do
3093this with a call to png_write_info().
3094
3095    png_write_info(png_ptr, info_ptr);
3096
3097Note that there is one transformation you may need to do before
3098png_write_info().  In PNG files, the alpha channel in an image is the
3099level of opacity.  If your data is supplied as a level of transparency,
3100you can invert the alpha channel before you write it, so that 0 is
3101fully transparent and 255 (in 8-bit or paletted images) or 65535
3102(in 16-bit images) is fully opaque, with
3103
3104    png_set_invert_alpha(png_ptr);
3105
3106This must appear before png_write_info() instead of later with the
3107other transformations because in the case of paletted images the tRNS
3108chunk data has to be inverted before the tRNS chunk is written.  If
3109your image is not a paletted image, the tRNS data (which in such cases
3110represents a single color to be rendered as transparent) won't need to
3111be changed, and you can safely do this transformation after your
3112png_write_info() call.
3113
3114If you need to write a private chunk that you want to appear before
3115the PLTE chunk when PLTE is present, you can write the PNG info in
3116two steps, and insert code to write your own chunk between them:
3117
3118    png_write_info_before_PLTE(png_ptr, info_ptr);
3119    png_set_unknown_chunks(png_ptr, info_ptr, ...);
3120    png_write_info(png_ptr, info_ptr);
3121
3122After you've written the file information, you can set up the library
3123to handle any special transformations of the image data.  The various
3124ways to transform the data will be described in the order that they
3125should occur.  This is important, as some of these change the color
3126type and/or bit depth of the data, and some others only work on
3127certain color types and bit depths.  Even though each transformation
3128checks to see if it has data that it can do something with, you should
3129make sure to only enable a transformation if it will be valid for the
3130data.  For example, don't swap red and blue on grayscale data.
3131
3132PNG files store RGB pixels packed into 3 or 6 bytes.  This code tells
3133the library to strip input data that has 4 or 8 bytes per pixel down
3134to 3 or 6 bytes (or strip 2 or 4-byte grayscale+filler data to 1 or 2
3135bytes per pixel).
3136
3137    png_set_filler(png_ptr, 0, PNG_FILLER_BEFORE);
3138
3139where the 0 is unused, and the location is either PNG_FILLER_BEFORE or
3140PNG_FILLER_AFTER, depending upon whether the filler byte in the pixel
3141is stored XRGB or RGBX.
3142
3143PNG files pack pixels of bit depths 1, 2, and 4 into bytes as small as
3144they can, resulting in, for example, 8 pixels per byte for 1 bit files.
3145If the data is supplied at 1 pixel per byte, use this code, which will
3146correctly pack the pixels into a single byte:
3147
3148    png_set_packing(png_ptr);
3149
3150PNG files reduce possible bit depths to 1, 2, 4, 8, and 16.  If your
3151data is of another bit depth, you can write an sBIT chunk into the
3152file so that decoders can recover the original data if desired.
3153
3154    /* Set the true bit depth of the image data */
3155    if (color_type & PNG_COLOR_MASK_COLOR)
3156    {
3157       sig_bit.red = true_bit_depth;
3158       sig_bit.green = true_bit_depth;
3159       sig_bit.blue = true_bit_depth;
3160    }
3161
3162    else
3163    {
3164       sig_bit.gray = true_bit_depth;
3165    }
3166
3167    if (color_type & PNG_COLOR_MASK_ALPHA)
3168    {
3169       sig_bit.alpha = true_bit_depth;
3170    }
3171
3172    png_set_sBIT(png_ptr, info_ptr, &sig_bit);
3173
3174If the data is stored in the row buffer in a bit depth other than
3175one supported by PNG (e.g. 3 bit data in the range 0-7 for a 4-bit PNG),
3176this will scale the values to appear to be the correct bit depth as
3177is required by PNG.
3178
3179    png_set_shift(png_ptr, &sig_bit);
3180
3181PNG files store 16-bit pixels in network byte order (big-endian,
3182ie. most significant bits first).  This code would be used if they are
3183supplied the other way (little-endian, i.e. least significant bits
3184first, the way PCs store them):
3185
3186    if (bit_depth > 8)
3187       png_set_swap(png_ptr);
3188
3189If you are using packed-pixel images (1, 2, or 4 bits/pixel), and you
3190need to change the order the pixels are packed into bytes, you can use:
3191
3192    if (bit_depth < 8)
3193       png_set_packswap(png_ptr);
3194
3195PNG files store 3 color pixels in red, green, blue order.  This code
3196would be used if they are supplied as blue, green, red:
3197
3198    png_set_bgr(png_ptr);
3199
3200PNG files describe monochrome as black being zero and white being
3201one. This code would be used if the pixels are supplied with this reversed
3202(black being one and white being zero):
3203
3204    png_set_invert_mono(png_ptr);
3205
3206Finally, you can write your own transformation function if none of
3207the existing ones meets your needs.  This is done by setting a callback
3208with
3209
3210    png_set_write_user_transform_fn(png_ptr,
3211       write_transform_fn);
3212
3213You must supply the function
3214
3215    void write_transform_fn(png_structp png_ptr, png_row_infop
3216       row_info, png_bytep data)
3217
3218See pngtest.c for a working example.  Your function will be called
3219before any of the other transformations are processed.  If supported
3220libpng also supplies an information routine that may be called from
3221your callback:
3222
3223   png_get_current_row_number(png_ptr);
3224   png_get_current_pass_number(png_ptr);
3225
3226This returns the current row passed to the transform.  With interlaced
3227images the value returned is the row in the input sub-image image.  Use
3228PNG_ROW_FROM_PASS_ROW(row, pass) and PNG_COL_FROM_PASS_COL(col, pass) to
3229find the output pixel (x,y) given an interlaced sub-image pixel (row,col,pass).
3230
3231The discussion of interlace handling above contains more information on how to
3232use these values.
3233
3234You can also set up a pointer to a user structure for use by your
3235callback function.
3236
3237    png_set_user_transform_info(png_ptr, user_ptr, 0, 0);
3238
3239The user_channels and user_depth parameters of this function are ignored
3240when writing; you can set them to zero as shown.
3241
3242You can retrieve the pointer via the function png_get_user_transform_ptr().
3243For example:
3244
3245    voidp write_user_transform_ptr =
3246       png_get_user_transform_ptr(png_ptr);
3247
3248It is possible to have libpng flush any pending output, either manually,
3249or automatically after a certain number of lines have been written.  To
3250flush the output stream a single time call:
3251
3252    png_write_flush(png_ptr);
3253
3254and to have libpng flush the output stream periodically after a certain
3255number of scanlines have been written, call:
3256
3257    png_set_flush(png_ptr, nrows);
3258
3259Note that the distance between rows is from the last time png_write_flush()
3260was called, or the first row of the image if it has never been called.
3261So if you write 50 lines, and then png_set_flush 25, it will flush the
3262output on the next scanline, and every 25 lines thereafter, unless
3263png_write_flush() is called before 25 more lines have been written.
3264If nrows is too small (less than about 10 lines for a 640 pixel wide
3265RGB image) the image compression may decrease noticeably (although this
3266may be acceptable for real-time applications).  Infrequent flushing will
3267only degrade the compression performance by a few percent over images
3268that do not use flushing.
3269
3270Writing the image data
3271
3272That's it for the transformations.  Now you can write the image data.
3273The simplest way to do this is in one function call.  If you have the
3274whole image in memory, you can just call png_write_image() and libpng
3275will write the image.  You will need to pass in an array of pointers to
3276each row.  This function automatically handles interlacing, so you don't
3277need to call png_set_interlace_handling() or call this function multiple
3278times, or any of that other stuff necessary with png_write_rows().
3279
3280    png_write_image(png_ptr, row_pointers);
3281
3282where row_pointers is:
3283
3284    png_byte *row_pointers[height];
3285
3286You can point to void or char or whatever you use for pixels.
3287
3288If you don't want to write the whole image at once, you can
3289use png_write_rows() instead.  If the file is not interlaced,
3290this is simple:
3291
3292    png_write_rows(png_ptr, row_pointers,
3293       number_of_rows);
3294
3295row_pointers is the same as in the png_write_image() call.
3296
3297If you are just writing one row at a time, you can do this with
3298a single row_pointer instead of an array of row_pointers:
3299
3300    png_bytep row_pointer = row;
3301
3302    png_write_row(png_ptr, row_pointer);
3303
3304When the file is interlaced, things can get a good deal more complicated.
3305The only currently (as of the PNG Specification version 1.2, dated July
33061999) defined interlacing scheme for PNG files is the "Adam7" interlace
3307scheme, that breaks down an image into seven smaller images of varying
3308size.  libpng will build these images for you, or you can do them
3309yourself.  If you want to build them yourself, see the PNG specification
3310for details of which pixels to write when.
