all repos — mgba @ 9de8f084ba55460b02d300c1dd8b8e6c56f691d5

mGBA Game Boy Advance Emulator

src/third-party/libpng/libpng-manual.txt (view raw)

   1libpng-manual.txt - A description on how to use and modify libpng
   2
   3 libpng version 1.6.34 - September 29, 2017
   4 Updated and distributed by Glenn Randers-Pehrson
   5 <glennrp at users.sourceforge.net>
   6 Copyright (c) 1998-2017 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 1.6.34 - September 29, 2017
  15 Updated and distributed by Glenn Randers-Pehrson
  16 Copyright (c) 1998-2017 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
  31 TABLE OF CONTENTS
  32
  33    I. Introduction
  34   II. Structures
  35  III. Reading
  36   IV. Writing
  37    V. Simplified API
  38   VI. Modifying/Customizing libpng
  39  VII. MNG support
  40 VIII. Changes to Libpng from version 0.88
  41   IX. Changes to Libpng from version 1.0.x to 1.2.x
  42    X. Changes to Libpng from version 1.0.x/1.2.x to 1.4.x
  43   XI. Changes to Libpng from version 1.4.x to 1.5.x
  44  XII. Changes to Libpng from version 1.5.x to 1.6.x
  45 XIII. Detecting libpng
  46  XIV. Source code repository
  47   XV. Coding style
  48  XVI. Y2K Compliance in libpng
  49
  50I. Introduction
  51
  52This file describes how to use and modify the PNG reference library
  53(known as libpng) for your own use.  In addition to this
  54file, example.c is a good starting point for using the library, as
  55it is heavily commented and should include everything most people
  56will need.  We assume that libpng is already installed; see the
  57INSTALL file for instructions on how to configure and install libpng.
  58
  59For examples of libpng usage, see the files "example.c", "pngtest.c",
  60and the files in the "contrib" directory, all of which are included in
  61the libpng distribution.
  62
  63Libpng was written as a companion to the PNG specification, as a way
  64of reducing the amount of time and effort it takes to support the PNG
  65file format in application programs.
  66
  67The PNG specification (second edition), November 2003, is available as
  68a W3C Recommendation and as an ISO Standard (ISO/IEC 15948:2004 (E)) at
  69<https://www.w3.org/TR/2003/REC-PNG-20031110/
  70The W3C and ISO documents have identical technical content.
  71
  72The PNG-1.2 specification is available at
  73<https://png-mng.sourceforge.io/pub/png/spec/1.2/>.
  74It is technically equivalent
  75to the PNG specification (second edition) but has some additional material.
  76
  77The PNG-1.0 specification is available as RFC 2083
  78<https://png-mng.sourceforge.io/pub/png/spec/1.0/> and as a
  79W3C Recommendation <https://www.w3.org/TR/REC-png-961001>.
  80
  81Some additional chunks are described in the special-purpose public chunks
  82documents at <http://www.libpng.org/pub/png/spec/register/>
  83
  84Other information
  85about PNG, and the latest version of libpng, can be found at the PNG home
  86page, <http://www.libpng.org/pub/png/>.
  87
  88Most users will not have to modify the library significantly; advanced
  89users may want to modify it more.  All attempts were made to make it as
  90complete as possible, while keeping the code easy to understand.
  91Currently, this library only supports C.  Support for other languages
  92is being considered.
  93
  94Libpng has been designed to handle multiple sessions at one time,
  95to be easily modifiable, to be portable to the vast majority of
  96machines (ANSI, K&R, 16-, 32-, and 64-bit) available, and to be easy
  97to use.  The ultimate goal of libpng is to promote the acceptance of
  98the PNG file format in whatever way possible.  While there is still
  99work to be done (see the TODO file), libpng should cover the
 100majority of the needs of its users.
 101
 102Libpng uses zlib for its compression and decompression of PNG files.
 103Further information about zlib, and the latest version of zlib, can
 104be found at the zlib home page, <https://zlib.net/>.
 105The zlib compression utility is a general purpose utility that is
 106useful for more than PNG files, and can be used without libpng.
 107See the documentation delivered with zlib for more details.
 108You can usually find the source files for the zlib utility wherever you
 109find the libpng source files.
 110
 111Libpng is thread safe, provided the threads are using different
 112instances of the structures.  Each thread should have its own
 113png_struct and png_info instances, and thus its own image.
 114Libpng does not protect itself against two threads using the
 115same instance of a structure.
 116
 117II. Structures
 118
 119There are two main structures that are important to libpng, png_struct
 120and png_info.  Both are internal structures that are no longer exposed
 121in the libpng interface (as of libpng 1.5.0).
 122
 123The png_info structure is designed to provide information about the
 124PNG file.  At one time, the fields of png_info were intended to be
 125directly accessible to the user.  However, this tended to cause problems
 126with applications using dynamically loaded libraries, and as a result
 127a set of interface functions for png_info (the png_get_*() and png_set_*()
 128functions) was developed, and direct access to the png_info fields was
 129deprecated..
 130
 131The png_struct structure is the object used by the library to decode a
 132single image.  As of 1.5.0 this structure is also not exposed.
 133
 134Almost all libpng APIs require a pointer to a png_struct as the first argument.
 135Many (in particular the png_set and png_get APIs) also require a pointer
 136to png_info as the second argument.  Some application visible macros
 137defined in png.h designed for basic data access (reading and writing
 138integers in the PNG format) don't take a png_info pointer, but it's almost
 139always safe to assume that a (png_struct*) has to be passed to call an API
 140function.
 141
 142You can have more than one png_info structure associated with an image,
 143as illustrated in pngtest.c, one for information valid prior to the
 144IDAT chunks and another (called "end_info" below) for things after them.
 145
 146The png.h header file is an invaluable reference for programming with libpng.
 147And while I'm on the topic, make sure you include the libpng header file:
 148
 149#include <png.h>
 150
 151and also (as of libpng-1.5.0) the zlib header file, if you need it:
 152
 153#include <zlib.h>
 154
 155Types
 156
 157The png.h header file defines a number of integral types used by the
 158APIs.  Most of these are fairly obvious; for example types corresponding
 159to integers of particular sizes and types for passing color values.
 160
 161One exception is how non-integral numbers are handled.  For application
 162convenience most APIs that take such numbers have C (double) arguments;
 163however, internally PNG, and libpng, use 32 bit signed integers and encode
 164the value by multiplying by 100,000.  As of libpng 1.5.0 a convenience
 165macro PNG_FP_1 is defined in png.h along with a type (png_fixed_point)
 166which is simply (png_int_32).
 167
 168All APIs that take (double) arguments also have a matching API that
 169takes the corresponding fixed point integer arguments.  The fixed point
 170API has the same name as the floating point one with "_fixed" appended.
 171The actual range of values permitted in the APIs is frequently less than
 172the full range of (png_fixed_point) (-21474 to +21474).  When APIs require
 173a non-negative argument the type is recorded as png_uint_32 above.  Consult
 174the header file and the text below for more information.
 175
 176Special care must be take with sCAL chunk handling because the chunk itself
 177uses non-integral values encoded as strings containing decimal floating point
 178numbers.  See the comments in the header file.
 179
 180Configuration
 181
 182The main header file function declarations are frequently protected by C
 183preprocessing directives of the form:
 184
 185    #ifdef PNG_feature_SUPPORTED
 186    declare-function
 187    #endif
 188    ...
 189    #ifdef PNG_feature_SUPPORTED
 190    use-function
 191    #endif
 192
 193The library can be built without support for these APIs, although a
 194standard build will have all implemented APIs.  Application programs
 195should check the feature macros before using an API for maximum
 196portability.  From libpng 1.5.0 the feature macros set during the build
 197of libpng are recorded in the header file "pnglibconf.h" and this file
 198is always included by png.h.
 199
 200If you don't need to change the library configuration from the default, skip to
 201the next section ("Reading").
 202
 203Notice that some of the makefiles in the 'scripts' directory and (in 1.5.0) all
 204of the build project files in the 'projects' directory simply copy
 205scripts/pnglibconf.h.prebuilt to pnglibconf.h.  This means that these build
 206systems do not permit easy auto-configuration of the library - they only
 207support the default configuration.
 208
 209The easiest way to make minor changes to the libpng configuration when
 210auto-configuration is supported is to add definitions to the command line
 211using (typically) CPPFLAGS.  For example:
 212
 213CPPFLAGS=-DPNG_NO_FLOATING_ARITHMETIC
 214
 215will change the internal libpng math implementation for gamma correction and
 216other arithmetic calculations to fixed point, avoiding the need for fast
 217floating point support.  The result can be seen in the generated pnglibconf.h -
 218make sure it contains the changed feature macro setting.
 219
 220If you need to make more extensive configuration changes - more than one or two
 221feature macro settings - you can either add -DPNG_USER_CONFIG to the build
 222command line and put a list of feature macro settings in pngusr.h or you can set
 223DFA_XTRA (a makefile variable) to a file containing the same information in the
 224form of 'option' settings.
 225
 226A. Changing pnglibconf.h
 227
 228A variety of methods exist to build libpng.  Not all of these support
 229reconfiguration of pnglibconf.h.  To reconfigure pnglibconf.h it must either be
 230rebuilt from scripts/pnglibconf.dfa using awk or it must be edited by hand.
 231
 232Hand editing is achieved by copying scripts/pnglibconf.h.prebuilt to
 233pnglibconf.h and changing the lines defining the supported features, paying
 234very close attention to the 'option' information in scripts/pnglibconf.dfa
 235that describes those features and their requirements.  This is easy to get
 236wrong.
 237
 238B. Configuration using DFA_XTRA
 239
 240Rebuilding from pnglibconf.dfa is easy if a functioning 'awk', or a later
 241variant such as 'nawk' or 'gawk', is available.  The configure build will
 242automatically find an appropriate awk and build pnglibconf.h.
 243The scripts/pnglibconf.mak file contains a set of make rules for doing the
 244same thing if configure is not used, and many of the makefiles in the scripts
 245directory use this approach.
 246
 247When rebuilding simply write a new file containing changed options and set
 248DFA_XTRA to the name of this file.  This causes the build to append the new file
 249to the end of scripts/pnglibconf.dfa.  The pngusr.dfa file should contain lines
 250of the following forms:
 251
 252everything = off
 253
 254This turns all optional features off.  Include it at the start of pngusr.dfa to
 255make it easier to build a minimal configuration.  You will need to turn at least
 256some features on afterward to enable either reading or writing code, or both.
 257
 258option feature on
 259option feature off
 260
 261Enable or disable a single feature.  This will automatically enable other
 262features required by a feature that is turned on or disable other features that
 263require a feature which is turned off.  Conflicting settings will cause an error
 264message to be emitted by awk.
 265
 266setting feature default value
 267
 268Changes the default value of setting 'feature' to 'value'.  There are a small
 269number of settings listed at the top of pnglibconf.h, they are documented in the
 270source code.  Most of these values have performance implications for the library
 271but most of them have no visible effect on the API.  Some can also be overridden
 272from the API.
 273
 274This method of building a customized pnglibconf.h is illustrated in
 275contrib/pngminim/*.  See the "$(PNGCONF):" target in the makefile and
 276pngusr.dfa in these directories.
 277
 278C. Configuration using PNG_USER_CONFIG
 279
 280If -DPNG_USER_CONFIG is added to the CPPFLAGS when pnglibconf.h is built,
 281the file pngusr.h will automatically be included before the options in
 282scripts/pnglibconf.dfa are processed.  Your pngusr.h file should contain only
 283macro definitions turning features on or off or setting settings.
 284
 285Apart from the global setting "everything = off" all the options listed above
 286can be set using macros in pngusr.h:
 287
 288#define PNG_feature_SUPPORTED
 289
 290is equivalent to:
 291
 292option feature on
 293
 294#define PNG_NO_feature
 295
 296is equivalent to:
 297
 298option feature off
 299
 300#define PNG_feature value
 301
 302is equivalent to:
 303
 304setting feature default value
 305
 306Notice that in both cases, pngusr.dfa and pngusr.h, the contents of the
 307pngusr file you supply override the contents of scripts/pnglibconf.dfa
 308
 309If confusing or incomprehensible behavior results it is possible to
 310examine the intermediate file pnglibconf.dfn to find the full set of
 311dependency information for each setting and option.  Simply locate the
 312feature in the file and read the C comments that precede it.
 313
 314This method is also illustrated in the contrib/pngminim/* makefiles and
 315pngusr.h.
 316
 317III. Reading
 318
 319We'll now walk you through the possible functions to call when reading
 320in a PNG file sequentially, briefly explaining the syntax and purpose
 321of each one.  See example.c and png.h for more detail.  While
 322progressive reading is covered in the next section, you will still
 323need some of the functions discussed in this section to read a PNG
 324file.
 325
 326Setup
 327
 328You will want to do the I/O initialization(*) before you get into libpng,
 329so if it doesn't work, you don't have much to undo.  Of course, you
 330will also want to insure that you are, in fact, dealing with a PNG
 331file.  Libpng provides a simple check to see if a file is a PNG file.
 332To use it, pass in the first 1 to 8 bytes of the file to the function
 333png_sig_cmp(), and it will return 0 (false) if the bytes match the
 334corresponding bytes of the PNG signature, or nonzero (true) otherwise.
 335Of course, the more bytes you pass in, the greater the accuracy of the
 336prediction.
 337
 338If you are intending to keep the file pointer open for use in libpng,
 339you must ensure you don't read more than 8 bytes from the beginning
 340of the file, and you also have to make a call to png_set_sig_bytes()
 341with the number of bytes you read from the beginning.  Libpng will
 342then only check the bytes (if any) that your program didn't read.
 343
 344(*): If you are not using the standard I/O functions, you will need
 345to replace them with custom functions.  See the discussion under
 346Customizing libpng.
 347
 348    FILE *fp = fopen(file_name, "rb");
 349    if (!fp)
 350    {
 351       return (ERROR);
 352    }
 353
 354    if (fread(header, 1, number, fp) != number)
 355    {
 356       return (ERROR);
 357    }
 358
 359    is_png = !png_sig_cmp(header, 0, number);
 360    if (!is_png)
 361    {
 362       return (NOT_PNG);
 363    }
 364
 365Next, png_struct and png_info need to be allocated and initialized.  In
 366order to ensure that the size of these structures is correct even with a
 367dynamically linked libpng, there are functions to initialize and
 368allocate the structures.  We also pass the library version, optional
 369pointers to error handling functions, and a pointer to a data struct for
 370use by the error functions, if necessary (the pointer and functions can
 371be NULL if the default error handlers are to be used).  See the section
 372on Changes to Libpng below regarding the old initialization functions.
 373The structure allocation functions quietly return NULL if they fail to
 374create the structure, so your application should check for that.
 375
 376    png_structp png_ptr = png_create_read_struct
 377        (PNG_LIBPNG_VER_STRING, (png_voidp)user_error_ptr,
 378        user_error_fn, user_warning_fn);
 379
 380    if (!png_ptr)
 381       return (ERROR);
 382
 383    png_infop info_ptr = png_create_info_struct(png_ptr);
 384
 385    if (!info_ptr)
 386    {
 387       png_destroy_read_struct(&png_ptr,
 388           (png_infopp)NULL, (png_infopp)NULL);
 389       return (ERROR);
 390    }
 391
 392If you want to use your own memory allocation routines,
 393use a libpng that was built with PNG_USER_MEM_SUPPORTED defined, and use
 394png_create_read_struct_2() instead of png_create_read_struct():
 395
 396    png_structp png_ptr = png_create_read_struct_2
 397        (PNG_LIBPNG_VER_STRING, (png_voidp)user_error_ptr,
 398        user_error_fn, user_warning_fn, (png_voidp)
 399        user_mem_ptr, user_malloc_fn, user_free_fn);
 400
 401The error handling routines passed to png_create_read_struct()
 402and the memory alloc/free routines passed to png_create_struct_2()
 403are only necessary if you are not using the libpng supplied error
 404handling and memory alloc/free functions.
 405
 406When libpng encounters an error, it expects to longjmp back
 407to your routine.  Therefore, you will need to call setjmp and pass
 408your png_jmpbuf(png_ptr).  If you read the file from different
 409routines, you will need to update the longjmp buffer every time you enter
 410a new routine that will call a png_*() function.
 411
 412See your documentation of setjmp/longjmp for your compiler for more
 413information on setjmp/longjmp.  See the discussion on libpng error
 414handling in the Customizing Libpng section below for more information
 415on the libpng error handling.  If an error occurs, and libpng longjmp's
 416back to your setjmp, you will want to call png_destroy_read_struct() to
 417free any memory.
 418
 419    if (setjmp(png_jmpbuf(png_ptr)))
 420    {
 421       png_destroy_read_struct(&png_ptr, &info_ptr,
 422           &end_info);
 423       fclose(fp);
 424       return (ERROR);
 425    }
 426
 427Pass (png_infopp)NULL instead of &end_info if you didn't create
 428an end_info structure.
 429
 430If you would rather avoid the complexity of setjmp/longjmp issues,
 431you can compile libpng with PNG_NO_SETJMP, in which case
 432errors will result in a call to PNG_ABORT() which defaults to abort().
 433
 434You can #define PNG_ABORT() to a function that does something
 435more useful than abort(), as long as your function does not
 436return.
 437
 438Now you need to set up the input code.  The default for libpng is to
 439use the C function fread().  If you use this, you will need to pass a
 440valid FILE * in the function png_init_io().  Be sure that the file is
 441opened in binary mode.  If you wish to handle reading data in another
 442way, you need not call the png_init_io() function, but you must then
 443implement the libpng I/O methods discussed in the Customizing Libpng
 444section below.
 445
 446    png_init_io(png_ptr, fp);
 447
 448If you had previously opened the file and read any of the signature from
 449the beginning in order to see if this was a PNG file, you need to let
 450libpng know that there are some bytes missing from the start of the file.
 451
 452    png_set_sig_bytes(png_ptr, number);
 453
 454You can change the zlib compression buffer size to be used while
 455reading compressed data with
 456
 457    png_set_compression_buffer_size(png_ptr, buffer_size);
 458
 459where the default size is 8192 bytes.  Note that the buffer size
 460is changed immediately and the buffer is reallocated immediately,
 461instead of setting a flag to be acted upon later.
 462
 463If you want CRC errors to be handled in a different manner than
 464the default, use
 465
 466    png_set_crc_action(png_ptr, crit_action, ancil_action);
 467
 468The values for png_set_crc_action() say how libpng is to handle CRC errors in
 469ancillary and critical chunks, and whether to use the data contained
 470therein. Starting with libpng-1.6.26, this also governs how an ADLER32 error
 471is handled while reading the IDAT chunk. Note that it is impossible to
 472"discard" data in a critical chunk.
 473
 474Choices for (int) crit_action are
 475   PNG_CRC_DEFAULT      0  error/quit
 476   PNG_CRC_ERROR_QUIT   1  error/quit
 477   PNG_CRC_WARN_USE     3  warn/use data
 478   PNG_CRC_QUIET_USE    4  quiet/use data
 479   PNG_CRC_NO_CHANGE    5  use the current value
 480
 481Choices for (int) ancil_action are
 482   PNG_CRC_DEFAULT      0  error/quit
 483   PNG_CRC_ERROR_QUIT   1  error/quit
 484   PNG_CRC_WARN_DISCARD 2  warn/discard data
 485   PNG_CRC_WARN_USE     3  warn/use data
 486   PNG_CRC_QUIET_USE    4  quiet/use data
 487   PNG_CRC_NO_CHANGE    5  use the current value
 488
 489When the setting for crit_action is PNG_CRC_QUIET_USE, the CRC and ADLER32
 490checksums are not only ignored, but they are not evaluated.
 491
 492Setting up callback code
 493
 494You can set up a callback function to handle any unknown chunks in the
 495input stream. You must supply the function
 496
 497    read_chunk_callback(png_structp png_ptr,
 498         png_unknown_chunkp chunk);
 499    {
 500       /* The unknown chunk structure contains your
 501          chunk data, along with similar data for any other
 502          unknown chunks: */
 503
 504           png_byte name[5];
 505           png_byte *data;
 506           png_size_t size;
 507
 508       /* Note that libpng has already taken care of
 509          the CRC handling */
 510
 511       /* put your code here.  Search for your chunk in the
 512          unknown chunk structure, process it, and return one
 513          of the following: */
 514
 515       return (-n); /* chunk had an error */
 516       return (0); /* did not recognize */
 517       return (n); /* success */
 518    }
 519
 520(You can give your function another name that you like instead of
 521"read_chunk_callback")
 522
 523To inform libpng about your function, use
 524
 525    png_set_read_user_chunk_fn(png_ptr, user_chunk_ptr,
 526        read_chunk_callback);
 527
 528This names not only the callback function, but also a user pointer that
 529you can retrieve with
 530
 531    png_get_user_chunk_ptr(png_ptr);
 532
 533If you call the png_set_read_user_chunk_fn() function, then all unknown
 534chunks which the callback does not handle will be saved when read.  You can
 535cause them to be discarded by returning '1' ("handled") instead of '0'.  This
 536behavior will change in libpng 1.7 and the default handling set by the
 537png_set_keep_unknown_chunks() function, described below, will be used when the
 538callback returns 0.  If you want the existing behavior you should set the global
 539default to PNG_HANDLE_CHUNK_IF_SAFE now; this is compatible with all current
 540versions of libpng and with 1.7.  Libpng 1.6 issues a warning if you keep the
 541default, or PNG_HANDLE_CHUNK_NEVER, and the callback returns 0.
 542
 543At this point, you can set up a callback function that will be
 544called after each row has been read, which you can use to control
 545a progress meter or the like.  It's demonstrated in pngtest.c.
 546You must supply a function
 547
 548    void read_row_callback(png_structp png_ptr,
 549       png_uint_32 row, int pass);
 550    {
 551      /* put your code here */
 552    }
 553
 554(You can give it another name that you like instead of "read_row_callback")
 555
 556To inform libpng about your function, use
 557
 558    png_set_read_status_fn(png_ptr, read_row_callback);
 559
 560When this function is called the row has already been completely processed and
 561the 'row' and 'pass' refer to the next row to be handled.  For the
 562non-interlaced case the row that was just handled is simply one less than the
 563passed in row number, and pass will always be 0.  For the interlaced case the
 564same applies unless the row value is 0, in which case the row just handled was
 565the last one from one of the preceding passes.  Because interlacing may skip a
 566pass you cannot be sure that the preceding pass is just 'pass-1'; if you really
 567need to know what the last pass is record (row,pass) from the callback and use
 568the last recorded value each time.
 569
 570As with the user transform you can find the output row using the
 571PNG_ROW_FROM_PASS_ROW macro.
 572
 573Unknown-chunk handling
 574
 575Now you get to set the way the library processes unknown chunks in the
 576input PNG stream. Both known and unknown chunks will be read.  Normal
 577behavior is that known chunks will be parsed into information in
 578various info_ptr members while unknown chunks will be discarded. This
 579behavior can be wasteful if your application will never use some known
 580chunk types. To change this, you can call:
 581
 582    png_set_keep_unknown_chunks(png_ptr, keep,
 583        chunk_list, num_chunks);
 584
 585    keep       - 0: default unknown chunk handling
 586                 1: ignore; do not keep
 587                 2: keep only if safe-to-copy
 588                 3: keep even if unsafe-to-copy
 589
 590               You can use these definitions:
 591                 PNG_HANDLE_CHUNK_AS_DEFAULT   0
 592                 PNG_HANDLE_CHUNK_NEVER        1
 593                 PNG_HANDLE_CHUNK_IF_SAFE      2
 594                 PNG_HANDLE_CHUNK_ALWAYS       3
 595
 596    chunk_list - list of chunks affected (a byte string,
 597                 five bytes per chunk, NULL or '\0' if
 598                 num_chunks is positive; ignored if
 599                 numchunks <= 0).
 600
 601    num_chunks - number of chunks affected; if 0, all
 602                 unknown chunks are affected.  If positive,
 603                 only the chunks in the list are affected,
 604                 and if negative all unknown chunks and
 605                 all known chunks except for the IHDR,
 606                 PLTE, tRNS, IDAT, and IEND chunks are
 607                 affected.
 608
 609Unknown chunks declared in this way will be saved as raw data onto a
 610list of png_unknown_chunk structures.  If a chunk that is normally
 611known to libpng is named in the list, it will be handled as unknown,
 612according to the "keep" directive.  If a chunk is named in successive
 613instances of png_set_keep_unknown_chunks(), the final instance will
 614take precedence.  The IHDR and IEND chunks should not be named in
 615chunk_list; if they are, libpng will process them normally anyway.
 616If you know that your application will never make use of some particular
 617chunks, use PNG_HANDLE_CHUNK_NEVER (or 1) as demonstrated below.
 618
 619Here is an example of the usage of png_set_keep_unknown_chunks(),
 620where the private "vpAg" chunk will later be processed by a user chunk
 621callback function:
 622
 623    png_byte vpAg[5]={118, 112,  65, 103, (png_byte) '\0'};
 624
 625    #if defined(PNG_UNKNOWN_CHUNKS_SUPPORTED)
 626      png_byte unused_chunks[]=
 627      {
 628        104,  73,  83,  84, (png_byte) '\0',   /* hIST */
 629        105,  84,  88, 116, (png_byte) '\0',   /* iTXt */
 630        112,  67,  65,  76, (png_byte) '\0',   /* pCAL */
 631        115,  67,  65,  76, (png_byte) '\0',   /* sCAL */
 632        115,  80,  76,  84, (png_byte) '\0',   /* sPLT */
 633        116,  73,  77,  69, (png_byte) '\0',   /* tIME */
 634      };
 635    #endif
 636
 637    ...
 638
 639    #if defined(PNG_UNKNOWN_CHUNKS_SUPPORTED)
 640      /* ignore all unknown chunks
 641       * (use global setting "2" for libpng16 and earlier):
 642       */
 643      png_set_keep_unknown_chunks(read_ptr, 2, NULL, 0);
 644
 645      /* except for vpAg: */
 646      png_set_keep_unknown_chunks(read_ptr, 2, vpAg, 1);
 647
 648      /* also ignore unused known chunks: */
 649      png_set_keep_unknown_chunks(read_ptr, 1, unused_chunks,
 650         (int)(sizeof unused_chunks)/5);
 651    #endif
 652
 653User limits
 654
 655The PNG specification allows the width and height of an image to be as
 656large as 2^31-1 (0x7fffffff), or about 2.147 billion rows and columns.
 657For safety, libpng imposes a default limit of 1 million rows and columns.
 658Larger images will be rejected immediately with a png_error() call. If
 659you wish to change these limits, you can use
 660
 661   png_set_user_limits(png_ptr, width_max, height_max);
 662
 663to set your own limits (libpng may reject some very wide images
 664anyway because of potential buffer overflow conditions).
 665
 666You should put this statement after you create the PNG structure and
 667before calling png_read_info(), png_read_png(), or png_process_data().
 668
 669When writing a PNG datastream, put this statement before calling
 670png_write_info() or png_write_png().
 671
 672If you need to retrieve the limits that are being applied, use
 673
 674   width_max = png_get_user_width_max(png_ptr);
 675   height_max = png_get_user_height_max(png_ptr);
 676
 677The PNG specification sets no limit on the number of ancillary chunks
 678allowed in a PNG datastream.  By default, libpng imposes a limit of
 679a total of 1000 sPLT, tEXt, iTXt, zTXt, and unknown chunks to be stored.
 680If you have set up both info_ptr and end_info_ptr, the limit applies
 681separately to each.  You can change the limit on the total number of such
 682chunks that will be stored, with
 683
 684   png_set_chunk_cache_max(png_ptr, user_chunk_cache_max);
 685
 686where 0x7fffffffL means unlimited.  You can retrieve this limit with
 687
 688   chunk_cache_max = png_get_chunk_cache_max(png_ptr);
 689
 690Libpng imposes a limit of 8 Megabytes (8,000,000 bytes) on the amount of
 691memory that any chunk other than IDAT can occupy, originally or when
 692decompressed (prior to libpng-1.6.32 the limit was only applied to compressed
 693chunks after decompression). You can change this limit with
 694
 695   png_set_chunk_malloc_max(png_ptr, user_chunk_malloc_max);
 696
 697and you can retrieve the limit with
 698
 699   chunk_malloc_max = png_get_chunk_malloc_max(png_ptr);
 700
 701Any chunks that would cause either of these limits to be exceeded will
 702be ignored.
 703
 704Information about your system
 705
 706If you intend to display the PNG or to incorporate it in other image data you
 707need to tell libpng information about your display or drawing surface so that
 708libpng can convert the values in the image to match the display.
 709
 710From libpng-1.5.4 this information can be set before reading the PNG file
 711header.  In earlier versions png_set_gamma() existed but behaved incorrectly if
 712called before the PNG file header had been read and png_set_alpha_mode() did not
 713exist.
 714
 715If you need to support versions prior to libpng-1.5.4 test the version number
 716as illustrated below using "PNG_LIBPNG_VER >= 10504" and follow the procedures
 717described in the appropriate manual page.
 718
 719You give libpng the encoding expected by your system expressed as a 'gamma'
 720value.  You can also specify a default encoding for the PNG file in
 721case the required information is missing from the file.  By default libpng
 722assumes that the PNG data matches your system, to keep this default call:
 723
 724   png_set_gamma(png_ptr, screen_gamma, output_gamma);
 725
 726or you can use the fixed point equivalent:
 727
 728   png_set_gamma_fixed(png_ptr, PNG_FP_1*screen_gamma,
 729      PNG_FP_1*output_gamma);
 730
 731If you don't know the gamma for your system it is probably 2.2 - a good
 732approximation to the IEC standard for display systems (sRGB).  If images are
 733too contrasty or washed out you got the value wrong - check your system
 734documentation!
 735
 736Many systems permit the system gamma to be changed via a lookup table in the
 737display driver, a few systems, including older Macs, change the response by
 738default.  As of 1.5.4 three special values are available to handle common
 739situations:
 740
 741   PNG_DEFAULT_sRGB: Indicates that the system conforms to the
 742                     IEC 61966-2-1 standard.  This matches almost
 743                     all systems.
 744   PNG_GAMMA_MAC_18: Indicates that the system is an older
 745                     (pre Mac OS 10.6) Apple Macintosh system with
 746                     the default settings.
 747   PNG_GAMMA_LINEAR: Just the fixed point value for 1.0 - indicates
 748                     that the system expects data with no gamma
 749                     encoding.
 750
 751You would use the linear (unencoded) value if you need to process the pixel
 752values further because this avoids the need to decode and re-encode each
 753component value whenever arithmetic is performed.  A lot of graphics software
 754uses linear values for this reason, often with higher precision component values
 755to preserve overall accuracy.