3311
3312If you don't want libpng to handle the interlacing details, just
3313use png_set_interlace_handling() and call png_write_rows() the
3314correct number of times to write all the sub-images
3315(png_set_interlace_handling() returns the number of sub-images.)
3316
3317If you want libpng to build the sub-images, call this before you start
3318writing any rows:
3319
3320    number_of_passes = png_set_interlace_handling(png_ptr);
3321
3322This will return the number of passes needed.  Currently, this is seven,
3323but may change if another interlace type is added.
3324
3325Then write the complete image number_of_passes times.
3326
3327    png_write_rows(png_ptr, row_pointers, number_of_rows);
3328
3329Think carefully before you write an interlaced image.  Typically code that
3330reads such images reads all the image data into memory, uncompressed, before
3331doing any processing.  Only code that can display an image on the fly can
3332take advantage of the interlacing and even then the image has to be exactly
3333the correct size for the output device, because scaling an image requires
3334adjacent pixels and these are not available until all the passes have been
3335read.
3336
3337If you do write an interlaced image you will hardly ever need to handle
3338the interlacing yourself.  Call png_set_interlace_handling() and use the
3339approach described above.
3340
3341The only time it is conceivable that you will really need to write an
3342interlaced image pass-by-pass is when you have read one pass by pass and
3343made some pixel-by-pixel transformation to it, as described in the read
3344code above.  In this case use the PNG_PASS_ROWS and PNG_PASS_COLS macros
3345to determine the size of each sub-image in turn and simply write the rows
3346you obtained from the read code.
3347
3348Finishing a sequential write
3349
3350After you are finished writing the image, you should finish writing
3351the file.  If you are interested in writing comments or time, you should
3352pass an appropriately filled png_info pointer.  If you are not interested,
3353you can pass NULL.
3354
3355    png_write_end(png_ptr, info_ptr);
3356
3357When you are done, you can free all memory used by libpng like this:
3358
3359    png_destroy_write_struct(&png_ptr, &info_ptr);
3360
3361It is also possible to individually free the info_ptr members that
3362point to libpng-allocated storage with the following function:
3363
3364    png_free_data(png_ptr, info_ptr, mask, seq)
3365
3366    mask  - identifies data to be freed, a mask
3367            containing the bitwise OR of one or
3368            more of
3369              PNG_FREE_PLTE, PNG_FREE_TRNS,
3370              PNG_FREE_HIST, PNG_FREE_ICCP,
3371              PNG_FREE_PCAL, PNG_FREE_ROWS,
3372              PNG_FREE_SCAL, PNG_FREE_SPLT,
3373              PNG_FREE_TEXT, PNG_FREE_UNKN,
3374            or simply PNG_FREE_ALL
3375
3376    seq   - sequence number of item to be freed
3377            (-1 for all items)
3378
3379This function may be safely called when the relevant storage has
3380already been freed, or has not yet been allocated, or was allocated
3381by the user  and not by libpng,  and will in those cases do nothing.
3382The "seq" parameter is ignored if only one item of the selected data
3383type, such as PLTE, is allowed.  If "seq" is not -1, and multiple items
3384are allowed for the data type identified in the mask, such as text or
3385sPLT, only the n'th item in the structure is freed, where n is "seq".
3386
3387If you allocated data such as a palette that you passed in to libpng
3388with png_set_*, you must not free it until just before the call to
3389png_destroy_write_struct().
3390
3391The default behavior is only to free data that was allocated internally
3392by libpng.  This can be changed, so that libpng will not free the data,
3393or so that it will free data that was allocated by the user with png_malloc()
3394or png_zalloc() and passed in via a png_set_*() function, with
3395
3396    png_data_freer(png_ptr, info_ptr, freer, mask)
3397
3398    freer  - one of
3399               PNG_DESTROY_WILL_FREE_DATA
3400               PNG_SET_WILL_FREE_DATA
3401               PNG_USER_WILL_FREE_DATA
3402
3403    mask   - which data elements are affected
3404             same choices as in png_free_data()
3405
3406For example, to transfer responsibility for some data from a read structure
3407to a write structure, you could use
3408
3409    png_data_freer(read_ptr, read_info_ptr,
3410       PNG_USER_WILL_FREE_DATA,
3411       PNG_FREE_PLTE|PNG_FREE_tRNS|PNG_FREE_hIST)
3412
3413    png_data_freer(write_ptr, write_info_ptr,
3414       PNG_DESTROY_WILL_FREE_DATA,
3415       PNG_FREE_PLTE|PNG_FREE_tRNS|PNG_FREE_hIST)
3416
3417thereby briefly reassigning responsibility for freeing to the user but
3418immediately afterwards reassigning it once more to the write_destroy
3419function.  Having done this, it would then be safe to destroy the read
3420structure and continue to use the PLTE, tRNS, and hIST data in the write
3421structure.
3422
3423This function only affects data that has already been allocated.
3424You can call this function before calling after the png_set_*() functions
3425to control whether the user or png_destroy_*() is supposed to free the data.
3426When the user assumes responsibility for libpng-allocated data, the
3427application must use
3428png_free() to free it, and when the user transfers responsibility to libpng
3429for data that the user has allocated, the user must have used png_malloc()
3430or png_zalloc() to allocate it.
3431
3432If you allocated text_ptr.text, text_ptr.lang, and text_ptr.translated_keyword
3433separately, do not transfer responsibility for freeing text_ptr to libpng,
3434because when libpng fills a png_text structure it combines these members with
3435the key member, and png_free_data() will free only text_ptr.key.  Similarly,
3436if you transfer responsibility for free'ing text_ptr from libpng to your
3437application, your application must not separately free those members.
3438For a more compact example of writing a PNG image, see the file example.c.
3439
3440V. Modifying/Customizing libpng:
3441
3442There are two issues here.  The first is changing how libpng does
3443standard things like memory allocation, input/output, and error handling.
3444The second deals with more complicated things like adding new chunks,
3445adding new transformations, and generally changing how libpng works.
3446Both of those are compile-time issues; that is, they are generally
3447determined at the time the code is written, and there is rarely a need
3448to provide the user with a means of changing them.
3449
3450Memory allocation, input/output, and error handling
3451
3452All of the memory allocation, input/output, and error handling in libpng
3453goes through callbacks that are user-settable.  The default routines are
3454in pngmem.c, pngrio.c, pngwio.c, and pngerror.c, respectively.  To change
3455these functions, call the appropriate png_set_*_fn() function.
3456
3457Memory allocation is done through the functions png_malloc(), png_calloc(),
3458and png_free().  These currently just call the standard C functions.
3459png_calloc() calls png_malloc() and then clears the newly
3460allocated memory to zero.  There is limited support for certain systems
3461with segmented memory architectures and the types of pointers declared by
3462png.h match this; you will have to use appropriate pointers in your
3463application.  Since it is
3464unlikely that the method of handling memory allocation on a platform
3465will change between applications, these functions must be modified in
3466the library at compile time.  If you prefer to use a different method
3467of allocating and freeing data, you can use png_create_read_struct_2() or
3468png_create_write_struct_2() to register your own functions as described
3469above.  These functions also provide a void pointer that can be retrieved
3470via
3471
3472    mem_ptr=png_get_mem_ptr(png_ptr);
3473
3474Your replacement memory functions must have prototypes as follows:
3475
3476    png_voidp malloc_fn(png_structp png_ptr,
3477       png_alloc_size_t size);
3478
3479    void free_fn(png_structp png_ptr, png_voidp ptr);
3480
3481Your malloc_fn() must return NULL in case of failure.  The png_malloc()
3482function will normally call png_error() if it receives a NULL from the
3483system memory allocator or from your replacement malloc_fn().
3484
3485Your free_fn() will never be called with a NULL ptr, since libpng's
3486png_free() checks for NULL before calling free_fn().
3487
3488Input/Output in libpng is done through png_read() and png_write(),
3489which currently just call fread() and fwrite().  The FILE * is stored in
3490png_struct and is initialized via png_init_io().  If you wish to change
3491the method of I/O, the library supplies callbacks that you can set
3492through the function png_set_read_fn() and png_set_write_fn() at run
3493time, instead of calling the png_init_io() function.  These functions
3494also provide a void pointer that can be retrieved via the function
3495png_get_io_ptr().  For example:
3496
3497    png_set_read_fn(png_structp read_ptr,
3498        voidp read_io_ptr, png_rw_ptr read_data_fn)
3499
3500    png_set_write_fn(png_structp write_ptr,
3501        voidp write_io_ptr, png_rw_ptr write_data_fn,
3502        png_flush_ptr output_flush_fn);
3503
3504    voidp read_io_ptr = png_get_io_ptr(read_ptr);
3505    voidp write_io_ptr = png_get_io_ptr(write_ptr);
3506
3507The replacement I/O functions must have prototypes as follows:
3508
3509    void user_read_data(png_structp png_ptr,
3510        png_bytep data, png_size_t length);
3511
3512    void user_write_data(png_structp png_ptr,
3513        png_bytep data, png_size_t length);
3514
3515    void user_flush_data(png_structp png_ptr);
3516
3517The user_read_data() function is responsible for detecting and
3518handling end-of-data errors.
3519
3520Supplying NULL for the read, write, or flush functions sets them back
3521to using the default C stream functions, which expect the io_ptr to
3522point to a standard *FILE structure.  It is probably a mistake
3523to use NULL for one of write_data_fn and output_flush_fn but not both
3524of them, unless you have built libpng with PNG_NO_WRITE_FLUSH defined.
3525It is an error to read from a write stream, and vice versa.
3526
3527Error handling in libpng is done through png_error() and png_warning().
3528Errors handled through png_error() are fatal, meaning that png_error()
3529should never return to its caller.  Currently, this is handled via
3530setjmp() and longjmp() (unless you have compiled libpng with
3531PNG_NO_SETJMP, in which case it is handled via PNG_ABORT()),
3532but you could change this to do things like exit() if you should wish,
3533as long as your function does not return.
3534
3535On non-fatal errors, png_warning() is called
3536to print a warning message, and then control returns to the calling code.
3537By default png_error() and png_warning() print a message on stderr via
3538fprintf() unless the library is compiled with PNG_NO_CONSOLE_IO defined
3539(because you don't want the messages) or PNG_NO_STDIO defined (because
3540fprintf() isn't available).  If you wish to change the behavior of the error
3541functions, you will need to set up your own message callbacks.  These
3542functions are normally supplied at the time that the png_struct is created.
3543It is also possible to redirect errors and warnings to your own replacement
3544functions after png_create_*_struct() has been called by calling:
3545
3546    png_set_error_fn(png_structp png_ptr,
3547        png_voidp error_ptr, png_error_ptr error_fn,
3548        png_error_ptr warning_fn);
3549
3550    png_voidp error_ptr = png_get_error_ptr(png_ptr);
3551
3552If NULL is supplied for either error_fn or warning_fn, then the libpng
3553default function will be used, calling fprintf() and/or longjmp() if a
3554problem is encountered.  The replacement error functions should have
3555parameters as follows:
3556
3557    void user_error_fn(png_structp png_ptr,
3558        png_const_charp error_msg);
3559
3560    void user_warning_fn(png_structp png_ptr,
3561        png_const_charp warning_msg);
3562
3563The motivation behind using setjmp() and longjmp() is the C++ throw and
3564catch exception handling methods.  This makes the code much easier to write,
3565as there is no need to check every return code of every function call.
3566However, there are some uncertainties about the status of local variables
3567after a longjmp, so the user may want to be careful about doing anything
3568after setjmp returns non-zero besides returning itself.  Consult your
3569compiler documentation for more details.  For an alternative approach, you
3570may wish to use the "cexcept" facility (see http://cexcept.sourceforge.net),
3571which is illustrated in pngvalid.c and in contrib/visupng.
3572
3573Custom chunks
3574
3575If you need to read or write custom chunks, you may need to get deeper
3576into the libpng code.  The library now has mechanisms for storing
3577and writing chunks of unknown type; you can even declare callbacks
3578for custom chunks.  However, this may not be good enough if the
3579library code itself needs to know about interactions between your
3580chunk and existing `intrinsic' chunks.
3581
3582If you need to write a new intrinsic chunk, first read the PNG
3583specification. Acquire a first level of understanding of how it works.
3584Pay particular attention to the sections that describe chunk names,
3585and look at how other chunks were designed, so you can do things
3586similarly.  Second, check out the sections of libpng that read and
3587write chunks.  Try to find a chunk that is similar to yours and use
3588it as a template.  More details can be found in the comments inside
3589the code.  It is best to handle private or unknown chunks in a generic method,
3590via callback functions, instead of by modifying libpng functions. This
3591is illustrated in pngtest.c, which uses a callback function to handle a
3592private "vpAg" chunk and the new "sTER" chunk, which are both unknown to
3593libpng.
3594
3595If you wish to write your own transformation for the data, look through
3596the part of the code that does the transformations, and check out some of
3597the simpler ones to get an idea of how they work.  Try to find a similar
3598transformation to the one you want to add and copy off of it.  More details
3599can be found in the comments inside the code itself.
3600
3601Configuring for 16-bit platforms
3602
3603You will want to look into zconf.h to tell zlib (and thus libpng) that
3604it cannot allocate more then 64K at a time.  Even if you can, the memory
3605won't be accessible.  So limit zlib and libpng to 64K by defining MAXSEG_64K.
3606
3607Configuring for DOS
3608
3609For DOS users who only have access to the lower 640K, you will
3610have to limit zlib's memory usage via a png_set_compression_mem_level()
3611call.  See zlib.h or zconf.h in the zlib library for more information.
3612
3613Configuring for Medium Model
3614
3615Libpng's support for medium model has been tested on most of the popular
3616compilers.  Make sure MAXSEG_64K gets defined, USE_FAR_KEYWORD gets
3617defined, and FAR gets defined to far in pngconf.h, and you should be
3618all set.  Everything in the library (except for zlib's structure) is
3619expecting far data.  You must use the typedefs with the p or pp on
3620the end for pointers (or at least look at them and be careful).  Make
3621note that the rows of data are defined as png_bytepp, which is
3622an "unsigned char far * far *".
3623
3624Configuring for gui/windowing platforms:
3625
3626You will need to write new error and warning functions that use the GUI
3627interface, as described previously, and set them to be the error and
3628warning functions at the time that png_create_*_struct() is called,
3629in order to have them available during the structure initialization.
3630They can be changed later via png_set_error_fn().  On some compilers,
3631you may also have to change the memory allocators (png_malloc, etc.).
3632
3633Configuring for compiler xxx:
3634
3635All includes for libpng are in pngconf.h.  If you need to add, change
3636or delete an include, this is the place to do it.
3637The includes that are not needed outside libpng are placed in pngpriv.h,
3638which is only used by the routines inside libpng itself.
3639The files in libpng proper only include pngpriv.h and png.h, which
3640in turn includes pngconf.h and, as of libpng-1.5.0, pnglibconf.h.
3641As of libpng-1.5.0, pngpriv.h also includes three other private header
3642files, pngstruct.h, pnginfo.h, and pngdebug.h, which contain material
3643that previously appeared in the public headers.
3644
3645Configuring zlib:
3646
3647There are special functions to configure the compression.  Perhaps the
3648most useful one changes the compression level, which currently uses
3649input compression values in the range 0 - 9.  The library normally
3650uses the default compression level (Z_DEFAULT_COMPRESSION = 6).  Tests
3651have shown that for a large majority of images, compression values in
3652the range 3-6 compress nearly as well as higher levels, and do so much
3653faster.  For online applications it may be desirable to have maximum speed
3654(Z_BEST_SPEED = 1).  With versions of zlib after v0.99, you can also
3655specify no compression (Z_NO_COMPRESSION = 0), but this would create
3656files larger than just storing the raw bitmap.  You can specify the
3657compression level by calling:
3658
3659    #include zlib.h
3660    png_set_compression_level(png_ptr, level);
3661
3662Another useful one is to reduce the memory level used by the library.
3663The memory level defaults to 8, but it can be lowered if you are
3664short on memory (running DOS, for example, where you only have 640K).
3665Note that the memory level does have an effect on compression; among
3666other things, lower levels will result in sections of incompressible
3667data being emitted in smaller stored blocks, with a correspondingly
3668larger relative overhead of up to 15% in the worst case.
3669
3670    #include zlib.h
3671    png_set_compression_mem_level(png_ptr, level);
3672
3673The other functions are for configuring zlib.  They are not recommended
3674for normal use and may result in writing an invalid PNG file.  See
3675zlib.h for more information on what these mean.
3676
3677    #include zlib.h
3678    png_set_compression_strategy(png_ptr,
3679        strategy);
3680
3681    png_set_compression_window_bits(png_ptr,
3682        window_bits);
3683
3684    png_set_compression_method(png_ptr, method);
3685
3686    png_set_compression_buffer_size(png_ptr, size);
3687
3688As of libpng version 1.5.4, additional APIs became
3689available to set these separately for non-IDAT
3690compressed chunks such as zTXt, iTXt, and iCCP:
3691
3692    #include zlib.h
3693    #if PNG_LIBPNG_VER >= 10504
3694    png_set_text_compression_level(png_ptr, level);
3695
3696    png_set_text_compression_mem_level(png_ptr, level);
3697
3698    png_set_text_compression_strategy(png_ptr,
3699        strategy);
3700
3701    png_set_text_compression_window_bits(png_ptr,
3702        window_bits);
3703
3704    png_set_text_compression_method(png_ptr, method);
3705    #endif
3706
3707Controlling row filtering
3708
3709If you want to control whether libpng uses filtering or not, which
3710filters are used, and how it goes about picking row filters, you
3711can call one of these functions.  The selection and configuration
3712of row filters can have a significant impact on the size and
3713encoding speed and a somewhat lesser impact on the decoding speed
3714of an image.  Filtering is enabled by default for RGB and grayscale
3715images (with and without alpha), but not for paletted images nor
3716for any images with bit depths less than 8 bits/pixel.