 756
 757
 758The output_gamma value expresses how to decode the output values, not how
 759they are encoded.  The values used correspond to the normal numbers used to
 760describe the overall gamma of a computer display system; for example 2.2 for
 761an sRGB conformant system.  The values are scaled by 100000 in the _fixed
 762version of the API (so 220000 for sRGB.)
 763
 764The inverse of the value is always used to provide a default for the PNG file
 765encoding if it has no gAMA chunk and if png_set_gamma() has not been called
 766to override the PNG gamma information.
 767
 768When the ALPHA_OPTIMIZED mode is selected the output gamma is used to encode
 769opaque pixels however pixels with lower alpha values are not encoded,
 770regardless of the output gamma setting.
 771
 772When the standard Porter Duff handling is requested with mode 1 the output
 773encoding is set to be linear and the output_gamma value is only relevant
 774as a default for input data that has no gamma information.  The linear output
 775encoding will be overridden if png_set_gamma() is called - the results may be
 776highly unexpected!
 777
 778The following numbers are derived from the sRGB standard and the research
 779behind it.  sRGB is defined to be approximated by a PNG gAMA chunk value of
 7800.45455 (1/2.2) for PNG.  The value implicitly includes any viewing
 781correction required to take account of any differences in the color
 782environment of the original scene and the intended display environment; the
 783value expresses how to *decode* the image for display, not how the original
 784data was *encoded*.
 785
 786sRGB provides a peg for the PNG standard by defining a viewing environment.
 787sRGB itself, and earlier TV standards, actually use a more complex transform
 788(a linear portion then a gamma 2.4 power law) than PNG can express.  (PNG is
 789limited to simple power laws.)  By saying that an image for direct display on
 790an sRGB conformant system should be stored with a gAMA chunk value of 45455
 791(11.3.3.2 and 11.3.3.5 of the ISO PNG specification) the PNG specification
 792makes it possible to derive values for other display systems and
 793environments.
 794
 795The Mac value is deduced from the sRGB based on an assumption that the actual
 796extra viewing correction used in early Mac display systems was implemented as
 797a power 1.45 lookup table.
 798
 799Any system where a programmable lookup table is used or where the behavior of
 800the final display device characteristics can be changed requires system
 801specific code to obtain the current characteristic.  However this can be
 802difficult and most PNG gamma correction only requires an approximate value.
 803
 804By default, if png_set_alpha_mode() is not called, libpng assumes that all
 805values are unencoded, linear, values and that the output device also has a
 806linear characteristic.  This is only very rarely correct - it is invariably
 807better to call png_set_alpha_mode() with PNG_DEFAULT_sRGB than rely on the
 808default if you don't know what the right answer is!
 809
 810The special value PNG_GAMMA_MAC_18 indicates an older Mac system (pre Mac OS
 81110.6) which used a correction table to implement a somewhat lower gamma on an
 812otherwise sRGB system.
 813
 814Both these values are reserved (not simple gamma values) in order to allow
 815more precise correction internally in the future.
 816
 817NOTE: the values can be passed to either the fixed or floating
 818point APIs, but the floating point API will also accept floating point
 819values.
 820
 821The second thing you may need to tell libpng about is how your system handles
 822alpha channel information.  Some, but not all, PNG files contain an alpha
 823channel.  To display these files correctly you need to compose the data onto a
 824suitable background, as described in the PNG specification.
 825
 826Libpng only supports composing onto a single color (using png_set_background;
 827see below).  Otherwise you must do the composition yourself and, in this case,
 828you may need to call png_set_alpha_mode:
 829
 830   #if PNG_LIBPNG_VER >= 10504
 831      png_set_alpha_mode(png_ptr, mode, screen_gamma);
 832   #else
 833      png_set_gamma(png_ptr, screen_gamma, 1.0/screen_gamma);
 834   #endif
 835
 836The screen_gamma value is the same as the argument to png_set_gamma; however,
 837how it affects the output depends on the mode.  png_set_alpha_mode() sets the
 838file gamma default to 1/screen_gamma, so normally you don't need to call
 839png_set_gamma.  If you need different defaults call png_set_gamma() before
 840png_set_alpha_mode() - if you call it after it will override the settings made
 841by png_set_alpha_mode().
 842
 843The mode is as follows:
 844
 845    PNG_ALPHA_PNG: The data is encoded according to the PNG
 846specification.  Red, green and blue, or gray, components are
 847gamma encoded color values and are not premultiplied by the
 848alpha value.  The alpha value is a linear measure of the
 849contribution of the pixel to the corresponding final output pixel.
 850
 851You should normally use this format if you intend to perform
 852color correction on the color values; most, maybe all, color
 853correction software has no handling for the alpha channel and,
 854anyway, the math to handle pre-multiplied component values is
 855unnecessarily complex.
 856
 857Before you do any arithmetic on the component values you need
 858to remove the gamma encoding and multiply out the alpha
 859channel.  See the PNG specification for more detail.  It is
 860important to note that when an image with an alpha channel is
 861scaled, linear encoded, pre-multiplied component values must
 862be used!
 863
 864The remaining modes assume you don't need to do any further color correction or
 865that if you do, your color correction software knows all about alpha (it
 866probably doesn't!).  They 'associate' the alpha with the color information by
 867storing color channel values that have been scaled by the alpha.  The
 868advantage is that the color channels can be resampled (the image can be
 869scaled) in this form.  The disadvantage is that normal practice is to store
 870linear, not (gamma) encoded, values and this requires 16-bit channels for
 871still images rather than the 8-bit channels that are just about sufficient if
 872gamma encoding is used.  In addition all non-transparent pixel values,
 873including completely opaque ones, must be gamma encoded to produce the final
 874image.  These are the 'STANDARD', 'ASSOCIATED' or 'PREMULTIPLIED' modes
 875described below (the latter being the two common names for associated alpha
 876color channels). Note that PNG files always contain non-associated color
 877channels; png_set_alpha_mode() with one of the modes causes the decoder to
 878convert the pixels to an associated form before returning them to your
 879application. 
 880
 881Since it is not necessary to perform arithmetic on opaque color values so
 882long as they are not to be resampled and are in the final color space it is
 883possible to optimize the handling of alpha by storing the opaque pixels in
 884the PNG format (adjusted for the output color space) while storing partially
 885opaque pixels in the standard, linear, format.  The accuracy required for
 886standard alpha composition is relatively low, because the pixels are
 887isolated, therefore typically the accuracy loss in storing 8-bit linear
 888values is acceptable.  (This is not true if the alpha channel is used to
 889simulate transparency over large areas - use 16 bits or the PNG mode in
 890this case!)  This is the 'OPTIMIZED' mode.  For this mode a pixel is
 891treated as opaque only if the alpha value is equal to the maximum value.
 892
 893    PNG_ALPHA_STANDARD:  The data libpng produces is encoded in the
 894standard way assumed by most correctly written graphics software.
 895The gamma encoding will be removed by libpng and the
 896linear component values will be pre-multiplied by the
 897alpha channel.
 898
 899With this format the final image must be re-encoded to
 900match the display gamma before the image is displayed.
 901If your system doesn't do that, yet still seems to
 902perform arithmetic on the pixels without decoding them,
 903it is broken - check out the modes below.
 904
 905With PNG_ALPHA_STANDARD libpng always produces linear
 906component values, whatever screen_gamma you supply.  The
 907screen_gamma value is, however, used as a default for
 908the file gamma if the PNG file has no gamma information.
 909
 910If you call png_set_gamma() after png_set_alpha_mode() you
 911will override the linear encoding.  Instead the
 912pre-multiplied pixel values will be gamma encoded but
 913the alpha channel will still be linear.  This may
 914actually match the requirements of some broken software,
 915but it is unlikely.
 916
 917While linear 8-bit data is often used it has
 918insufficient precision for any image with a reasonable
 919dynamic range.  To avoid problems, and if your software
 920supports it, use png_set_expand_16() to force all
 921components to 16 bits.
 922
 923    PNG_ALPHA_OPTIMIZED: This mode is the same as PNG_ALPHA_STANDARD
 924except that completely opaque pixels are gamma encoded according to
 925the screen_gamma value.  Pixels with alpha less than 1.0
 926will still have linear components.
 927
 928Use this format if you have control over your
 929compositing software and so don't do other arithmetic
 930(such as scaling) on the data you get from libpng.  Your
 931compositing software can simply copy opaque pixels to
 932the output but still has linear values for the
 933non-opaque pixels.
 934
 935In normal compositing, where the alpha channel encodes
 936partial pixel coverage (as opposed to broad area
 937translucency), the inaccuracies of the 8-bit
 938representation of non-opaque pixels are irrelevant.
 939
 940You can also try this format if your software is broken;
 941it might look better.
 942
 943    PNG_ALPHA_BROKEN: This is PNG_ALPHA_STANDARD; however, all component
 944values, including the alpha channel are gamma encoded.  This is
 945broken because, in practice, no implementation that uses this choice
 946correctly undoes the encoding before handling alpha composition.  Use this
 947choice only if other serious errors in the software or hardware you use
 948mandate it.  In most cases of broken software or hardware the bug in the
 949final display manifests as a subtle halo around composited parts of the
 950image.  You may not even perceive this as a halo; the composited part of
 951the image may simply appear separate from the background, as though it had
 952been cut out of paper and pasted on afterward.
 953
 954If you don't have to deal with bugs in software or hardware, or if you can fix
 955them, there are three recommended ways of using png_set_alpha_mode():
 956
 957   png_set_alpha_mode(png_ptr, PNG_ALPHA_PNG,
 958       screen_gamma);
 959
 960You can do color correction on the result (libpng does not currently
 961support color correction internally).  When you handle the alpha channel
 962you need to undo the gamma encoding and multiply out the alpha.
 963
 964   png_set_alpha_mode(png_ptr, PNG_ALPHA_STANDARD,
 965       screen_gamma);
 966   png_set_expand_16(png_ptr);
 967
 968If you are using the high level interface, don't call png_set_expand_16();
 969instead pass PNG_TRANSFORM_EXPAND_16 to the interface.
 970
 971With this mode you can't do color correction, but you can do arithmetic,
 972including composition and scaling, on the data without further processing.
 973
 974   png_set_alpha_mode(png_ptr, PNG_ALPHA_OPTIMIZED,
 975       screen_gamma);
 976
 977You can avoid the expansion to 16-bit components with this mode, but you
 978lose the ability to scale the image or perform other linear arithmetic.
 979All you can do is compose the result onto a matching output.  Since this
 980mode is libpng-specific you also need to write your own composition
 981software.
 982
 983The following are examples of calls to png_set_alpha_mode to achieve the
 984required overall gamma correction and, where necessary, alpha
 985premultiplication.
 986
 987    png_set_alpha_mode(pp, PNG_ALPHA_PNG, PNG_DEFAULT_sRGB);
 988
 989Choices for the alpha_mode are
 990
 991    PNG_ALPHA_PNG           0 /* according to the PNG standard */
 992    PNG_ALPHA_STANDARD      1 /* according to Porter/Duff */
 993    PNG_ALPHA_ASSOCIATED    1 /* as above; this is the normal practice */
 994    PNG_ALPHA_PREMULTIPLIED 1 /* as above */
 995    PNG_ALPHA_OPTIMIZED     2 /* 'PNG' for opaque pixels, else 'STANDARD' */
 996    PNG_ALPHA_BROKEN        3 /* the alpha channel is gamma encoded */
 997
 998PNG_ALPHA_PNG is the default libpng handling of the alpha channel. It is not
 999pre-multiplied into the color components. In addition the call states
1000that the output is for a sRGB system and causes all PNG files without gAMA
1001chunks to be assumed to be encoded using sRGB.
1002
1003    png_set_alpha_mode(pp, PNG_ALPHA_PNG, PNG_GAMMA_MAC);
1004
1005In this case the output is assumed to be something like an sRGB conformant
1006display preceeded by a power-law lookup table of power 1.45.  This is how
1007early Mac systems behaved.
1008
1009    png_set_alpha_mode(pp, PNG_ALPHA_STANDARD, PNG_GAMMA_LINEAR);
1010
1011This is the classic Jim Blinn approach and will work in academic
1012environments where everything is done by the book.  It has the shortcoming
1013of assuming that input PNG data with no gamma information is linear - this
1014is unlikely to be correct unless the PNG files were generated locally.
1015Most of the time the output precision will be so low as to show
1016significant banding in dark areas of the image.
1017
1018    png_set_expand_16(pp);
1019    png_set_alpha_mode(pp, PNG_ALPHA_STANDARD, PNG_DEFAULT_sRGB);
1020
1021This is a somewhat more realistic Jim Blinn inspired approach.  PNG files
1022are assumed to have the sRGB encoding if not marked with a gamma value and
1023the output is always 16 bits per component.  This permits accurate scaling
1024and processing of the data.  If you know that your input PNG files were
1025generated locally you might need to replace PNG_DEFAULT_sRGB with the
1026correct value for your system.
1027
1028    png_set_alpha_mode(pp, PNG_ALPHA_OPTIMIZED, PNG_DEFAULT_sRGB);
1029
1030If you just need to composite the PNG image onto an existing background
1031and if you control the code that does this you can use the optimization
1032setting.  In this case you just copy completely opaque pixels to the
1033output.  For pixels that are not completely transparent (you just skip
1034those) you do the composition math using png_composite or png_composite_16
1035below then encode the resultant 8-bit or 16-bit values to match the output
1036encoding.
1037
1038    Other cases
1039
1040If neither the PNG nor the standard linear encoding work for you because
1041of the software or hardware you use then you have a big problem.  The PNG
1042case will probably result in halos around the image.  The linear encoding
1043will probably result in a washed out, too bright, image (it's actually too
1044contrasty.)  Try the ALPHA_OPTIMIZED mode above - this will probably
1045substantially reduce the halos.  Alternatively try:
1046
1047    png_set_alpha_mode(pp, PNG_ALPHA_BROKEN, PNG_DEFAULT_sRGB);
1048
1049This option will also reduce the halos, but there will be slight dark
1050halos round the opaque parts of the image where the background is light.
1051In the OPTIMIZED mode the halos will be light halos where the background
1052is dark.  Take your pick - the halos are unavoidable unless you can get
1053your hardware/software fixed!  (The OPTIMIZED approach is slightly
1054faster.)
1055
1056When the default gamma of PNG files doesn't match the output gamma.
1057If you have PNG files with no gamma information png_set_alpha_mode allows
1058you to provide a default gamma, but it also sets the ouput gamma to the
1059matching value.  If you know your PNG files have a gamma that doesn't
1060match the output you can take advantage of the fact that
1061png_set_alpha_mode always sets the output gamma but only sets the PNG
1062default if it is not already set:
1063
1064    png_set_alpha_mode(pp, PNG_ALPHA_PNG, PNG_DEFAULT_sRGB);
1065    png_set_alpha_mode(pp, PNG_ALPHA_PNG, PNG_GAMMA_MAC);
1066
1067The first call sets both the default and the output gamma values, the
1068second call overrides the output gamma without changing the default.  This
1069is easier than achieving the same effect with png_set_gamma.  You must use
1070PNG_ALPHA_PNG for the first call - internal checking in png_set_alpha will
1071fire if more than one call to png_set_alpha_mode and png_set_background is
1072made in the same read operation, however multiple calls with PNG_ALPHA_PNG
1073are ignored.
1074
1075If you don't need, or can't handle, the alpha channel you can call
1076png_set_background() to remove it by compositing against a fixed color.  Don't
1077call png_set_strip_alpha() to do this - it will leave spurious pixel values in
1078transparent parts of this image.
1079
1080   png_set_background(png_ptr, &background_color,
1081       PNG_BACKGROUND_GAMMA_SCREEN, 0, 1);
1082
1083The background_color is an RGB or grayscale value according to the data format
1084libpng will produce for you.  Because you don't yet know the format of the PNG
1085file, if you call png_set_background at this point you must arrange for the
1086format produced by libpng to always have 8-bit or 16-bit components and then
1087store the color as an 8-bit or 16-bit color as appropriate.  The color contains
1088separate gray and RGB component values, so you can let libpng produce gray or
1089RGB output according to the input format, but low bit depth grayscale images
1090must always be converted to at least 8-bit format.  (Even though low bit depth
1091grayscale images can't have an alpha channel they can have a transparent
1092color!)
1093
1094You set the transforms you need later, either as flags to the high level
1095interface or libpng API calls for the low level interface.  For reference the
1096settings and API calls required are:
1097
10988-bit values:
1099   PNG_TRANSFORM_SCALE_16 | PNG_EXPAND
1100   png_set_expand(png_ptr); png_set_scale_16(png_ptr);
1101
1102   If you must get exactly the same inaccurate results
1103   produced by default in versions prior to libpng-1.5.4,
1104   use PNG_TRANSFORM_STRIP_16 and png_set_strip_16(png_ptr)
1105   instead.
1106
110716-bit values:
1108   PNG_TRANSFORM_EXPAND_16
1109   png_set_expand_16(png_ptr);
1110
1111In either case palette image data will be expanded to RGB.  If you just want
1112color data you can add PNG_TRANSFORM_GRAY_TO_RGB or png_set_gray_to_rgb(png_ptr)
1113to the list.
1114
1115Calling png_set_background before the PNG file header is read will not work
1116prior to libpng-1.5.4.  Because the failure may result in unexpected warnings or
1117errors it is therefore much safer to call png_set_background after the head has
1118been read.  Unfortunately this means that prior to libpng-1.5.4 it cannot be
1119used with the high level interface.
1120
1121The high-level read interface
1122
1123At this point there are two ways to proceed; through the high-level
1124read interface, or through a sequence of low-level read operations.
1125You can use the high-level interface if (a) you are willing to read
1126the entire image into memory, and (b) the input transformations
1127you want to do are limited to the following set:
1128
1129    PNG_TRANSFORM_IDENTITY      No transformation
1130    PNG_TRANSFORM_SCALE_16      Strip 16-bit samples to
1131                                8-bit accurately
1132    PNG_TRANSFORM_STRIP_16      Chop 16-bit samples to
1133                                8-bit less accurately
1134    PNG_TRANSFORM_STRIP_ALPHA   Discard the alpha channel
1135    PNG_TRANSFORM_PACKING       Expand 1, 2 and 4-bit
1136                                samples to bytes
1137    PNG_TRANSFORM_PACKSWAP      Change order of packed
1138                                pixels to LSB first
1139    PNG_TRANSFORM_EXPAND        Perform set_expand()
1140    PNG_TRANSFORM_INVERT_MONO   Invert monochrome images
1141    PNG_TRANSFORM_SHIFT         Normalize pixels to the
1142                                sBIT depth
1143    PNG_TRANSFORM_BGR           Flip RGB to BGR, RGBA
1144                                to BGRA
1145    PNG_TRANSFORM_SWAP_ALPHA    Flip RGBA to ARGB or GA
1146                                to AG
1147    PNG_TRANSFORM_INVERT_ALPHA  Change alpha from opacity
1148                                to transparency
1149    PNG_TRANSFORM_SWAP_ENDIAN   Byte-swap 16-bit samples
1150    PNG_TRANSFORM_GRAY_TO_RGB   Expand grayscale samples
1151                                to RGB (or GA to RGBA)
1152    PNG_TRANSFORM_EXPAND_16     Expand samples to 16 bits
1153
1154(This excludes setting a background color, doing gamma transformation,
1155quantizing, and setting filler.)  If this is the case, simply do this:
1156
1157    png_read_png(png_ptr, info_ptr, png_transforms, NULL)
1158
1159where png_transforms is an integer containing the bitwise OR of some
1160set of transformation flags.  This call is equivalent to png_read_info(),
1161followed the set of transformations indicated by the transform mask,
1162then png_read_image(), and finally png_read_end().
1163
1164(The final parameter of this call is not yet used.  Someday it might point
1165to transformation parameters required by some future input transform.)
1166
1167You must use png_transforms and not call any png_set_transform() functions
1168when you use png_read_png().
1169
1170After you have called png_read_png(), you can retrieve the image data
1171with
1172
1173   row_pointers = png_get_rows(png_ptr, info_ptr);
1174
1175where row_pointers is an array of pointers to the pixel data for each row:
1176
1177   png_bytep row_pointers[height];
1178
1179If you know your image size and pixel size ahead of time, you can allocate
1180row_pointers prior to calling png_read_png() with
1181
1182   if (height > PNG_UINT_32_MAX/(sizeof (png_byte)))
1183      png_error (png_ptr,
1184          "Image is too tall to process in memory");
1185
1186   if (width > PNG_UINT_32_MAX/pixel_size)
1187      png_error (png_ptr,
1188          "Image is too wide to process in memory");
1189
1190   row_pointers = png_malloc(png_ptr,
1191       height*(sizeof (png_bytep)));
1192
1193   for (int i=0; i<height, i++)
1194      row_pointers[i]=NULL;  /* security precaution */
1195
1196   for (int i=0; i<height, i++)
1197      row_pointers[i]=png_malloc(png_ptr,
1198          width*pixel_size);
1199
1200   png_set_rows(png_ptr, info_ptr, &row_pointers);
1201
1202Alternatively you could allocate your image in one big block and define
1203row_pointers[i] to point into the proper places in your block, but first
1204be sure that your platform is able to allocate such a large buffer:
1205
1206   /* Guard against integer overflow */
1207   if (height > PNG_SIZE_MAX/(width*pixel_size)) {
1208        png_error(png_ptr,"image_data buffer would be too large");
1209   }
1210
1211   png_bytep buffer=png_malloc(png_ptr,height*width*pixel_size);
1212
1213   for (int i=0; i<height, i++)
1214      row_pointers[i]=buffer+i*width*pixel_size;
1215
1216   png_set_rows(png_ptr, info_ptr, &row_pointers);
1217
1218If you use png_set_rows(), the application is responsible for freeing
1219row_pointers (and row_pointers[i], if they were separately allocated).
1220
1221If you don't allocate row_pointers ahead of time, png_read_png() will
1222do it, and it'll be free'ed by libpng when you call png_destroy_*().
1223
1224The low-level read interface
1225
1226If you are going the low-level route, you are now ready to read all
1227the file information up to the actual image data.  You do this with a
1228call to png_read_info().
1229
1230    png_read_info(png_ptr, info_ptr);
1231
1232This will process all chunks up to but not including the image data.
1233
1234This also copies some of the data from the PNG file into the decode structure
1235for use in later transformations.  Important information copied in is:
1236
12371) The PNG file gamma from the gAMA chunk.  This overwrites the default value
1238provided by an earlier call to png_set_gamma or png_set_alpha_mode.
1239
12402) Prior to libpng-1.5.4 the background color from a bKGd chunk.  This
1241damages the information provided by an earlier call to png_set_background
1242resulting in unexpected behavior.  Libpng-1.5.4 no longer does this.
1243
12443) The number of significant bits in each component value.  Libpng uses this to
1245optimize gamma handling by reducing the internal lookup table sizes.
1246
12474) The transparent color information from a tRNS chunk.  This can be modified by
1248a later call to png_set_tRNS.
1249
1250Querying the info structure
1251
1252Functions are used to get the information from the info_ptr once it
1253has been read.  Note that these fields may not be completely filled
1254in until png_read_end() has read the chunk data following the image.
1255
1256    png_get_IHDR(png_ptr, info_ptr, &width, &height,
1257       &bit_depth, &color_type, &interlace_type,
1258       &compression_type, &filter_method);
1259
1260    width          - holds the width of the image
1261                     in pixels (up to 2^31).
1262
1263    height         - holds the height of the image
1264                     in pixels (up to 2^31).
1265
1266    bit_depth      - holds the bit depth of one of the
1267                     image channels.  (valid values are
1268                     1, 2, 4, 8, 16 and depend also on
1269                     the color_type.  See also
1270                     significant bits (sBIT) below).
1271
1272    color_type     - describes which color/alpha channels
1273                         are present.
1274                     PNG_COLOR_TYPE_GRAY
1275                        (bit depths 1, 2, 4, 8, 16)
1276                     PNG_COLOR_TYPE_GRAY_ALPHA
1277                        (bit depths 8, 16)
1278                     PNG_COLOR_TYPE_PALETTE
1279                        (bit depths 1, 2, 4, 8)
1280                     PNG_COLOR_TYPE_RGB
1281                        (bit_depths 8, 16)
1282                     PNG_COLOR_TYPE_RGB_ALPHA
1283                        (bit_depths 8, 16)
1284
1285                     PNG_COLOR_MASK_PALETTE
1286                     PNG_COLOR_MASK_COLOR
1287                     PNG_COLOR_MASK_ALPHA
1288
1289    interlace_type - (PNG_INTERLACE_NONE or
1290                     PNG_INTERLACE_ADAM7)
1291
1292    compression_type - (must be PNG_COMPRESSION_TYPE_BASE
1293                     for PNG 1.0)
1294
1295    filter_method  - (must be PNG_FILTER_TYPE_BASE
1296                     for PNG 1.0, and can also be
1297                     PNG_INTRAPIXEL_DIFFERENCING if
1298                     the PNG datastream is embedded in
1299                     a MNG-1.0 datastream)
1300
1301    Any of width, height, color_type, bit_depth,
1302    interlace_type, compression_type, or filter_method can
1303    be NULL if you are not interested in their values.
1304
1305    Note that png_get_IHDR() returns 32-bit data into
1306    the application's width and height variables.
1307    This is an unsafe situation if these are not png_uint_32
1308    variables.  In such situations, the
1309    png_get_image_width() and png_get_image_height()
1310    functions described below are safer.
1311
1312    width            = png_get_image_width(png_ptr,
1313                         info_ptr);
1314
1315    height           = png_get_image_height(png_ptr,
1316                         info_ptr);
1317
1318    bit_depth        = png_get_bit_depth(png_ptr,
1319                         info_ptr);
1320
1321    color_type       = png_get_color_type(png_ptr,
1322                         info_ptr);
1323
1324    interlace_type   = png_get_interlace_type(png_ptr,
1325                         info_ptr);
1326
1327    compression_type = png_get_compression_type(png_ptr,
1328                         info_ptr);
1329
1330    filter_method    = png_get_filter_type(png_ptr,
1331                         info_ptr);
1332
1333    channels = png_get_channels(png_ptr, info_ptr);
1334
1335    channels       - number of channels of info for the
1336                     color type (valid values are 1 (GRAY,
1337                     PALETTE), 2 (GRAY_ALPHA), 3 (RGB),
1338                     4 (RGB_ALPHA or RGB + filler byte))
1339
1340    rowbytes = png_get_rowbytes(png_ptr, info_ptr);
1341
1342    rowbytes       - number of bytes needed to hold a row
1343                     This value, the bit_depth, color_type,
1344                     and the number of channels can change
1345                     if you use transforms such as
1346                     png_set_expand(). See
1347                     png_read_update_info(), below.
1348
1349    signature = png_get_signature(png_ptr, info_ptr);
1350
1351    signature      - holds the signature read from the
1352                     file (if any).  The data is kept in
1353                     the same offset it would be if the
1354                     whole signature were read (i.e. if an
1355                     application had already read in 4
1356                     bytes of signature before starting
1357                     libpng, the remaining 4 bytes would
1358                     be in signature[4] through signature[7]
1359                     (see png_set_sig_bytes())).
1360
1361These are also important, but their validity depends on whether the chunk
1362has been read.  The png_get_valid(png_ptr, info_ptr, PNG_INFO_<chunk>) and
1363png_get_<chunk>(png_ptr, info_ptr, ...) functions return non-zero if the
1364data has been read, or zero if it is missing.  The parameters to the
1365png_get_<chunk> are set directly if they are simple data types, or a
1366pointer into the info_ptr is returned for any complex types.
1367
1368The colorspace data from gAMA, cHRM, sRGB, iCCP, and sBIT chunks
1369is simply returned to give the application information about how the
1370image was encoded.  Libpng itself only does transformations using the file
1371gamma when combining semitransparent pixels with the background color, and,
1372since libpng-1.6.0, when converting between 8-bit sRGB and 16-bit linear pixels
1373within the simplified API.  Libpng also uses the file gamma when converting
1374RGB to gray, beginning with libpng-1.0.5, if the application calls
1375png_set_rgb_to_gray()).
1376
1377    png_get_PLTE(png_ptr, info_ptr, &palette,
1378                     &num_palette);
1379
1380    palette        - the palette for the file
1381                     (array of png_color)
1382
1383    num_palette    - number of entries in the palette
1384
1385    png_get_gAMA(png_ptr, info_ptr, &file_gamma);
1386    png_get_gAMA_fixed(png_ptr, info_ptr, &int_file_gamma);
1387
1388    file_gamma     - the gamma at which the file is
1389                     written (PNG_INFO_gAMA)
1390
1391    int_file_gamma - 100,000 times the gamma at which the
1392                     file is written
1393
1394    png_get_cHRM(png_ptr, info_ptr,  &white_x, &white_y, &red_x,
1395                     &red_y, &green_x, &green_y, &blue_x, &blue_y)
1396    png_get_cHRM_XYZ(png_ptr, info_ptr, &red_X, &red_Y, &red_Z,
1397                     &green_X, &green_Y, &green_Z, &blue_X, &blue_Y,
1398                     &blue_Z)
1399    png_get_cHRM_fixed(png_ptr, info_ptr, &int_white_x,
1400                     &int_white_y, &int_red_x, &int_red_y,
1401                     &int_green_x, &int_green_y, &int_blue_x,
1402                     &int_blue_y)
1403    png_get_cHRM_XYZ_fixed(png_ptr, info_ptr, &int_red_X, &int_red_Y,
1404                     &int_red_Z, &int_green_X, &int_green_Y,
1405                     &int_green_Z, &int_blue_X, &int_blue_Y,
1406                     &int_blue_Z)
1407
1408    {white,red,green,blue}_{x,y}
1409                     A color space encoding specified using the
1410                     chromaticities of the end points and the
1411                     white point. (PNG_INFO_cHRM)
1412
1413    {red,green,blue}_{X,Y,Z}
1414                     A color space encoding specified using the
1415                     encoding end points - the CIE tristimulus
1416                     specification of the intended color of the red,
1417                     green and blue channels in the PNG RGB data.
1418                     The white point is simply the sum of the three
1419                     end points. (PNG_INFO_cHRM)
1420
1421    png_get_sRGB(png_ptr, info_ptr, &srgb_intent);
1422
1423    srgb_intent -    the rendering intent (PNG_INFO_sRGB)
1424                     The presence of the sRGB chunk
1425                     means that the pixel data is in the
1426                     sRGB color space.  This chunk also
1427                     implies specific values of gAMA and
1428                     cHRM.
1429
1430    png_get_iCCP(png_ptr, info_ptr, &name,
1431       &compression_type, &profile, &proflen);
1432
1433    name             - The profile name.
1434
1435    compression_type - The compression type; always
1436                       PNG_COMPRESSION_TYPE_BASE for PNG 1.0.
1437                       You may give NULL to this argument to
1438                       ignore it.