3717
3718The 'method' parameter sets the main filtering method, which is
3719currently only '0' in the PNG 1.2 specification.  The 'filters'
3720parameter sets which filter(s), if any, should be used for each
3721scanline.  Possible values are PNG_ALL_FILTERS and PNG_NO_FILTERS
3722to turn filtering on and off, respectively.
3723
3724Individual filter types are PNG_FILTER_NONE, PNG_FILTER_SUB,
3725PNG_FILTER_UP, PNG_FILTER_AVG, PNG_FILTER_PAETH, which can be bitwise
3726ORed together with '|' to specify one or more filters to use.
3727These filters are described in more detail in the PNG specification.
3728If you intend to change the filter type during the course of writing
3729the image, you should start with flags set for all of the filters
3730you intend to use so that libpng can initialize its internal
3731structures appropriately for all of the filter types.  (Note that this
3732means the first row must always be adaptively filtered, because libpng
3733currently does not allocate the filter buffers until png_write_row()
3734is called for the first time.)
3735
3736    filters = PNG_FILTER_NONE | PNG_FILTER_SUB
3737              PNG_FILTER_UP | PNG_FILTER_AVG |
3738              PNG_FILTER_PAETH | PNG_ALL_FILTERS;
3739
3740    png_set_filter(png_ptr, PNG_FILTER_TYPE_BASE,
3741       filters);
3742              The second parameter can also be
3743              PNG_INTRAPIXEL_DIFFERENCING if you are
3744              writing a PNG to be embedded in a MNG
3745              datastream.  This parameter must be the
3746              same as the value of filter_method used
3747              in png_set_IHDR().
3748
3749It is also possible to influence how libpng chooses from among the
3750available filters.  This is done in one or both of two ways - by
3751telling it how important it is to keep the same filter for successive
3752rows, and by telling it the relative computational costs of the filters.
3753
3754    double weights[3] = {1.5, 1.3, 1.1},
3755       costs[PNG_FILTER_VALUE_LAST] =
3756       {1.0, 1.3, 1.3, 1.5, 1.7};
3757
3758    png_set_filter_heuristics(png_ptr,
3759       PNG_FILTER_HEURISTIC_WEIGHTED, 3,
3760       weights, costs);
3761
3762The weights are multiplying factors that indicate to libpng that the
3763row filter should be the same for successive rows unless another row filter
3764is that many times better than the previous filter.  In the above example,
3765if the previous 3 filters were SUB, SUB, NONE, the SUB filter could have a
3766"sum of absolute differences" 1.5 x 1.3 times higher than other filters
3767and still be chosen, while the NONE filter could have a sum 1.1 times
3768higher than other filters and still be chosen.  Unspecified weights are
3769taken to be 1.0, and the specified weights should probably be declining
3770like those above in order to emphasize recent filters over older filters.
3771
3772The filter costs specify for each filter type a relative decoding cost
3773to be considered when selecting row filters.  This means that filters
3774with higher costs are less likely to be chosen over filters with lower
3775costs, unless their "sum of absolute differences" is that much smaller.
3776The costs do not necessarily reflect the exact computational speeds of
3777the various filters, since this would unduly influence the final image
3778size.
3779
3780Note that the numbers above were invented purely for this example and
3781are given only to help explain the function usage.  Little testing has
3782been done to find optimum values for either the costs or the weights.
3783
3784Removing unwanted object code
3785
3786There are a bunch of #define's in pngconf.h that control what parts of
3787libpng are compiled.  All the defines end in _SUPPORTED.  If you are
3788never going to use a capability, you can change the #define to #undef
3789before recompiling libpng and save yourself code and data space, or
3790you can turn off individual capabilities with defines that begin with
3791PNG_NO_.
3792
3793In libpng-1.5.0 and later, the #define's are in pnglibconf.h instead.
3794
3795You can also turn all of the transforms and ancillary chunk capabilities
3796off en masse with compiler directives that define
3797PNG_NO_READ[or WRITE]_TRANSFORMS, or PNG_NO_READ[or WRITE]_ANCILLARY_CHUNKS,
3798or all four,
3799along with directives to turn on any of the capabilities that you do
3800want.  The PNG_NO_READ[or WRITE]_TRANSFORMS directives disable the extra
3801transformations but still leave the library fully capable of reading
3802and writing PNG files with all known public chunks. Use of the
3803PNG_NO_READ[or WRITE]_ANCILLARY_CHUNKS directive produces a library
3804that is incapable of reading or writing ancillary chunks.  If you are
3805not using the progressive reading capability, you can turn that off
3806with PNG_NO_PROGRESSIVE_READ (don't confuse this with the INTERLACING
3807capability, which you'll still have).
3808
3809All the reading and writing specific code are in separate files, so the
3810linker should only grab the files it needs.  However, if you want to
3811make sure, or if you are building a stand alone library, all the
3812reading files start with "pngr" and all the writing files start with "pngw".
3813The files that don't match either (like png.c, pngtrans.c, etc.)
3814are used for both reading and writing, and always need to be included.
3815The progressive reader is in pngpread.c
3816
3817If you are creating or distributing a dynamically linked library (a .so
3818or DLL file), you should not remove or disable any parts of the library,
3819as this will cause applications linked with different versions of the
3820library to fail if they call functions not available in your library.
3821The size of the library itself should not be an issue, because only
3822those sections that are actually used will be loaded into memory.
3823
3824Requesting debug printout
3825
3826The macro definition PNG_DEBUG can be used to request debugging
3827printout.  Set it to an integer value in the range 0 to 3.  Higher
3828numbers result in increasing amounts of debugging information.  The
3829information is printed to the "stderr" file, unless another file
3830name is specified in the PNG_DEBUG_FILE macro definition.
3831
3832When PNG_DEBUG > 0, the following functions (macros) become available:
3833
3834   png_debug(level, message)
3835   png_debug1(level, message, p1)
3836   png_debug2(level, message, p1, p2)
3837
3838in which "level" is compared to PNG_DEBUG to decide whether to print
3839the message, "message" is the formatted string to be printed,
3840and p1 and p2 are parameters that are to be embedded in the string
3841according to printf-style formatting directives.  For example,
3842
3843   png_debug1(2, "foo=%d\n", foo);
3844
3845is expanded to
3846
3847   if (PNG_DEBUG > 2)
3848      fprintf(PNG_DEBUG_FILE, "foo=%d\n", foo);
3849
3850When PNG_DEBUG is defined but is zero, the macros aren't defined, but you
3851can still use PNG_DEBUG to control your own debugging:
3852
3853   #ifdef PNG_DEBUG
3854       fprintf(stderr, ...
3855   #endif
3856
3857When PNG_DEBUG = 1, the macros are defined, but only png_debug statements
3858having level = 0 will be printed.  There aren't any such statements in
3859this version of libpng, but if you insert some they will be printed.
3860
3861VI.  MNG support
3862
3863The MNG specification (available at http://www.libpng.org/pub/mng) allows
3864certain extensions to PNG for PNG images that are embedded in MNG datastreams.
3865Libpng can support some of these extensions.  To enable them, use the
3866png_permit_mng_features() function:
3867
3868   feature_set = png_permit_mng_features(png_ptr, mask)
3869
3870   mask is a png_uint_32 containing the bitwise OR of the
3871        features you want to enable.  These include
3872        PNG_FLAG_MNG_EMPTY_PLTE
3873        PNG_FLAG_MNG_FILTER_64
3874        PNG_ALL_MNG_FEATURES
3875
3876   feature_set is a png_uint_32 that is the bitwise AND of
3877      your mask with the set of MNG features that is
3878      supported by the version of libpng that you are using.
3879
3880It is an error to use this function when reading or writing a standalone
3881PNG file with the PNG 8-byte signature.  The PNG datastream must be wrapped
3882in a MNG datastream.  As a minimum, it must have the MNG 8-byte signature
3883and the MHDR and MEND chunks.  Libpng does not provide support for these
3884or any other MNG chunks; your application must provide its own support for
3885them.  You may wish to consider using libmng (available at
3886http://www.libmng.com) instead.
3887
3888VII.  Changes to Libpng from version 0.88
3889
3890It should be noted that versions of libpng later than 0.96 are not
3891distributed by the original libpng author, Guy Schalnat, nor by
3892Andreas Dilger, who had taken over from Guy during 1996 and 1997, and
3893distributed versions 0.89 through 0.96, but rather by another member
3894of the original PNG Group, Glenn Randers-Pehrson.  Guy and Andreas are
3895still alive and well, but they have moved on to other things.
3896
3897The old libpng functions png_read_init(), png_write_init(),
3898png_info_init(), png_read_destroy(), and png_write_destroy() have been
3899moved to PNG_INTERNAL in version 0.95 to discourage their use.  These
3900functions will be removed from libpng version 1.4.0.
3901
3902The preferred method of creating and initializing the libpng structures is
3903via the png_create_read_struct(), png_create_write_struct(), and
3904png_create_info_struct() because they isolate the size of the structures
3905from the application, allow version error checking, and also allow the
3906use of custom error handling routines during the initialization, which
3907the old functions do not.  The functions png_read_destroy() and
3908png_write_destroy() do not actually free the memory that libpng
3909allocated for these structs, but just reset the data structures, so they
3910can be used instead of png_destroy_read_struct() and
3911png_destroy_write_struct() if you feel there is too much system overhead
3912allocating and freeing the png_struct for each image read.
3913
3914Setting the error callbacks via png_set_message_fn() before
3915png_read_init() as was suggested in libpng-0.88 is no longer supported
3916because this caused applications that do not use custom error functions
3917to fail if the png_ptr was not initialized to zero.  It is still possible
3918to set the error callbacks AFTER png_read_init(), or to change them with
3919png_set_error_fn(), which is essentially the same function, but with a new
3920name to force compilation errors with applications that try to use the old
3921method.
3922
3923Starting with version 1.0.7, you can find out which version of the library
3924you are using at run-time:
3925
3926   png_uint_32 libpng_vn = png_access_version_number();
3927
3928The number libpng_vn is constructed from the major version, minor
3929version with leading zero, and release number with leading zero,
3930(e.g., libpng_vn for version 1.0.7 is 10007).
3931
3932Note that this function does not take a png_ptr, so you can call it
3933before you've created one.
3934
3935You can also check which version of png.h you used when compiling your
3936application:
3937
3938   png_uint_32 application_vn = PNG_LIBPNG_VER;
3939
3940VIII.  Changes to Libpng from version 1.0.x to 1.2.x
3941
3942Support for user memory management was enabled by default.  To
3943accomplish this, the functions png_create_read_struct_2(),
3944png_create_write_struct_2(), png_set_mem_fn(), png_get_mem_ptr(),
3945png_malloc_default(), and png_free_default() were added.
3946
3947Support for the iTXt chunk has been enabled by default as of
3948version 1.2.41.
3949
3950Support for certain MNG features was enabled.
3951
3952Support for numbered error messages was added.  However, we never got
3953around to actually numbering the error messages.  The function
3954png_set_strip_error_numbers() was added (Note: the prototype for this
3955function was inadvertently removed from png.h in PNG_NO_ASSEMBLER_CODE
3956builds of libpng-1.2.15.  It was restored in libpng-1.2.36).
3957
3958The png_malloc_warn() function was added at libpng-1.2.3.  This issues
3959a png_warning and returns NULL instead of aborting when it fails to
3960acquire the requested memory allocation.
3961
3962Support for setting user limits on image width and height was enabled
3963by default.  The functions png_set_user_limits(), png_get_user_width_max(),
3964and png_get_user_height_max() were added at libpng-1.2.6.
3965
3966The png_set_add_alpha() function was added at libpng-1.2.7.
3967
3968The function png_set_expand_gray_1_2_4_to_8() was added at libpng-1.2.9.
3969Unlike png_set_gray_1_2_4_to_8(), the new function does not expand the
3970tRNS chunk to alpha. The png_set_gray_1_2_4_to_8() function is
3971deprecated.
3972
3973A number of macro definitions in support of runtime selection of
3974assembler code features (especially Intel MMX code support) were
3975added at libpng-1.2.0:
3976
3977    PNG_ASM_FLAG_MMX_SUPPORT_COMPILED
3978    PNG_ASM_FLAG_MMX_SUPPORT_IN_CPU
3979    PNG_ASM_FLAG_MMX_READ_COMBINE_ROW
3980    PNG_ASM_FLAG_MMX_READ_INTERLACE
3981    PNG_ASM_FLAG_MMX_READ_FILTER_SUB
3982    PNG_ASM_FLAG_MMX_READ_FILTER_UP
3983    PNG_ASM_FLAG_MMX_READ_FILTER_AVG
3984    PNG_ASM_FLAG_MMX_READ_FILTER_PAETH
3985    PNG_ASM_FLAGS_INITIALIZED
3986    PNG_MMX_READ_FLAGS
3987    PNG_MMX_FLAGS
3988    PNG_MMX_WRITE_FLAGS
3989    PNG_MMX_FLAGS
3990
3991We added the following functions in support of runtime
3992selection of assembler code features:
3993
3994    png_get_mmx_flagmask()
3995    png_set_mmx_thresholds()
3996    png_get_asm_flags()
3997    png_get_mmx_bitdepth_threshold()
3998    png_get_mmx_rowbytes_threshold()
3999    png_set_asm_flags()
4000
4001We replaced all of these functions with simple stubs in libpng-1.2.20,
4002when the Intel assembler code was removed due to a licensing issue.
4003
4004These macros are deprecated:
4005
4006    PNG_READ_TRANSFORMS_NOT_SUPPORTED
4007    PNG_PROGRESSIVE_READ_NOT_SUPPORTED
4008    PNG_NO_SEQUENTIAL_READ_SUPPORTED
4009    PNG_WRITE_TRANSFORMS_NOT_SUPPORTED
4010    PNG_READ_ANCILLARY_CHUNKS_NOT_SUPPORTED
4011    PNG_WRITE_ANCILLARY_CHUNKS_NOT_SUPPORTED
4012
4013They have been replaced, respectively, by:
4014
4015    PNG_NO_READ_TRANSFORMS
4016    PNG_NO_PROGRESSIVE_READ
4017    PNG_NO_SEQUENTIAL_READ
4018    PNG_NO_WRITE_TRANSFORMS
4019    PNG_NO_READ_ANCILLARY_CHUNKS
4020    PNG_NO_WRITE_ANCILLARY_CHUNKS
4021
4022PNG_MAX_UINT was replaced with PNG_UINT_31_MAX.  It has been
4023deprecated since libpng-1.0.16 and libpng-1.2.6.
4024
4025The function
4026    png_check_sig(sig, num)
4027was replaced with
4028    !png_sig_cmp(sig, 0, num)
4029It has been deprecated since libpng-0.90.
4030
4031The function
4032    png_set_gray_1_2_4_to_8()
4033which also expands tRNS to alpha was replaced with
4034    png_set_expand_gray_1_2_4_to_8()
4035which does not. It has been deprecated since libpng-1.0.18 and 1.2.9.
4036
4037IX.  Changes to Libpng from version 1.0.x/1.2.x to 1.4.x
4038
4039Private libpng prototypes and macro definitions were moved from
4040png.h and pngconf.h into a new pngpriv.h header file.
4041
4042Functions png_set_benign_errors(), png_benign_error(), and
4043png_chunk_benign_error() were added.
4044
4045Support for setting the maximum amount of memory that the application
4046will allocate for reading chunks was added, as a security measure.
4047The functions png_set_chunk_cache_max() and png_get_chunk_cache_max()
4048were added to the library.
4049
4050We implemented support for I/O states by adding png_ptr member io_state
4051and functions png_get_io_chunk_name() and png_get_io_state() in pngget.c
4052
4053We added PNG_TRANSFORM_GRAY_TO_RGB to the available high-level
4054input transforms.
4055
4056Checking for and reporting of errors in the IHDR chunk is more thorough.
4057
4058Support for global arrays was removed, to improve thread safety.
4059
4060Some obsolete/deprecated macros and functions have been removed.
4061
4062Typecasted NULL definitions such as
4063   #define png_voidp_NULL            (png_voidp)NULL
4064were eliminated.  If you used these in your application, just use
4065NULL instead.
4066
4067The png_struct and info_struct members "trans" and "trans_values" were
4068changed to "trans_alpha" and "trans_color", respectively.
4069
4070The obsolete, unused pnggccrd.c and pngvcrd.c files and related makefiles
4071were removed.
4072
4073The PNG_1_0_X and PNG_1_2_X macros were eliminated.
4074
4075The PNG_LEGACY_SUPPORTED macro was eliminated.
4076
4077Many WIN32_WCE #ifdefs were removed.