1439
1440    profile          - International Color Consortium color
1441                       profile data. May contain NULs.
1442
1443    proflen          - length of profile data in bytes.
1444
1445    png_get_sBIT(png_ptr, info_ptr, &sig_bit);
1446
1447    sig_bit        - the number of significant bits for
1448                     (PNG_INFO_sBIT) each of the gray,
1449                     red, green, and blue channels,
1450                     whichever are appropriate for the
1451                     given color type (png_color_16)
1452
1453    png_get_tRNS(png_ptr, info_ptr, &trans_alpha,
1454                     &num_trans, &trans_color);
1455
1456    trans_alpha    - array of alpha (transparency)
1457                     entries for palette (PNG_INFO_tRNS)
1458
1459    num_trans      - number of transparent entries
1460                     (PNG_INFO_tRNS)
1461
1462    trans_color    - graylevel or color sample values of
1463                     the single transparent color for
1464                     non-paletted images (PNG_INFO_tRNS)
1465
1466    png_get_eXIf_1(png_ptr, info_ptr, &num_exif, &exif);
1467                     (PNG_INFO_eXIf)
1468
1469    exif           - Exif profile (array of png_byte)
1470
1471    png_get_hIST(png_ptr, info_ptr, &hist);
1472                     (PNG_INFO_hIST)
1473
1474    hist           - histogram of palette (array of
1475                     png_uint_16)
1476
1477    png_get_tIME(png_ptr, info_ptr, &mod_time);
1478
1479    mod_time       - time image was last modified
1480                    (PNG_VALID_tIME)
1481
1482    png_get_bKGD(png_ptr, info_ptr, &background);
1483
1484    background     - background color (of type
1485                     png_color_16p) (PNG_VALID_bKGD)
1486                     valid 16-bit red, green and blue
1487                     values, regardless of color_type
1488
1489    num_comments   = png_get_text(png_ptr, info_ptr,
1490                     &text_ptr, &num_text);
1491
1492    num_comments   - number of comments
1493
1494    text_ptr       - array of png_text holding image
1495                     comments
1496
1497    text_ptr[i].compression - type of compression used
1498                 on "text" PNG_TEXT_COMPRESSION_NONE
1499                           PNG_TEXT_COMPRESSION_zTXt
1500                           PNG_ITXT_COMPRESSION_NONE
1501                           PNG_ITXT_COMPRESSION_zTXt
1502
1503    text_ptr[i].key   - keyword for comment.  Must contain
1504                         1-79 characters.
1505
1506    text_ptr[i].text  - text comments for current
1507                         keyword.  Can be empty.
1508
1509    text_ptr[i].text_length - length of text string,
1510                 after decompression, 0 for iTXt
1511
1512    text_ptr[i].itxt_length - length of itxt string,
1513                 after decompression, 0 for tEXt/zTXt
1514
1515    text_ptr[i].lang  - language of comment (empty
1516                         string for unknown).
1517
1518    text_ptr[i].lang_key  - keyword in UTF-8
1519                         (empty string for unknown).
1520
1521    Note that the itxt_length, lang, and lang_key
1522    members of the text_ptr structure only exist when the
1523    library is built with iTXt chunk support.  Prior to
1524    libpng-1.4.0 the library was built by default without
1525    iTXt support. Also note that when iTXt is supported,
1526    they contain NULL pointers when the "compression"
1527    field contains PNG_TEXT_COMPRESSION_NONE or
1528    PNG_TEXT_COMPRESSION_zTXt.
1529
1530    num_text       - number of comments (same as
1531                     num_comments; you can put NULL here
1532                     to avoid the duplication)
1533
1534    Note while png_set_text() will accept text, language,
1535    and translated keywords that can be NULL pointers, the
1536    structure returned by png_get_text will always contain
1537    regular zero-terminated C strings.  They might be
1538    empty strings but they will never be NULL pointers.
1539
1540    num_spalettes = png_get_sPLT(png_ptr, info_ptr,
1541       &palette_ptr);
1542
1543    num_spalettes  - number of sPLT chunks read.
1544
1545    palette_ptr    - array of palette structures holding
1546                     contents of one or more sPLT chunks
1547                     read.
1548
1549    png_get_oFFs(png_ptr, info_ptr, &offset_x, &offset_y,
1550       &unit_type);
1551
1552    offset_x       - positive offset from the left edge
1553                     of the screen (can be negative)
1554
1555    offset_y       - positive offset from the top edge
1556                     of the screen (can be negative)
1557
1558    unit_type      - PNG_OFFSET_PIXEL, PNG_OFFSET_MICROMETER
1559
1560    png_get_pHYs(png_ptr, info_ptr, &res_x, &res_y,
1561       &unit_type);
1562
1563    res_x          - pixels/unit physical resolution in
1564                     x direction
1565
1566    res_y          - pixels/unit physical resolution in
1567                     x direction
1568
1569    unit_type      - PNG_RESOLUTION_UNKNOWN,
1570                     PNG_RESOLUTION_METER
1571
1572    png_get_sCAL(png_ptr, info_ptr, &unit, &width,
1573       &height)
1574
1575    unit        - physical scale units (an integer)
1576
1577    width       - width of a pixel in physical scale units
1578
1579    height      - height of a pixel in physical scale units
1580                 (width and height are doubles)
1581
1582    png_get_sCAL_s(png_ptr, info_ptr, &unit, &width,
1583       &height)
1584
1585    unit        - physical scale units (an integer)
1586
1587    width       - width of a pixel in physical scale units
1588                  (expressed as a string)
1589
1590    height      - height of a pixel in physical scale units
1591                 (width and height are strings like "2.54")
1592
1593    num_unknown_chunks = png_get_unknown_chunks(png_ptr,
1594       info_ptr, &unknowns)
1595
1596    unknowns          - array of png_unknown_chunk
1597                        structures holding unknown chunks
1598
1599    unknowns[i].name  - name of unknown chunk
1600
1601    unknowns[i].data  - data of unknown chunk
1602
1603    unknowns[i].size  - size of unknown chunk's data
1604
1605    unknowns[i].location - position of chunk in file
1606
1607    The value of "i" corresponds to the order in which the
1608    chunks were read from the PNG file or inserted with the
1609    png_set_unknown_chunks() function.
1610
1611    The value of "location" is a bitwise "or" of
1612
1613         PNG_HAVE_IHDR  (0x01)
1614         PNG_HAVE_PLTE  (0x02)
1615         PNG_AFTER_IDAT (0x08)
1616
1617The data from the pHYs chunk can be retrieved in several convenient
1618forms:
1619
1620    res_x = png_get_x_pixels_per_meter(png_ptr,
1621       info_ptr)
1622
1623    res_y = png_get_y_pixels_per_meter(png_ptr,
1624       info_ptr)
1625
1626    res_x_and_y = png_get_pixels_per_meter(png_ptr,
1627       info_ptr)
1628
1629    res_x = png_get_x_pixels_per_inch(png_ptr,
1630       info_ptr)
1631
1632    res_y = png_get_y_pixels_per_inch(png_ptr,
1633       info_ptr)
1634
1635    res_x_and_y = png_get_pixels_per_inch(png_ptr,
1636       info_ptr)
1637
1638    aspect_ratio = png_get_pixel_aspect_ratio(png_ptr,
1639       info_ptr)
1640
1641    Each of these returns 0 [signifying "unknown"] if
1642       the data is not present or if res_x is 0;
1643       res_x_and_y is 0 if res_x != res_y
1644
1645    Note that because of the way the resolutions are
1646       stored internally, the inch conversions won't
1647       come out to exactly even number.  For example,
1648       72 dpi is stored as 0.28346 pixels/meter, and
1649       when this is retrieved it is 71.9988 dpi, so
1650       be sure to round the returned value appropriately
1651       if you want to display a reasonable-looking result.
1652
1653The data from the oFFs chunk can be retrieved in several convenient
1654forms:
1655
1656    x_offset = png_get_x_offset_microns(png_ptr, info_ptr);
1657
1658    y_offset = png_get_y_offset_microns(png_ptr, info_ptr);
1659
1660    x_offset = png_get_x_offset_inches(png_ptr, info_ptr);
1661
1662    y_offset = png_get_y_offset_inches(png_ptr, info_ptr);
1663
1664    Each of these returns 0 [signifying "unknown" if both
1665       x and y are 0] if the data is not present or if the
1666       chunk is present but the unit is the pixel.  The
1667       remark about inexact inch conversions applies here
1668       as well, because a value in inches can't always be
1669       converted to microns and back without some loss
1670       of precision.
1671
1672For more information, see the
1673PNG specification for chunk contents.  Be careful with trusting
1674rowbytes, as some of the transformations could increase the space
1675needed to hold a row (expand, filler, gray_to_rgb, etc.).
1676See png_read_update_info(), below.
1677
1678A quick word about text_ptr and num_text.  PNG stores comments in
1679keyword/text pairs, one pair per chunk, with no limit on the number
1680of text chunks, and a 2^31 byte limit on their size.  While there are
1681suggested keywords, there is no requirement to restrict the use to these
1682strings.  It is strongly suggested that keywords and text be sensible
1683to humans (that's the point), so don't use abbreviations.  Non-printing
1684symbols are not allowed.  See the PNG specification for more details.
1685There is also no requirement to have text after the keyword.
1686
1687Keywords should be limited to 79 Latin-1 characters without leading or
1688trailing spaces, but non-consecutive spaces are allowed within the
1689keyword.  It is possible to have the same keyword any number of times.
1690The text_ptr is an array of png_text structures, each holding a
1691pointer to a language string, a pointer to a keyword and a pointer to
1692a text string.  The text string, language code, and translated
1693keyword may be empty or NULL pointers.  The keyword/text
1694pairs are put into the array in the order that they are received.
1695However, some or all of the text chunks may be after the image, so, to
1696make sure you have read all the text chunks, don't mess with these
1697until after you read the stuff after the image.  This will be
1698mentioned again below in the discussion that goes with png_read_end().
1699
1700Input transformations
1701
1702After you've read the header information, you can set up the library
1703to handle any special transformations of the image data.  The various
1704ways to transform the data will be described in the order that they
1705should occur.  This is important, as some of these change the color
1706type and/or bit depth of the data, and some others only work on
1707certain color types and bit depths.
1708
1709Transformations you request are ignored if they don't have any meaning for a
1710particular input data format.  However some transformations can have an effect
1711as a result of a previous transformation.  If you specify a contradictory set of
1712transformations, for example both adding and removing the alpha channel, you
1713cannot predict the final result.
1714
1715The color used for the transparency values should be supplied in the same
1716format/depth as the current image data.  It is stored in the same format/depth
1717as the image data in a tRNS chunk, so this is what libpng expects for this data.
1718
1719The color used for the background value depends on the need_expand argument as
1720described below.
1721
1722Data will be decoded into the supplied row buffers packed into bytes
1723unless the library has been told to transform it into another format.
1724For example, 4 bit/pixel paletted or grayscale data will be returned
17252 pixels/byte with the leftmost pixel in the high-order bits of the byte,
1726unless png_set_packing() is called.  8-bit RGB data will be stored
1727in RGB RGB RGB format unless png_set_filler() or png_set_add_alpha()
1728is called to insert filler bytes, either before or after each RGB triplet.
1729
173016-bit RGB data will be returned RRGGBB RRGGBB, with the most significant
1731byte of the color value first, unless png_set_scale_16() is called to
1732transform it to regular RGB RGB triplets, or png_set_filler() or
1733png_set_add alpha() is called to insert two filler bytes, either before
1734or after each RRGGBB triplet.  Similarly, 8-bit or 16-bit grayscale data can
1735be modified with png_set_filler(), png_set_add_alpha(), png_set_strip_16(),
1736or png_set_scale_16().
1737
1738The following code transforms grayscale images of less than 8 to 8 bits,
1739changes paletted images to RGB, and adds a full alpha channel if there is
1740transparency information in a tRNS chunk.  This is most useful on
1741grayscale images with bit depths of 2 or 4 or if there is a multiple-image
1742viewing application that wishes to treat all images in the same way.
1743
1744    if (color_type == PNG_COLOR_TYPE_PALETTE)
1745        png_set_palette_to_rgb(png_ptr);
1746
1747    if (png_get_valid(png_ptr, info_ptr,
1748        PNG_INFO_tRNS)) png_set_tRNS_to_alpha(png_ptr);
1749
1750    if (color_type == PNG_COLOR_TYPE_GRAY &&
1751        bit_depth < 8) png_set_expand_gray_1_2_4_to_8(png_ptr);
1752
1753The first two functions are actually aliases for png_set_expand(), added
1754in libpng version 1.0.4, with the function names expanded to improve code
1755readability.  In some future version they may actually do different
1756things.
1757
1758As of libpng version 1.2.9, png_set_expand_gray_1_2_4_to_8() was
1759added.  It expands the sample depth without changing tRNS to alpha.
1760
1761As of libpng version 1.5.2, png_set_expand_16() was added.  It behaves as
1762png_set_expand(); however, the resultant channels have 16 bits rather than 8.
1763Use this when the output color or gray channels are made linear to avoid fairly
1764severe accuracy loss.
1765
1766   if (bit_depth < 16)
1767      png_set_expand_16(png_ptr);
1768
1769PNG can have files with 16 bits per channel.  If you only can handle
17708 bits per channel, this will strip the pixels down to 8-bit.
1771
1772    if (bit_depth == 16)
1773#if PNG_LIBPNG_VER >= 10504
1774       png_set_scale_16(png_ptr);
1775#else
1776       png_set_strip_16(png_ptr);
1777#endif
1778
1779(The more accurate "png_set_scale_16()" API became available in libpng version
17801.5.4).
1781
1782If you need to process the alpha channel on the image separately from the image
1783data (for example if you convert it to a bitmap mask) it is possible to have
1784libpng strip the channel leaving just RGB or gray data:
1785
1786    if (color_type & PNG_COLOR_MASK_ALPHA)
1787       png_set_strip_alpha(png_ptr);
1788
1789If you strip the alpha channel you need to find some other way of dealing with
1790the information.  If, instead, you want to convert the image to an opaque
1791version with no alpha channel use png_set_background; see below.
1792
1793As of libpng version 1.5.2, almost all useful expansions are supported, the
1794major ommissions are conversion of grayscale to indexed images (which can be
1795done trivially in the application) and conversion of indexed to grayscale (which
1796can be done by a trivial manipulation of the palette.)
1797
1798In the following table, the 01 means grayscale with depth<8, 31 means
1799indexed with depth<8, other numerals represent the color type, "T" means
1800the tRNS chunk is present, A means an alpha channel is present, and O
1801means tRNS or alpha is present but all pixels in the image are opaque.