4078
4079The functions png_read_init(info_ptr), png_write_init(info_ptr),
4080png_info_init(info_ptr), png_read_destroy(), and png_write_destroy()
4081have been removed.  They have been deprecated since libpng-0.95.
4082
4083The png_permit_empty_plte() was removed. It has been deprecated
4084since libpng-1.0.9.  Use png_permit_mng_features() instead.
4085
4086We removed the obsolete stub functions png_get_mmx_flagmask(),
4087png_set_mmx_thresholds(), png_get_asm_flags(),
4088png_get_mmx_bitdepth_threshold(), png_get_mmx_rowbytes_threshold(),
4089png_set_asm_flags(), and png_mmx_supported()
4090
4091We removed the obsolete png_check_sig(), png_memcpy_check(), and
4092png_memset_check() functions.  Instead use !png_sig_cmp(), memcpy(),
4093and memset(), respectively.
4094
4095The function png_set_gray_1_2_4_to_8() was removed. It has been
4096deprecated since libpng-1.0.18 and 1.2.9, when it was replaced with
4097png_set_expand_gray_1_2_4_to_8() because the former function also
4098expanded any tRNS chunk to an alpha channel.
4099
4100Macros for png_get_uint_16, png_get_uint_32, and png_get_int_32
4101were added and are used by default instead of the corresponding
4102functions. Unfortunately,
4103from libpng-1.4.0 until 1.4.4, the png_get_uint_16 macro (but not the
4104function) incorrectly returned a value of type png_uint_32.
4105
4106We changed the prototype for png_malloc() from
4107    png_malloc(png_structp png_ptr, png_uint_32 size)
4108to
4109    png_malloc(png_structp png_ptr, png_alloc_size_t size)
4110
4111This also applies to the prototype for the user replacement malloc_fn().
4112
4113The png_calloc() function was added and is used in place of
4114of "png_malloc(); memset();" except in the case in png_read_png()
4115where the array consists of pointers; in this case a "for" loop is used
4116after the png_malloc() to set the pointers to NULL, to give robust.
4117behavior in case the application runs out of memory part-way through
4118the process.
4119
4120We changed the prototypes of png_get_compression_buffer_size() and
4121png_set_compression_buffer_size() to work with png_size_t instead of
4122png_uint_32.
4123
4124Support for numbered error messages was removed by default, since we
4125never got around to actually numbering the error messages. The function
4126png_set_strip_error_numbers() was removed from the library by default.
4127
4128The png_zalloc() and png_zfree() functions are no longer exported.
4129The png_zalloc() function no longer zeroes out the memory that it
4130allocates.
4131
4132Support for dithering was disabled by default in libpng-1.4.0, because
4133it has not been well tested and doesn't actually "dither".
4134The code was not
4135removed, however, and could be enabled by building libpng with
4136PNG_READ_DITHER_SUPPORTED defined.  In libpng-1.4.2, this support
4137was reenabled, but the function was renamed png_set_quantize() to
4138reflect more accurately what it actually does.  At the same time,
4139the PNG_DITHER_[RED,GREEN_BLUE]_BITS macros were also renamed to
4140PNG_QUANTIZE_[RED,GREEN,BLUE]_BITS, and PNG_READ_DITHER_SUPPORTED
4141was renamed to PNG_READ_QUANTIZE_SUPPORTED.
4142
4143We removed the trailing '.' from the warning and error messages.
4144
4145X.  Changes to Libpng from version 1.4.x to 1.5.x
4146
4147From libpng-1.4.0 until 1.4.4, the png_get_uint_16 macro (but not the
4148function) incorrectly returned a value of type png_uint_32.
4149
4150Checking for invalid palette index on read or write was added at libpng
41511.5.10.  When an invalid index is found, libpng issues a benign error.
4152This is enabled by default but can be disabled in each png_ptr with
4153
4154   png_set_check_for_invalid_index(png_ptr, allowed);
4155
4156      allowed  - one of
4157                 0: disable
4158                 1: enable
4159
4160A. Changes that affect users of libpng
4161
4162There are no substantial API changes between the non-deprecated parts of
4163the 1.4.5 API and the 1.5.0 API; however, the ability to directly access
4164the main libpng control structures, png_struct and png_info, deprecated
4165in earlier versions of libpng, has been completely removed from
4166libpng 1.5.
4167
4168We no longer include zlib.h in png.h.  Applications that need access
4169to information in zlib.h will need to add the '#include "zlib.h"'
4170directive.  It does not matter whether it is placed prior to or after
4171the '"#include png.h"' directive.
4172
4173We moved the png_strcpy(), png_strncpy(), png_strlen(), png_memcpy(),
4174png_memcmp(), png_sprintf, and png_memcpy() macros into a private
4175header file (pngpriv.h) that is not accessible to applications.
4176
4177In png_get_iCCP, the type of "profile" was changed from png_charpp
4178to png_bytepp, and in png_set_iCCP, from png_charp to png_const_bytep.
4179
4180There are changes of form in png.h, including new and changed macros to
4181declare parts of the API.  Some API functions with arguments that are
4182pointers to data not modified within the function have been corrected to
4183declare these arguments with PNG_CONST.
4184
4185Much of the internal use of C macros to control the library build has also
4186changed and some of this is visible in the exported header files, in
4187particular the use of macros to control data and API elements visible
4188during application compilation may require significant revision to
4189application code.  (It is extremely rare for an application to do this.)
4190
4191Any program that compiled against libpng 1.4 and did not use deprecated
4192features or access internal library structures should compile and work
4193against libpng 1.5, except for the change in the prototype for
4194png_get_iCCP() and png_set_iCCP() API functions mentioned above.
4195
4196libpng 1.5.0 adds PNG_ PASS macros to help in the reading and writing of
4197interlaced images.  The macros return the number of rows and columns in
4198each pass and information that can be used to de-interlace and (if
4199absolutely necessary) interlace an image.
4200
4201libpng 1.5.0 adds an API png_longjmp(png_ptr, value).  This API calls
4202the application-provided png_longjmp_ptr on the internal, but application
4203initialized, longjmp buffer.  It is provided as a convenience to avoid
4204the need to use the png_jmpbuf macro, which had the unnecessary side
4205effect of resetting the internal png_longjmp_ptr value.
4206
4207libpng 1.5.0 includes a complete fixed point API.  By default this is
4208present along with the corresponding floating point API.  In general the
4209fixed point API is faster and smaller than the floating point one because
4210the PNG file format used fixed point, not floating point.  This applies
4211even if the library uses floating point in internal calculations.  A new
4212macro, PNG_FLOATING_ARITHMETIC_SUPPORTED, reveals whether the library
4213uses floating point arithmetic (the default) or fixed point arithmetic
4214internally for performance critical calculations such as gamma correction.
4215In some cases, the gamma calculations may produce slightly different
4216results.  This has changed the results in png_rgb_to_gray and in alpha
4217composition (png_set_background for example). This applies even if the
4218original image was already linear (gamma == 1.0) and, therefore, it is
4219not necessary to linearize the image.  This is because libpng has *not*
4220been changed to optimize that case correctly, yet.
4221
4222Fixed point support for the sCAL chunk comes with an important caveat;
4223the sCAL specification uses a decimal encoding of floating point values
4224and the accuracy of PNG fixed point values is insufficient for
4225representation of these values. Consequently a "string" API
4226(png_get_sCAL_s and png_set_sCAL_s) is the only reliable way of reading
4227arbitrary sCAL chunks in the absence of either the floating point API or
4228internal floating point calculations.
4229
4230Applications no longer need to include the optional distribution header
4231file pngusr.h or define the corresponding macros during application
4232build in order to see the correct variant of the libpng API.  From 1.5.0
4233application code can check for the corresponding _SUPPORTED macro:
4234
4235#ifdef PNG_INCH_CONVERSIONS_SUPPORTED
4236   /* code that uses the inch conversion APIs. */
4237#endif
4238
4239This macro will only be defined if the inch conversion functions have been
4240compiled into libpng.  The full set of macros, and whether or not support
4241has been compiled in, are available in the header file pnglibconf.h.
4242This header file is specific to the libpng build.  Notice that prior to
42431.5.0 the _SUPPORTED macros would always have the default definition unless
4244reset by pngusr.h or by explicit settings on the compiler command line.
4245These settings may produce compiler warnings or errors in 1.5.0 because
4246of macro redefinition.
4247
4248From libpng-1.4.0 until 1.4.4, the png_get_uint_16 macro (but not the
4249function) incorrectly returned a value of type png_uint_32.  libpng 1.5.0
4250is consistent with the implementation in 1.4.5 and 1.2.x (where the macro
4251did not exist.)