1802
1803  FROM  01  31   0  0T  0O   2  2T  2O   3  3T  3O  4A  4O  6A  6O
1804   TO
1805   01    -  [G]  -   -   -   -   -   -   -   -   -   -   -   -   -
1806   31   [Q]  Q  [Q] [Q] [Q]  Q   Q   Q   Q   Q   Q  [Q] [Q]  Q   Q
1807    0    1   G   +   .   .   G   G   G   G   G   G   B   B  GB  GB
1808   0T    lt  Gt  t   +   .   Gt  G   G   Gt  G   G   Bt  Bt GBt GBt
1809   0O    lt  Gt  t   .   +   Gt  Gt  G   Gt  Gt  G   Bt  Bt GBt GBt
1810    2    C   P   C   C   C   +   .   .   C   -   -  CB  CB   B   B
1811   2T    Ct  -   Ct  C   C   t   +   t   -   -   -  CBt CBt  Bt  Bt
1812   2O    Ct  -   Ct  C   C   t   t   +   -   -   -  CBt CBt  Bt  Bt
1813    3   [Q]  p  [Q] [Q] [Q]  Q   Q   Q   +   .   .  [Q] [Q]  Q   Q
1814   3T   [Qt] p  [Qt][Q] [Q]  Qt  Qt  Qt  t   +   t  [Qt][Qt] Qt  Qt
1815   3O   [Qt] p  [Qt][Q] [Q]  Qt  Qt  Qt  t   t   +  [Qt][Qt] Qt  Qt
1816   4A    lA  G   A   T   T   GA  GT  GT  GA  GT  GT  +   BA  G  GBA
1817   4O    lA GBA  A   T   T   GA  GT  GT  GA  GT  GT  BA  +  GBA  G
1818   6A    CA  PA  CA  C   C   A   T  tT   PA  P   P   C  CBA  +   BA
1819   6O    CA PBA  CA  C   C   A  tT   T   PA  P   P  CBA  C   BA  +
1820
1821Within the matrix,
1822     "+" identifies entries where 'from' and 'to' are the same.
1823     "-" means the transformation is not supported.
1824     "." means nothing is necessary (a tRNS chunk can just be ignored).
1825     "t" means the transformation is obtained by png_set_tRNS.
1826     "A" means the transformation is obtained by png_set_add_alpha().
1827     "X" means the transformation is obtained by png_set_expand().
1828     "1" means the transformation is obtained by
1829         png_set_expand_gray_1_2_4_to_8() (and by png_set_expand()
1830         if there is no transparency in the original or the final
1831         format).
1832     "C" means the transformation is obtained by png_set_gray_to_rgb().
1833     "G" means the transformation is obtained by png_set_rgb_to_gray().
1834     "P" means the transformation is obtained by
1835         png_set_expand_palette_to_rgb().
1836     "p" means the transformation is obtained by png_set_packing().
1837     "Q" means the transformation is obtained by png_set_quantize().
1838     "T" means the transformation is obtained by
1839         png_set_tRNS_to_alpha().
1840     "B" means the transformation is obtained by
1841         png_set_background(), or png_strip_alpha().
1842
1843When an entry has multiple transforms listed all are required to cause the
1844right overall transformation.  When two transforms are separated by a comma
1845either will do the job.  When transforms are enclosed in [] the transform should
1846do the job but this is currently unimplemented - a different format will result
1847if the suggested transformations are used.
1848
1849In PNG files, the alpha channel in an image
1850is the level of opacity.  If you need the alpha channel in an image to
1851be the level of transparency instead of opacity, you can invert the
1852alpha channel (or the tRNS chunk data) after it's read, so that 0 is
1853fully opaque and 255 (in 8-bit or paletted images) or 65535 (in 16-bit
1854images) is fully transparent, with
1855
1856    png_set_invert_alpha(png_ptr);
1857
1858PNG files pack pixels of bit depths 1, 2, and 4 into bytes as small as
1859they can, resulting in, for example, 8 pixels per byte for 1 bit
1860files.  This code expands to 1 pixel per byte without changing the
1861values of the pixels:
1862
1863    if (bit_depth < 8)
1864       png_set_packing(png_ptr);
1865
1866PNG files have possible bit depths of 1, 2, 4, 8, and 16.  All pixels
1867stored in a PNG image have been "scaled" or "shifted" up to the next
1868higher possible bit depth (e.g. from 5 bits/sample in the range [0,31]
1869to 8 bits/sample in the range [0, 255]).  However, it is also possible
1870to convert the PNG pixel data back to the original bit depth of the
1871image.  This call reduces the pixels back down to the original bit depth:
1872
1873    png_color_8p sig_bit;
1874
1875    if (png_get_sBIT(png_ptr, info_ptr, &sig_bit))
1876       png_set_shift(png_ptr, sig_bit);
1877
1878PNG files store 3-color pixels in red, green, blue order.  This code
1879changes the storage of the pixels to blue, green, red:
1880
1881    if (color_type == PNG_COLOR_TYPE_RGB ||
1882        color_type == PNG_COLOR_TYPE_RGB_ALPHA)
1883       png_set_bgr(png_ptr);
1884
1885PNG files store RGB pixels packed into 3 or 6 bytes. This code expands them
1886into 4 or 8 bytes for windowing systems that need them in this format:
1887
1888    if (color_type == PNG_COLOR_TYPE_RGB)
1889       png_set_filler(png_ptr, filler, PNG_FILLER_BEFORE);
1890
1891where "filler" is the 8-bit or 16-bit number to fill with, and the location
1892is either PNG_FILLER_BEFORE or PNG_FILLER_AFTER, depending upon whether
1893you want the filler before the RGB or after. When filling an 8-bit pixel,
1894the least significant 8 bits of the number are used, if a 16-bit number is
1895supplied.  This transformation does not affect images that already have full
1896alpha channels.  To add an opaque alpha channel, use filler=0xffff and
1897PNG_FILLER_AFTER which will generate RGBA pixels.
1898
1899Note that png_set_filler() does not change the color type.  If you want
1900to do that, you can add a true alpha channel with
1901
1902    if (color_type == PNG_COLOR_TYPE_RGB ||
1903       color_type == PNG_COLOR_TYPE_GRAY)
1904       png_set_add_alpha(png_ptr, filler, PNG_FILLER_AFTER);
1905
1906where "filler" contains the alpha value to assign to each pixel.
1907The png_set_add_alpha() function was added in libpng-1.2.7.
1908
1909If you are reading an image with an alpha channel, and you need the
1910data as ARGB instead of the normal PNG format RGBA:
1911
1912    if (color_type == PNG_COLOR_TYPE_RGB_ALPHA)
1913       png_set_swap_alpha(png_ptr);
1914
1915For some uses, you may want a grayscale image to be represented as
1916RGB.  This code will do that conversion:
1917
1918    if (color_type == PNG_COLOR_TYPE_GRAY ||
1919        color_type == PNG_COLOR_TYPE_GRAY_ALPHA)
1920       png_set_gray_to_rgb(png_ptr);
1921
1922Conversely, you can convert an RGB or RGBA image to grayscale or grayscale
1923with alpha.
1924
1925    if (color_type == PNG_COLOR_TYPE_RGB ||
1926        color_type == PNG_COLOR_TYPE_RGB_ALPHA)
1927       png_set_rgb_to_gray(png_ptr, error_action,
1928          double red_weight, double green_weight);
1929
1930    error_action = 1: silently do the conversion
1931
1932    error_action = 2: issue a warning if the original
1933                      image has any pixel where
1934                      red != green or red != blue
1935
1936    error_action = 3: issue an error and abort the
1937                      conversion if the original
1938                      image has any pixel where
1939                      red != green or red != blue
1940
1941    red_weight:       weight of red component
1942
1943    green_weight:     weight of green component
1944                      If either weight is negative, default
1945                      weights are used.
1946
1947In the corresponding fixed point API the red_weight and green_weight values are
1948simply scaled by 100,000:
1949
1950    png_set_rgb_to_gray(png_ptr, error_action,
1951       png_fixed_point red_weight,
1952       png_fixed_point green_weight);
1953
1954If you have set error_action = 1 or 2, you can
1955later check whether the image really was gray, after processing
1956the image rows, with the png_get_rgb_to_gray_status(png_ptr) function.
1957It will return a png_byte that is zero if the image was gray or
19581 if there were any non-gray pixels.  Background and sBIT data
1959will be silently converted to grayscale, using the green channel
1960data for sBIT, regardless of the error_action setting.
1961
1962The default values come from the PNG file cHRM chunk if present; otherwise, the
1963defaults correspond to the ITU-R recommendation 709, and also the sRGB color
1964space, as recommended in the Charles Poynton's Colour FAQ,
1965Copyright (c) 2006-11-28 Charles Poynton, in section 9:
1966
1967<http://www.poynton.com/notes/colour_and_gamma/ColorFAQ.html#RTFToC9>
1968
1969    Y = 0.2126 * R + 0.7152 * G + 0.0722 * B
1970
1971Previous versions of this document, 1998 through 2002, recommended a slightly
1972different formula:
1973
1974    Y = 0.212671 * R + 0.715160 * G + 0.072169 * B
1975
1976Libpng uses an integer approximation:
1977
1978    Y = (6968 * R + 23434 * G + 2366 * B)/32768
1979
1980The calculation is done in a linear colorspace, if the image gamma
1981can be determined.
1982
1983The png_set_background() function has been described already; it tells libpng to
1984composite images with alpha or simple transparency against the supplied
1985background color.  For compatibility with versions of libpng earlier than
1986libpng-1.5.4 it is recommended that you call the function after reading the file
1987header, even if you don't want to use the color in a bKGD chunk, if one exists.
1988
1989If the PNG file contains a bKGD chunk (PNG_INFO_bKGD valid),
1990you may use this color, or supply another color more suitable for
1991the current display (e.g., the background color from a web page).  You
1992need to tell libpng how the color is represented, both the format of the
1993component values in the color (the number of bits) and the gamma encoding of the
1994color.  The function takes two arguments, background_gamma_mode and need_expand
1995to convey this information; however, only two combinations are likely to be
1996useful:
1997
1998    png_color_16 my_background;
1999    png_color_16p image_background;
2000
2001    if (png_get_bKGD(png_ptr, info_ptr, &image_background))
2002       png_set_background(png_ptr, image_background,
2003           PNG_BACKGROUND_GAMMA_FILE, 1/*needs to be expanded*/, 1);
2004    else
2005       png_set_background(png_ptr, &my_background,
2006           PNG_BACKGROUND_GAMMA_SCREEN, 0/*do not expand*/, 1);
2007
2008The second call was described above - my_background is in the format of the
2009final, display, output produced by libpng.  Because you now know the format of
2010the PNG it is possible to avoid the need to choose either 8-bit or 16-bit
2011output and to retain palette images (the palette colors will be modified
2012appropriately and the tRNS chunk removed.)  However, if you are doing this,
2013take great care not to ask for transformations without checking first that
2014they apply!
2015
2016In the first call the background color has the original bit depth and color type
2017of the PNG file.  So, for palette images the color is supplied as a palette
2018index and for low bit greyscale images the color is a reduced bit value in
2019image_background->gray.
2020
2021If you didn't call png_set_gamma() before reading the file header, for example
2022if you need your code to remain compatible with older versions of libpng prior
2023to libpng-1.5.4, this is the place to call it.
2024
2025Do not call it if you called png_set_alpha_mode(); doing so will damage the
2026settings put in place by png_set_alpha_mode().  (If png_set_alpha_mode() is
2027supported then you can certainly do png_set_gamma() before reading the PNG
2028header.)
2029
2030This API unconditionally sets the screen and file gamma values, so it will
2031override the value in the PNG file unless it is called before the PNG file
2032reading starts.  For this reason you must always call it with the PNG file
2033value when you call it in this position:
2034
2035   if (png_get_gAMA(png_ptr, info_ptr, &file_gamma))
2036      png_set_gamma(png_ptr, screen_gamma, file_gamma);
2037
2038   else
2039      png_set_gamma(png_ptr, screen_gamma, 0.45455);
2040
2041If you need to reduce an RGB file to a paletted file, or if a paletted
2042file has more entries than will fit on your screen, png_set_quantize()
2043will do that.  Note that this is a simple match quantization that merely
2044finds the closest color available.  This should work fairly well with
2045optimized palettes, but fairly badly with linear color cubes.  If you
2046pass a palette that is larger than maximum_colors, the file will
2047reduce the number of colors in the palette so it will fit into
2048maximum_colors.  If there is a histogram, libpng will use it to make
2049more intelligent choices when reducing the palette.  If there is no
2050histogram, it may not do as good a job.
2051
2052   if (color_type & PNG_COLOR_MASK_COLOR)
2053   {
2054      if (png_get_valid(png_ptr, info_ptr,
2055          PNG_INFO_PLTE))
2056      {
2057         png_uint_16p histogram = NULL;
2058
2059         png_get_hIST(png_ptr, info_ptr,
2060             &histogram);
2061         png_set_quantize(png_ptr, palette, num_palette,
2062            max_screen_colors, histogram, 1);
2063      }
2064
2065      else
2066      {
2067         png_color std_color_cube[MAX_SCREEN_COLORS] =
2068            { ... colors ... };
2069
2070         png_set_quantize(png_ptr, std_color_cube,
2071            MAX_SCREEN_COLORS, MAX_SCREEN_COLORS,
2072            NULL,0);
2073      }
2074   }
2075
2076PNG files describe monochrome as black being zero and white being one.
2077The following code will reverse this (make black be one and white be
2078zero):
2079
2080   if (bit_depth == 1 && color_type == PNG_COLOR_TYPE_GRAY)
2081      png_set_invert_mono(png_ptr);
2082
2083This function can also be used to invert grayscale and gray-alpha images:
2084
2085   if (color_type == PNG_COLOR_TYPE_GRAY ||
2086       color_type == PNG_COLOR_TYPE_GRAY_ALPHA)
2087      png_set_invert_mono(png_ptr);
2088
2089PNG files store 16-bit pixels in network byte order (big-endian,
2090ie. most significant bits first).  This code changes the storage to the
2091other way (little-endian, i.e. least significant bits first, the
2092way PCs store them):
2093
2094    if (bit_depth == 16)
2095       png_set_swap(png_ptr);
2096
2097If you are using packed-pixel images (1, 2, or 4 bits/pixel), and you
2098need to change the order the pixels are packed into bytes, you can use:
2099
2100    if (bit_depth < 8)
2101       png_set_packswap(png_ptr);
2102
2103Finally, you can write your own transformation function if none of
2104the existing ones meets your needs.  This is done by setting a callback
2105with
2106
2107    png_set_read_user_transform_fn(png_ptr,
2108        read_transform_fn);
2109
2110You must supply the function
2111
2112    void read_transform_fn(png_structp png_ptr, png_row_infop
2113        row_info, png_bytep data)
2114
2115See pngtest.c for a working example.  Your function will be called
2116after all of the other transformations have been processed.  Take care with
2117interlaced images if you do the interlace yourself - the width of the row is the
2118width in 'row_info', not the overall image width.
2119
2120If supported, libpng provides two information routines that you can use to find
2121where you are in processing the image:
2122
2123   png_get_current_pass_number(png_structp png_ptr);
2124   png_get_current_row_number(png_structp png_ptr);
2125
2126Don't try using these outside a transform callback - firstly they are only
2127supported if user transforms are supported, secondly they may well return
2128unexpected results unless the row is actually being processed at the moment they
2129are called.
2130
2131With interlaced
2132images the value returned is the row in the input sub-image image.  Use
2133PNG_ROW_FROM_PASS_ROW(row, pass) and PNG_COL_FROM_PASS_COL(col, pass) to
2134find the output pixel (x,y) given an interlaced sub-image pixel (row,col,pass).
2135
2136The discussion of interlace handling above contains more information on how to
2137use these values.
2138
2139You can also set up a pointer to a user structure for use by your
2140callback function, and you can inform libpng that your transform
2141function will change the number of channels or bit depth with the
2142function
2143
2144    png_set_user_transform_info(png_ptr, user_ptr,
2145        user_depth, user_channels);
2146
2147The user's application, not libpng, is responsible for allocating and
2148freeing any memory required for the user structure.
2149
2150You can retrieve the pointer via the function
2151png_get_user_transform_ptr().  For example:
2152
2153    voidp read_user_transform_ptr =
2154        png_get_user_transform_ptr(png_ptr);
2155
2156The last thing to handle is interlacing; this is covered in detail below,
2157but you must call the function here if you want libpng to handle expansion
2158of the interlaced image.
2159
2160    number_of_passes = png_set_interlace_handling(png_ptr);
2161
2162After setting the transformations, libpng can update your png_info
2163structure to reflect any transformations you've requested with this
2164call.
2165
2166    png_read_update_info(png_ptr, info_ptr);
2167
2168This is most useful to update the info structure's rowbytes
2169field so you can use it to allocate your image memory.  This function
2170will also update your palette with the correct screen_gamma and
2171background if these have been given with the calls above.  You may
2172only call png_read_update_info() once with a particular info_ptr.
2173
2174After you call png_read_update_info(), you can allocate any
2175memory you need to hold the image.  The row data is simply
2176raw byte data for all forms of images.  As the actual allocation
2177varies among applications, no example will be given.  If you
2178are allocating one large chunk, you will need to build an
2179array of pointers to each row, as it will be needed for some
2180of the functions below.
2181
2182Be sure that your platform can allocate the buffer that you'll need.
2183libpng internally checks for oversize width, but you'll need to
2184do your own check for number_of_rows*width*pixel_size if you are using
2185a multiple-row buffer:
2186
2187   /* Guard against integer overflow */
2188   if (number_of_rows > PNG_SIZE_MAX/(width*pixel_size)) {
2189        png_error(png_ptr,"image_data buffer would be too large");
2190   }
2191
2192Remember: Before you call png_read_update_info(), the png_get_*()
2193functions return the values corresponding to the original PNG image.
2194After you call png_read_update_info the values refer to the image
2195that libpng will output.  Consequently you must call all the png_set_
2196functions before you call png_read_update_info().  This is particularly
2197important for png_set_interlace_handling() - if you are going to call
2198png_read_update_info() you must call png_set_interlace_handling() before
2199it unless you want to receive interlaced output.
2200
2201Reading image data
2202
2203After you've allocated memory, you can read the image data.
2204The simplest way to do this is in one function call.  If you are
2205allocating enough memory to hold the whole image, you can just
2206call png_read_image() and libpng will read in all the image data
2207and put it in the memory area supplied.  You will need to pass in
2208an array of pointers to each row.
2209
2210This function automatically handles interlacing, so you don't
2211need to call png_set_interlace_handling() (unless you call
2212png_read_update_info()) or call this function multiple times, or any
2213of that other stuff necessary with png_read_rows().
2214
2215   png_read_image(png_ptr, row_pointers);
2216
2217where row_pointers is:
2218
2219   png_bytep row_pointers[height];
2220
2221You can point to void or char or whatever you use for pixels.
2222
2223If you don't want to read in the whole image at once, you can
2224use png_read_rows() instead.  If there is no interlacing (check
2225interlace_type == PNG_INTERLACE_NONE), this is simple:
2226
2227    png_read_rows(png_ptr, row_pointers, NULL,
2228        number_of_rows);
2229
2230where row_pointers is the same as in the png_read_image() call.
2231
2232If you are doing this just one row at a time, you can do this with
2233a single row_pointer instead of an array of row_pointers:
2234
2235    png_bytep row_pointer = row;
2236    png_read_row(png_ptr, row_pointer, NULL);
2237
2238If the file is interlaced (interlace_type != 0 in the IHDR chunk), things
2239get somewhat harder.  The only current (PNG Specification version 1.2)
2240interlacing type for PNG is (interlace_type == PNG_INTERLACE_ADAM7);
2241a somewhat complicated 2D interlace scheme, known as Adam7, that
2242breaks down an image into seven smaller images of varying size, based
2243on an 8x8 grid.  This number is defined (from libpng 1.5) as
2244PNG_INTERLACE_ADAM7_PASSES in png.h
2245
2246libpng can fill out those images or it can give them to you "as is".
2247It is almost always better to have libpng handle the interlacing for you.
2248If you want the images filled out, there are two ways to do that.  The one
2249mentioned in the PNG specification is to expand each pixel to cover
2250those pixels that have not been read yet (the "rectangle" method).
2251This results in a blocky image for the first pass, which gradually
2252smooths out as more pixels are read.  The other method is the "sparkle"
2253method, where pixels are drawn only in their final locations, with the
2254rest of the image remaining whatever colors they were initialized to
2255before the start of the read.  The first method usually looks better,
2256but tends to be slower, as there are more pixels to put in the rows.
2257
2258If, as is likely, you want libpng to expand the images, call this before
2259calling png_start_read_image() or png_read_update_info():
2260
2261    if (interlace_type == PNG_INTERLACE_ADAM7)
2262       number_of_passes
2263           = png_set_interlace_handling(png_ptr);
2264
2265This will return the number of passes needed.  Currently, this is seven,
2266but may change if another interlace type is added.  This function can be
2267called even if the file is not interlaced, where it will return one pass.
2268You then need to read the whole image 'number_of_passes' times.  Each time
2269will distribute the pixels from the current pass to the correct place in
2270the output image, so you need to supply the same rows to png_read_rows in
2271each pass.
2272
2273If you are not going to display the image after each pass, but are
2274going to wait until the entire image is read in, use the sparkle
2275effect.  This effect is faster and the end result of either method
2276is exactly the same.  If you are planning on displaying the image
2277after each pass, the "rectangle" effect is generally considered the
2278better looking one.
2279
2280If you only want the "sparkle" effect, just call png_read_row() or
2281png_read_rows() as
2282normal, with the third parameter NULL.  Make sure you make pass over
2283the image number_of_passes times, and you don't change the data in the
2284rows between calls.  You can change the locations of the data, just
2285not the data.  Each pass only writes the pixels appropriate for that
2286pass, and assumes the data from previous passes is still valid.
2287
2288    png_read_rows(png_ptr, row_pointers, NULL,
2289        number_of_rows);
2290    or
2291    png_read_row(png_ptr, row_pointers, NULL);
2292
2293If you only want the first effect (the rectangles), do the same as
2294before except pass the row buffer in the third parameter, and leave
2295the second parameter NULL.
2296
2297    png_read_rows(png_ptr, NULL, row_pointers,
2298        number_of_rows);
2299    or
2300    png_read_row(png_ptr, NULL, row_pointers);
2301
2302If you don't want libpng to handle the interlacing details, just call
2303png_read_rows() PNG_INTERLACE_ADAM7_PASSES times to read in all the images.
2304Each of the images is a valid image by itself; however, you will almost
2305certainly need to distribute the pixels from each sub-image to the
2306correct place.  This is where everything gets very tricky.
2307
2308If you want to retrieve the separate images you must pass the correct
2309number of rows to each successive call of png_read_rows().  The calculation
2310gets pretty complicated for small images, where some sub-images may
2311not even exist because either their width or height ends up zero.
2312libpng provides two macros to help you in 1.5 and later versions:
2313
2314   png_uint_32 width = PNG_PASS_COLS(image_width, pass_number);
2315   png_uint_32 height = PNG_PASS_ROWS(image_height, pass_number);
2316
2317Respectively these tell you the width and height of the sub-image
2318corresponding to the numbered pass.  'pass' is in in the range 0 to 6 -
2319this can be confusing because the specification refers to the same passes
2320as 1 to 7!  Be careful, you must check both the width and height before
2321calling png_read_rows() and not call it for that pass if either is zero.
2322
2323You can, of course, read each sub-image row by row.  If you want to
2324produce optimal code to make a pixel-by-pixel transformation of an
2325interlaced image this is the best approach; read each row of each pass,
2326transform it, and write it out to a new interlaced image.
2327
2328If you want to de-interlace the image yourself libpng provides further
2329macros to help that tell you where to place the pixels in the output image.
2330Because the interlacing scheme is rectangular - sub-image pixels are always
2331arranged on a rectangular grid - all you need to know for each pass is the
2332starting column and row in the output image of the first pixel plus the
2333spacing between each pixel.  As of libpng 1.5 there are four macros to
2334retrieve this information:
2335
2336   png_uint_32 x = PNG_PASS_START_COL(pass);
2337   png_uint_32 y = PNG_PASS_START_ROW(pass);
2338   png_uint_32 xStep = 1U << PNG_PASS_COL_SHIFT(pass);
2339   png_uint_32 yStep = 1U << PNG_PASS_ROW_SHIFT(pass);
2340
2341These allow you to write the obvious loop:
2342
2343   png_uint_32 input_y = 0;
2344   png_uint_32 output_y = PNG_PASS_START_ROW(pass);
2345
2346   while (output_y < output_image_height)
2347   {
2348      png_uint_32 input_x = 0;
2349      png_uint_32 output_x = PNG_PASS_START_COL(pass);
2350
2351      while (output_x < output_image_width)
2352      {
2353         image[output_y][output_x] =
2354             subimage[pass][input_y][input_x++];
2355
2356         output_x += xStep;
2357      }
2358
2359      ++input_y;
2360      output_y += yStep;
2361   }
2362
2363Notice that the steps between successive output rows and columns are
2364returned as shifts.  This is possible because the pixels in the subimages
2365are always a power of 2 apart - 1, 2, 4 or 8 pixels - in the original
2366image.  In practice you may need to directly calculate the output coordinate
2367given an input coordinate.  libpng provides two further macros for this
2368purpose:
2369
2370   png_uint_32 output_x = PNG_COL_FROM_PASS_COL(input_x, pass);
2371   png_uint_32 output_y = PNG_ROW_FROM_PASS_ROW(input_y, pass);
2372
2373Finally a pair of macros are provided to tell you if a particular image
2374row or column appears in a given pass:
2375
2376   int col_in_pass = PNG_COL_IN_INTERLACE_PASS(output_x, pass);
2377   int row_in_pass = PNG_ROW_IN_INTERLACE_PASS(output_y, pass);
2378
2379Bear in mind that you will probably also need to check the width and height
2380of the pass in addition to the above to be sure the pass even exists!
2381
2382With any luck you are convinced by now that you don't want to do your own
2383interlace handling.  In reality normally the only good reason for doing this
2384is if you are processing PNG files on a pixel-by-pixel basis and don't want
2385to load the whole file into memory when it is interlaced.
2386
2387libpng includes a test program, pngvalid, that illustrates reading and
2388writing of interlaced images.  If you can't get interlacing to work in your
2389code and don't want to leave it to libpng (the recommended approach), see
2390how pngvalid.c does it.
2391
2392Finishing a sequential read
2393
2394After you are finished reading the image through the
2395low-level interface, you can finish reading the file.
2396
2397If you want to use a different crc action for handling CRC errors in
2398chunks after the image data, you can call png_set_crc_action()
2399again at this point.
2400
2401If you are interested in comments or time, which may be stored either
2402before or after the image data, you should pass the separate png_info
2403struct if you want to keep the comments from before and after the image
2404separate.
2405
2406    png_infop end_info = png_create_info_struct(png_ptr);
2407
2408    if (!end_info)
2409    {
2410       png_destroy_read_struct(&png_ptr, &info_ptr,
2411           (png_infopp)NULL);
2412       return (ERROR);
2413    }
2414
2415   png_read_end(png_ptr, end_info);
2416
2417If you are not interested, you should still call png_read_end()
2418but you can pass NULL, avoiding the need to create an end_info structure.
2419If you do this, libpng will not process any chunks after IDAT other than
2420skipping over them and perhaps (depending on whether you have called
2421png_set_crc_action) checking their CRCs while looking for the IEND chunk.
2422
2423   png_read_end(png_ptr, (png_infop)NULL);
2424
2425If you don't call png_read_end(), then your file pointer will be
2426left pointing to the first chunk after the last IDAT, which is probably
2427not what you want if you expect to read something beyond the end of
2428the PNG datastream.
2429
2430When you are done, you can free all memory allocated by libpng like this:
2431
2432   png_destroy_read_struct(&png_ptr, &info_ptr,
2433       &end_info);
2434
2435or, if you didn't create an end_info structure,
2436
2437   png_destroy_read_struct(&png_ptr, &info_ptr,
2438       (png_infopp)NULL);
2439
2440It is also possible to individually free the info_ptr members that
2441point to libpng-allocated storage with the following function:
2442
2443    png_free_data(png_ptr, info_ptr, mask, seq)
2444
2445    mask - identifies data to be freed, a mask
2446           containing the bitwise OR of one or
2447           more of
2448             PNG_FREE_PLTE, PNG_FREE_TRNS,
2449             PNG_FREE_HIST, PNG_FREE_ICCP,
2450             PNG_FREE_PCAL, PNG_FREE_ROWS,
2451             PNG_FREE_SCAL, PNG_FREE_SPLT,
2452             PNG_FREE_TEXT, PNG_FREE_UNKN,
2453           or simply PNG_FREE_ALL
2454
2455    seq  - sequence number of item to be freed
2456           (-1 for all items)
2457
2458This function may be safely called when the relevant storage has
2459already been freed, or has not yet been allocated, or was allocated
2460by the user and not by libpng,  and will in those cases do nothing.
2461The "seq" parameter is ignored if only one item of the selected data
2462type, such as PLTE, is allowed.  If "seq" is not -1, and multiple items
2463are allowed for the data type identified in the mask, such as text or
2464sPLT, only the n'th item in the structure is freed, where n is "seq".
2465
2466The default behavior is only to free data that was allocated internally
2467by libpng.  This can be changed, so that libpng will not free the data,
2468or so that it will free data that was allocated by the user with png_malloc()
2469or png_calloc() and passed in via a png_set_*() function, with
2470
2471    png_data_freer(png_ptr, info_ptr, freer, mask)
2472
2473    freer  - one of
2474               PNG_DESTROY_WILL_FREE_DATA
2475               PNG_SET_WILL_FREE_DATA
2476               PNG_USER_WILL_FREE_DATA
2477
2478    mask   - which data elements are affected
2479             same choices as in png_free_data()
2480
2481This function only affects data that has already been allocated.
2482You can call this function after reading the PNG data but before calling
2483any png_set_*() functions, to control whether the user or the png_set_*()
2484function is responsible for freeing any existing data that might be present,
2485and again after the png_set_*() functions to control whether the user
2486or png_destroy_*() is supposed to free the data.  When the user assumes
2487responsibility for libpng-allocated data, the application must use
2488png_free() to free it, and when the user transfers responsibility to libpng
2489for data that the user has allocated, the user must have used png_malloc()
2490or png_calloc() to allocate it.
2491
2492If you allocated your row_pointers in a single block, as suggested above in
2493the description of the high level read interface, you must not transfer
2494responsibility for freeing it to the png_set_rows or png_read_destroy function,
2495because they would also try to free the individual row_pointers[i].
2496
2497If you allocated text_ptr.text, text_ptr.lang, and text_ptr.translated_keyword
2498separately, do not transfer responsibility for freeing text_ptr to libpng,
2499because when libpng fills a png_text structure it combines these members with
2500the key member, and png_free_data() will free only text_ptr.key.  Similarly,
2501if you transfer responsibility for free'ing text_ptr from libpng to your
2502application, your application must not separately free those members.