4252
4253Applications can now choose whether to use these macros or to call the
4254corresponding function by defining PNG_USE_READ_MACROS or
4255PNG_NO_USE_READ_MACROS before including png.h.  Notice that this is
4256only supported from 1.5.0 -defining PNG_NO_USE_READ_MACROS prior to 1.5.0
4257will lead to a link failure.
4258
4259Prior to libpng-1.5.4, the zlib compressor used the same set of parameters
4260when compressing the IDAT data and textual data such as zTXt and iCCP.
4261In libpng-1.5.4 we reinitialized the zlib stream for each type of data.
4262We added five png_set_text_*() functions for setting the parameters to
4263use with textual data.
4264
4265Prior to libpng-1.5.4, the PNG_READ_16_TO_8_ACCURATE_SCALE_SUPPORTED
4266option was off by default, and slightly inaccurate scaling occurred.
4267This option can no longer be turned off, and the choice of accurate
4268or inaccurate 16-to-8 scaling is by using the new png_set_scale_16_to_8()
4269API for accurate scaling or the old png_set_strip_16_to_8() API for simple
4270chopping.
4271
4272Prior to libpng-1.5.4, the png_set_user_limits() function could only be
4273used to reduce the width and height limits from the value of
4274PNG_USER_WIDTH_MAX and PNG_USER_HEIGHT_MAX, although this document said
4275that it could be used to override them.  Now this function will reduce or
4276increase the limits.
4277
4278Starting in libpng-1.5.10, the user limits can be set en masse with the
4279configuration option PNG_SAFE_LIMITS_SUPPORTED.  If this option is enabled,
4280a set of "safe" limits is applied in pngpriv.h.  These can be overridden by
4281application calls to png_set_user_limits(), png_set_user_chunk_cache_max(),
4282and/or png_set_user_malloc_max() that increase or decrease the limits.  Also,
4283in libpng-1.5.10 the default width and height limits were increased
4284from 1,000,000 to 0x7ffffff (i.e., made unlimited).  Therefore, the
4285limits are now
4286                               default      safe
4287   png_user_width_max        0x7fffffff    1,000,000
4288   png_user_height_max       0x7fffffff    1,000,000
4289   png_user_chunk_cache_max  0 (unlimited)   128
4290   png_user_chunk_malloc_max 0 (unlimited) 8,000,000
4291
4292B. Changes to the build and configuration of libpng
4293
4294Details of internal changes to the library code can be found in the CHANGES
4295file and in the GIT repository logs.  These will be of no concern to the vast
4296majority of library users or builders; however, the few who configure libpng
4297to a non-default feature set may need to change how this is done.
4298
4299There should be no need for library builders to alter build scripts if
4300these use the distributed build support - configure or the makefiles -
4301however, users of the makefiles may care to update their build scripts
4302to build pnglibconf.h where the corresponding makefile does not do so.
4303
4304Building libpng with a non-default configuration has changed completely.
4305The old method using pngusr.h should still work correctly even though the
4306way pngusr.h is used in the build has been changed; however, library
4307builders will probably want to examine the changes to take advantage of
4308new capabilities and to simplify their build system.
4309
4310B.1 Specific changes to library configuration capabilities
4311
4312The library now supports a complete fixed point implementation and can
4313thus be used on systems that have no floating point support or very
4314limited or slow support.  Previously gamma correction, an essential part
4315of complete PNG support, required reasonably fast floating point.
4316
4317As part of this the choice of internal implementation has been made
4318independent of the choice of fixed versus floating point APIs and all the
4319missing fixed point APIs have been implemented.
4320
4321The exact mechanism used to control attributes of API functions has
4322changed.  A single set of operating system independent macro definitions
4323is used and operating system specific directives are defined in
4324pnglibconf.h
4325
4326As part of this the mechanism used to choose procedure call standards on
4327those systems that allow a choice has been changed.  At present this only
4328affects certain Microsoft (DOS, Windows) and IBM (OS/2) operating systems
4329running on Intel processors.  As before, PNGAPI is defined where required
4330to control the exported API functions; however, two new macros, PNGCBAPI
4331and PNGCAPI, are used instead for callback functions (PNGCBAPI) and
4332(PNGCAPI) for functions that must match a C library prototype (currently
4333only png_longjmp_ptr, which must match the C longjmp function.)  The new
4334approach is documented in pngconf.h
4335
4336Despite these changes, libpng 1.5.0 only supports the native C function
4337calling standard on those platforms tested so far (__cdecl on Microsoft
4338Windows).  This is because the support requirements for alternative
4339calling conventions seem to no longer exist.  Developers who find it
4340necessary to set PNG_API_RULE to 1 should advise the mailing list
4341(png-mng-implement) of this and library builders who use Openwatcom and
4342therefore set PNG_API_RULE to 2 should also contact the mailing list.
4343
4344A new test program, pngvalid, is provided in addition to pngtest.
4345pngvalid validates the arithmetic accuracy of the gamma correction
4346calculations and includes a number of validations of the file format.
4347A subset of the full range of tests is run when "make check" is done
4348(in the 'configure' build.)  pngvalid also allows total allocated memory
4349usage to be evaluated and performs additional memory overwrite validation.
4350
4351Many changes to individual feature macros have been made. The following
4352are the changes most likely to be noticed by library builders who
4353configure libpng:
4354
43551) All feature macros now have consistent naming:
4356
4357#define PNG_NO_feature turns the feature off
4358#define PNG_feature_SUPPORTED turns the feature on
4359
4360pnglibconf.h contains one line for each feature macro which is either:
4361
4362#define PNG_feature_SUPPORTED
4363
4364if the feature is supported or:
4365
4366/*#undef PNG_feature_SUPPORTED*/
4367
4368if it is not.  Library code consistently checks for the 'SUPPORTED' macro.
4369It does not, and libpng applications should not, check for the 'NO' macro
4370which will not normally be defined even if the feature is not supported.
4371The 'NO' macros are only used internally for setting or not setting the
4372corresponding 'SUPPORTED' macros.
4373
4374Compatibility with the old names is provided as follows:
4375
4376PNG_INCH_CONVERSIONS turns on PNG_INCH_CONVERSIONS_SUPPORTED
4377
4378And the following definitions disable the corresponding feature:
4379
4380PNG_SETJMP_NOT_SUPPORTED disables SETJMP
4381PNG_READ_TRANSFORMS_NOT_SUPPORTED disables READ_TRANSFORMS
4382PNG_NO_READ_COMPOSITED_NODIV disables READ_COMPOSITE_NODIV
4383PNG_WRITE_TRANSFORMS_NOT_SUPPORTED disables WRITE_TRANSFORMS
4384PNG_READ_ANCILLARY_CHUNKS_NOT_SUPPORTED disables READ_ANCILLARY_CHUNKS
4385PNG_WRITE_ANCILLARY_CHUNKS_NOT_SUPPORTED disables WRITE_ANCILLARY_CHUNKS
4386
4387Library builders should remove use of the above, inconsistent, names.
4388
43892) Warning and error message formatting was previously conditional on
4390the STDIO feature. The library has been changed to use the
4391CONSOLE_IO feature instead. This means that if CONSOLE_IO is disabled
4392the library no longer uses the printf(3) functions, even though the
4393default read/write implementations use (FILE) style stdio.h functions.
4394
43953) Three feature macros now control the fixed/floating point decisions:
4396
4397PNG_FLOATING_POINT_SUPPORTED enables the floating point APIs
4398
4399PNG_FIXED_POINT_SUPPORTED enables the fixed point APIs; however, in
4400practice these are normally required internally anyway (because the PNG
4401file format is fixed point), therefore in most cases PNG_NO_FIXED_POINT
4402merely stops the function from being exported.
4403
4404PNG_FLOATING_ARITHMETIC_SUPPORTED chooses between the internal floating
4405point implementation or the fixed point one.  Typically the fixed point
4406implementation is larger and slower than the floating point implementation
4407on a system that supports floating point; however, it may be faster on a
4408system which lacks floating point hardware and therefore uses a software
4409emulation.
4410
44114) Added PNG_{READ,WRITE}_INT_FUNCTIONS_SUPPORTED.  This allows the
4412functions to read and write ints to be disabled independently of
4413PNG_USE_READ_MACROS, which allows libpng to be built with the functions
4414even though the default is to use the macros - this allows applications
4415to choose at app buildtime whether or not to use macros (previously
4416impossible because the functions weren't in the default build.)
4417
4418B.2 Changes to the configuration mechanism
4419
4420Prior to libpng-1.5.0 library builders who needed to configure libpng
4421had either to modify the exported pngconf.h header file to add system
4422specific configuration or had to write feature selection macros into
4423pngusr.h and cause this to be included into pngconf.h by defining
4424PNG_USER_CONFIG. The latter mechanism had the disadvantage that an
4425application built without PNG_USER_CONFIG defined would see the
4426unmodified, default, libpng API and thus would probably fail to link.
4427
4428These mechanisms still work in the configure build and in any makefile
4429build that builds pnglibconf.h, although the feature selection macros
4430have changed somewhat as described above.  In 1.5.0, however, pngusr.h is
4431processed only once, when the exported header file pnglibconf.h is built.
4432pngconf.h no longer includes pngusr.h, therefore pngusr.h is ignored after the
4433build of pnglibconf.h and it is never included in an application build.
4434
4435The rarely used alternative of adding a list of feature macros to the
4436CFLAGS setting in the build also still works; however, the macros will be
4437copied to pnglibconf.h and this may produce macro redefinition warnings
4438when the individual C files are compiled.
4439
4440All configuration now only works if pnglibconf.h is built from
4441scripts/pnglibconf.dfa.  This requires the program awk.  Brian Kernighan
4442(the original author of awk) maintains C source code of that awk and this
4443and all known later implementations (often called by subtly different
4444names - nawk and gawk for example) are adequate to build pnglibconf.h.
4445The Sun Microsystems (now Oracle) program 'awk' is an earlier version
4446and does not work; this may also apply to other systems that have a
4447functioning awk called 'nawk'.
4448
4449Configuration options are now documented in scripts/pnglibconf.dfa.  This
4450file also includes dependency information that ensures a configuration is
4451consistent; that is, if a feature is switched off dependent features are
4452also removed.  As a recommended alternative to using feature macros in
4453pngusr.h a system builder may also define equivalent options in pngusr.dfa
4454(or, indeed, any file) and add that to the configuration by setting
4455DFA_XTRA to the file name.  The makefiles in contrib/pngminim illustrate
4456how to do this, and a case where pngusr.h is still required.
4457
4458XI. Detecting libpng
4459
4460The png_get_io_ptr() function has been present since libpng-0.88, has never
4461changed, and is unaffected by conditional compilation macros.  It is the
4462best choice for use in configure scripts for detecting the presence of any
4463libpng version since 0.88.  In an autoconf "configure.in" you could use
4464
4465    AC_CHECK_LIB(png, png_get_io_ptr, ...
4466
4467XII. Source code repository
4468
4469Since about February 2009, version 1.2.34, libpng has been under "git" source
4470control.  The git repository was built from old libpng-x.y.z.tar.gz files
4471going back to version 0.70.  You can access the git repository (read only)
4472at
4473
4474    git://libpng.git.sourceforge.net/gitroot/libpng
4475
4476or you can browse it via "gitweb" at
4477
4478    http://libpng.git.sourceforge.net/git/gitweb.cgi?p=libpng
4479
4480Patches can be sent to glennrp at users.sourceforge.net or to
4481png-mng-implement at lists.sourceforge.net or you can upload them to
4482the libpng bug tracker at
4483
4484    http://libpng.sourceforge.net
4485
4486We also accept patches built from the tar or zip distributions, and
4487simple verbal discriptions of bug fixes, reported either to the
4488SourceForge bug tracker, to the png-mng-implement at lists.sf.net
4489mailing list, or directly to glennrp.
4490
4491XIII. Coding style
4492
4493Our coding style is similar to the "Allman" style, with curly
4494braces on separate lines:
4495
4496    if (condition)
4497    {
4498       action;
4499    }
4500
4501    else if (another condition)
4502    {
4503       another action;
4504    }
4505
4506The braces can be omitted from simple one-line actions:
4507
4508    if (condition)
4509       return (0);
4510
4511We use 3-space indentation, except for continued statements which
4512are usually indented the same as the first line of the statement
4513plus four more spaces.
4514
4515For macro definitions we use 2-space indentation, always leaving the "#"
4516in the first column.
4517
4518    #ifndef PNG_NO_FEATURE
4519    #  ifndef PNG_FEATURE_SUPPORTED
4520    #    define PNG_FEATURE_SUPPORTED
4521    #  endif
4522    #endif
4523
4524Comments appear with the leading "/*" at the same indentation as
4525the statement that follows the comment:
4526
4527    /* Single-line comment */
4528    statement;
4529
4530    /* This is a multiple-line
4531     * comment.
4532     */
4533    statement;
4534
4535Very short comments can be placed after the end of the statement
4536to which they pertain:
4537
4538    statement;    /* comment */
4539
4540We don't use C++ style ("//") comments. We have, however,
4541used them in the past in some now-abandoned MMX assembler
4542code.
4543
4544Functions and their curly braces are not indented, and
4545exported functions are marked with PNGAPI:
4546
4547 /* This is a public function that is visible to
4548  * application programmers. It does thus-and-so.
4549  */
4550 void PNGAPI
4551 png_exported_function(png_ptr, png_info, foo)
4552 {
4553    body;
4554 }
4555
4556The prototypes for all exported functions appear in png.h,
4557above the comment that says
4558
4559    /* Maintainer: Put new public prototypes here ... */
4560
4561We mark all non-exported functions with "/* PRIVATE */"":
4562
4563 void /* PRIVATE */
4564 png_non_exported_function(png_ptr, png_info, foo)
4565 {
4566    body;
4567 }
4568
4569The prototypes for non-exported functions (except for those in
4570pngtest) appear in
4571pngpriv.h
4572above the comment that says
4573
4574  /* Maintainer: Put new private prototypes here ^ and in libpngpf.3 */
4575
4576To avoid polluting the global namespace, the names of all exported
4577functions and variables begin with "png_", and all publicly visible C
4578preprocessor macros begin with "PNG".  We request that applications that
4579use libpng *not* begin any of their own symbols with either of these strings.
4580
4581We put a space after each comma and after each semicolon
4582in "for" statements, and we put spaces before and after each
4583C binary operator and after "for" or "while", and before
4584"?".  We don't put a space between a typecast and the expression
4585being cast, nor do we put one between a function name and the
4586left parenthesis that follows it:
4587
4588    for (i = 2; i > 0; --i)
4589       y[i] = a(x) + (int)b;
4590
4591We prefer #ifdef and #ifndef to #if defined() and if !defined()
4592when there is only one macro being tested.
4593
4594We prefer to express integers that are used as bit masks in hex format,
4595with an even number of lower-case hex digits (e.g., 0x00, 0xff, 0x0100).
4596
4597We do not use the TAB character for indentation in the C sources.
4598
4599Lines do not exceed 80 characters.
4600
4601Other rules can be inferred by inspecting the libpng source.
4602
4603XIV. Y2K Compliance in libpng
4604
4605July 11, 2012
4606
4607Since the PNG Development group is an ad-hoc body, we can't make
4608an official declaration.
4609
4610This is your unofficial assurance that libpng from version 0.71 and
4611upward through 1.5.12 are Y2K compliant.  It is my belief that earlier
4612versions were also Y2K compliant.
4613
4614Libpng only has two year fields.  One is a 2-byte unsigned integer that
4615will hold years up to 65535.  The other holds the date in text
4616format, and will hold years up to 9999.
4617
4618The integer is
4619    "png_uint_16 year" in png_time_struct.
4620
4621The string is
4622    "char time_buffer[29]" in png_struct.  This will no
4623longer be used in libpng-1.6.x and will be removed from libpng-1.7.0.
4624
4625There are seven time-related functions:
4626
4627    png_convert_to_rfc_1123() in png.c
4628      (formerly png_convert_to_rfc_1152() in error)
4629    png_convert_from_struct_tm() in pngwrite.c, called
4630      in pngwrite.c
4631    png_convert_from_time_t() in pngwrite.c
4632    png_get_tIME() in pngget.c
4633    png_handle_tIME() in pngrutil.c, called in pngread.c
4634    png_set_tIME() in pngset.c
4635    png_write_tIME() in pngwutil.c, called in pngwrite.c
4636
4637All appear to handle dates properly in a Y2K environment.  The
4638png_convert_from_time_t() function calls gmtime() to convert from system
4639clock time, which returns (year - 1900), which we properly convert to
4640the full 4-digit year.  There is a possibility that applications using
4641libpng are not passing 4-digit years into the png_convert_to_rfc_1123()
4642function, or that they are incorrectly passing only a 2-digit year
4643instead of "year - 1900" into the png_convert_from_struct_tm() function,
4644but this is not under our control.  The libpng documentation has always
4645stated that it works with 4-digit years, and the APIs have been
4646documented as such.
4647
4648The tIME chunk itself is also Y2K compliant.  It uses a 2-byte unsigned
4649integer to hold the year, and can hold years as large as 65535.
4650
4651zlib, upon which libpng depends, is also Y2K compliant.  It contains
4652no date-related code.
4653
4654
4655   Glenn Randers-Pehrson
4656   libpng maintainer
4657   PNG Development Group
Note: See TracBrowser for help on using the repository browser.