2503
2504The png_free_data() function will turn off the "valid" flag for anything
2505it frees.  If you need to turn the flag off for a chunk that was freed by
2506your application instead of by libpng, you can use
2507
2508    png_set_invalid(png_ptr, info_ptr, mask);
2509
2510    mask - identifies the chunks to be made invalid,
2511           containing the bitwise OR of one or
2512           more of
2513             PNG_INFO_gAMA, PNG_INFO_sBIT,
2514             PNG_INFO_cHRM, PNG_INFO_PLTE,
2515             PNG_INFO_tRNS, PNG_INFO_bKGD,
2516             PNG_INFO_eXIf,
2517             PNG_INFO_hIST, PNG_INFO_pHYs,
2518             PNG_INFO_oFFs, PNG_INFO_tIME,
2519             PNG_INFO_pCAL, PNG_INFO_sRGB,
2520             PNG_INFO_iCCP, PNG_INFO_sPLT,
2521             PNG_INFO_sCAL, PNG_INFO_IDAT
2522
2523For a more compact example of reading a PNG image, see the file example.c.
2524
2525Reading PNG files progressively
2526
2527The progressive reader is slightly different from the non-progressive
2528reader.  Instead of calling png_read_info(), png_read_rows(), and
2529png_read_end(), you make one call to png_process_data(), which calls
2530callbacks when it has the info, a row, or the end of the image.  You
2531set up these callbacks with png_set_progressive_read_fn().  You don't
2532have to worry about the input/output functions of libpng, as you are
2533giving the library the data directly in png_process_data().  I will
2534assume that you have read the section on reading PNG files above,
2535so I will only highlight the differences (although I will show
2536all of the code).
2537
2538png_structp png_ptr;
2539png_infop info_ptr;
2540
2541 /*  An example code fragment of how you would
2542     initialize the progressive reader in your
2543     application. */
2544 int
2545 initialize_png_reader()
2546 {
2547    png_ptr = png_create_read_struct
2548        (PNG_LIBPNG_VER_STRING, (png_voidp)user_error_ptr,
2549         user_error_fn, user_warning_fn);
2550
2551    if (!png_ptr)
2552        return (ERROR);
2553
2554    info_ptr = png_create_info_struct(png_ptr);
2555
2556    if (!info_ptr)
2557    {
2558       png_destroy_read_struct(&png_ptr,
2559          (png_infopp)NULL, (png_infopp)NULL);
2560       return (ERROR);
2561    }
2562
2563    if (setjmp(png_jmpbuf(png_ptr)))
2564    {
2565       png_destroy_read_struct(&png_ptr, &info_ptr,
2566          (png_infopp)NULL);
2567       return (ERROR);
2568    }
2569
2570    /* This one's new.  You can provide functions
2571       to be called when the header info is valid,
2572       when each row is completed, and when the image
2573       is finished.  If you aren't using all functions,
2574       you can specify NULL parameters.  Even when all
2575       three functions are NULL, you need to call
2576       png_set_progressive_read_fn().  You can use
2577       any struct as the user_ptr (cast to a void pointer
2578       for the function call), and retrieve the pointer
2579       from inside the callbacks using the function
2580
2581          png_get_progressive_ptr(png_ptr);
2582
2583       which will return a void pointer, which you have
2584       to cast appropriately.
2585     */
2586    png_set_progressive_read_fn(png_ptr, (void *)user_ptr,
2587        info_callback, row_callback, end_callback);
2588
2589    return 0;
2590 }
2591
2592 /* A code fragment that you call as you receive blocks
2593   of data */
2594 int
2595 process_data(png_bytep buffer, png_uint_32 length)
2596 {
2597    if (setjmp(png_jmpbuf(png_ptr)))
2598    {
2599       png_destroy_read_struct(&png_ptr, &info_ptr,
2600           (png_infopp)NULL);
2601       return (ERROR);
2602    }
2603
2604    /* This one's new also.  Simply give it a chunk
2605       of data from the file stream (in order, of
2606       course).  On machines with segmented memory
2607       models machines, don't give it any more than
2608       64K.  The library seems to run fine with sizes
2609       of 4K. Although you can give it much less if
2610       necessary (I assume you can give it chunks of
2611       1 byte, I haven't tried less than 256 bytes
2612       yet).  When this function returns, you may
2613       want to display any rows that were generated
2614       in the row callback if you don't already do
2615       so there.
2616     */
2617    png_process_data(png_ptr, info_ptr, buffer, length);
2618
2619    /* At this point you can call png_process_data_skip if
2620       you want to handle data the library will skip yourself;
2621       it simply returns the number of bytes to skip (and stops
2622       libpng skipping that number of bytes on the next
2623       png_process_data call).
2624    return 0;
2625 }
2626
2627 /* This function is called (as set by
2628    png_set_progressive_read_fn() above) when enough data
2629    has been supplied so all of the header has been
2630    read.
2631 */
2632 void
2633 info_callback(png_structp png_ptr, png_infop info)
2634 {
2635    /* Do any setup here, including setting any of
2636       the transformations mentioned in the Reading
2637       PNG files section.  For now, you _must_ call
2638       either png_start_read_image() or
2639       png_read_update_info() after all the
2640       transformations are set (even if you don't set
2641       any).  You may start getting rows before
2642       png_process_data() returns, so this is your
2643       last chance to prepare for that.
2644
2645       This is where you turn on interlace handling,
2646       assuming you don't want to do it yourself.
2647
2648       If you need to you can stop the processing of
2649       your original input data at this point by calling
2650       png_process_data_pause.  This returns the number
2651       of unprocessed bytes from the last png_process_data
2652       call - it is up to you to ensure that the next call
2653       sees these bytes again.  If you don't want to bother
2654       with this you can get libpng to cache the unread
2655       bytes by setting the 'save' parameter (see png.h) but
2656       then libpng will have to copy the data internally.
2657     */
2658 }
2659
2660 /* This function is called when each row of image
2661    data is complete */
2662 void
2663 row_callback(png_structp png_ptr, png_bytep new_row,
2664    png_uint_32 row_num, int pass)
2665 {
2666    /* If the image is interlaced, and you turned
2667       on the interlace handler, this function will
2668       be called for every row in every pass.  Some
2669       of these rows will not be changed from the
2670       previous pass.  When the row is not changed,
2671       the new_row variable will be NULL.  The rows
2672       and passes are called in order, so you don't
2673       really need the row_num and pass, but I'm
2674       supplying them because it may make your life
2675       easier.
2676
2677       If you did not turn on interlace handling then
2678       the callback is called for each row of each
2679       sub-image when the image is interlaced.  In this
2680       case 'row_num' is the row in the sub-image, not
2681       the row in the output image as it is in all other
2682       cases.
2683
2684       For the non-NULL rows of interlaced images when
2685       you have switched on libpng interlace handling,
2686       you must call png_progressive_combine_row()
2687       passing in the row and the old row.  You can
2688       call this function for NULL rows (it will just
2689       return) and for non-interlaced images (it just
2690       does the memcpy for you) if it will make the
2691       code easier.  Thus, you can just do this for
2692       all cases if you switch on interlace handling;
2693     */
2694
2695        png_progressive_combine_row(png_ptr, old_row,
2696          new_row);
2697
2698    /* where old_row is what was displayed
2699       previously for the row.  Note that the first
2700       pass (pass == 0, really) will completely cover
2701       the old row, so the rows do not have to be
2702       initialized.  After the first pass (and only
2703       for interlaced images), you will have to pass
2704       the current row, and the function will combine
2705       the old row and the new row.
2706
2707       You can also call png_process_data_pause in this
2708       callback - see above.
2709    */
2710 }
2711
2712 void
2713 end_callback(png_structp png_ptr, png_infop info)
2714 {
2715    /* This function is called after the whole image
2716       has been read, including any chunks after the
2717       image (up to and including the IEND).  You
2718       will usually have the same info chunk as you
2719       had in the header, although some data may have
2720       been added to the comments and time fields.
2721
2722       Most people won't do much here, perhaps setting
2723       a flag that marks the image as finished.
2724     */
2725 }
2726
2727
2728
2729IV. Writing
2730
2731Much of this is very similar to reading.  However, everything of
2732importance is repeated here, so you won't have to constantly look
2733back up in the reading section to understand writing.
2734
2735Setup
2736
2737You will want to do the I/O initialization before you get into libpng,
2738so if it doesn't work, you don't have anything to undo. If you are not
2739using the standard I/O functions, you will need to replace them with
2740custom writing functions.  See the discussion under Customizing libpng.
2741
2742    FILE *fp = fopen(file_name, "wb");
2743
2744    if (!fp)
2745       return (ERROR);
2746
2747Next, png_struct and png_info need to be allocated and initialized.
2748As these can be both relatively large, you may not want to store these
2749on the stack, unless you have stack space to spare.  Of course, you
2750will want to check if they return NULL.  If you are also reading,
2751you won't want to name your read structure and your write structure
2752both "png_ptr"; you can call them anything you like, such as
2753"read_ptr" and "write_ptr".  Look at pngtest.c, for example.
2754
2755    png_structp png_ptr = png_create_write_struct
2756       (PNG_LIBPNG_VER_STRING, (png_voidp)user_error_ptr,
2757        user_error_fn, user_warning_fn);
2758
2759    if (!png_ptr)
2760       return (ERROR);
2761
2762    png_infop info_ptr = png_create_info_struct(png_ptr);
2763    if (!info_ptr)
2764    {
2765       png_destroy_write_struct(&png_ptr,
2766           (png_infopp)NULL);
2767       return (ERROR);
2768    }
2769
2770If you want to use your own memory allocation routines,
2771define PNG_USER_MEM_SUPPORTED and use
2772png_create_write_struct_2() instead of png_create_write_struct():
2773
2774    png_structp png_ptr = png_create_write_struct_2
2775       (PNG_LIBPNG_VER_STRING, (png_voidp)user_error_ptr,
2776        user_error_fn, user_warning_fn, (png_voidp)
2777        user_mem_ptr, user_malloc_fn, user_free_fn);
2778
2779After you have these structures, you will need to set up the
2780error handling.  When libpng encounters an error, it expects to
2781longjmp() back to your routine.  Therefore, you will need to call
2782setjmp() and pass the png_jmpbuf(png_ptr).  If you
2783write the file from different routines, you will need to update
2784the png_jmpbuf(png_ptr) every time you enter a new routine that will
2785call a png_*() function.  See your documentation of setjmp/longjmp
2786for your compiler for more information on setjmp/longjmp.  See
2787the discussion on libpng error handling in the Customizing Libpng
2788section below for more information on the libpng error handling.
2789
2790    if (setjmp(png_jmpbuf(png_ptr)))
2791    {
2792    png_destroy_write_struct(&png_ptr, &info_ptr);
2793       fclose(fp);
2794       return (ERROR);
2795    }
2796    ...
2797    return;
2798
2799If you would rather avoid the complexity of setjmp/longjmp issues,
2800you can compile libpng with PNG_NO_SETJMP, in which case
2801errors will result in a call to PNG_ABORT() which defaults to abort().
2802
2803You can #define PNG_ABORT() to a function that does something
2804more useful than abort(), as long as your function does not
2805return.
2806
2807Checking for invalid palette index on write was added at libpng
28081.5.10.  If a pixel contains an invalid (out-of-range) index libpng issues
2809a benign error.  This is enabled by default because this condition is an
2810error according to the PNG specification, Clause 11.3.2, but the error can
2811be ignored in each png_ptr with
2812
2813   png_set_check_for_invalid_index(png_ptr, 0);
2814
2815If the error is ignored, or if png_benign_error() treats it as a warning,
2816any invalid pixels are written as-is by the encoder, resulting in an
2817invalid PNG datastream as output.  In this case the application is
2818responsible for ensuring that the pixel indexes are in range when it writes
2819a PLTE chunk with fewer entries than the bit depth would allow.
2820
2821Now you need to set up the output code.  The default for libpng is to
2822use the C function fwrite().  If you use this, you will need to pass a
2823valid FILE * in the function png_init_io().  Be sure that the file is
2824opened in binary mode.  Again, if you wish to handle writing data in
2825another way, see the discussion on libpng I/O handling in the Customizing
2826Libpng section below.
2827
2828    png_init_io(png_ptr, fp);
2829
2830If you are embedding your PNG into a datastream such as MNG, and don't
2831want libpng to write the 8-byte signature, or if you have already
2832written the signature in your application, use
2833
2834    png_set_sig_bytes(png_ptr, 8);
2835
2836to inform libpng that it should not write a signature.
2837
2838Write callbacks
2839
2840At this point, you can set up a callback function that will be
2841called after each row has been written, which you can use to control
2842a progress meter or the like.  It's demonstrated in pngtest.c.
2843You must supply a function
2844
2845    void write_row_callback(png_structp png_ptr, png_uint_32 row,
2846       int pass);
2847    {
2848      /* put your code here */
2849    }
2850
2851(You can give it another name that you like instead of "write_row_callback")
2852
2853To inform libpng about your function, use
2854
2855    png_set_write_status_fn(png_ptr, write_row_callback);
2856
2857When this function is called the row has already been completely processed and
2858it has also been written out.  The 'row' and 'pass' refer to the next row to be
2859handled.  For the
2860non-interlaced case the row that was just handled is simply one less than the
2861passed in row number, and pass will always be 0.  For the interlaced case the
2862same applies unless the row value is 0, in which case the row just handled was
2863the last one from one of the preceding passes.  Because interlacing may skip a
2864pass you cannot be sure that the preceding pass is just 'pass-1', if you really
2865need to know what the last pass is record (row,pass) from the callback and use
2866the last recorded value each time.
2867
2868As with the user transform you can find the output row using the
2869PNG_ROW_FROM_PASS_ROW macro.
2870
2871You now have the option of modifying how the compression library will
2872run.  The following functions are mainly for testing, but may be useful
2873in some cases, like if you need to write PNG files extremely fast and
2874are willing to give up some compression, or if you want to get the
2875maximum possible compression at the expense of slower writing.  If you
2876have no special needs in this area, let the library do what it wants by
2877not calling this function at all, as it has been tuned to deliver a good
2878speed/compression ratio. The second parameter to png_set_filter() is
2879the filter method, for which the only valid values are 0 (as of the
2880July 1999 PNG specification, version 1.2) or 64 (if you are writing
2881a PNG datastream that is to be embedded in a MNG datastream).  The third
2882parameter is a flag that indicates which filter type(s) are to be tested
2883for each scanline.  See the PNG specification for details on the specific
2884filter types.
2885
2886
2887    /* turn on or off filtering, and/or choose
2888       specific filters.  You can use either a single
2889       PNG_FILTER_VALUE_NAME or the bitwise OR of one
2890       or more PNG_FILTER_NAME masks.
2891     */
2892    png_set_filter(png_ptr, 0,
2893       PNG_FILTER_NONE  | PNG_FILTER_VALUE_NONE |
2894       PNG_FILTER_SUB   | PNG_FILTER_VALUE_SUB  |
2895       PNG_FILTER_UP    | PNG_FILTER_VALUE_UP   |
2896       PNG_FILTER_AVG   | PNG_FILTER_VALUE_AVG  |
2897       PNG_FILTER_PAETH | PNG_FILTER_VALUE_PAETH|
2898       PNG_ALL_FILTERS  | PNG_FAST_FILTERS);
2899
2900If an application wants to start and stop using particular filters during
2901compression, it should start out with all of the filters (to ensure that
2902the previous row of pixels will be stored in case it's needed later),
2903and then add and remove them after the start of compression.
2904
2905If you are writing a PNG datastream that is to be embedded in a MNG
2906datastream, the second parameter can be either 0 or 64.
2907
2908The png_set_compression_*() functions interface to the zlib compression
2909library, and should mostly be ignored unless you really know what you are
2910doing.  The only generally useful call is png_set_compression_level()
2911which changes how much time zlib spends on trying to compress the image
2912data.  See the Compression Library (zlib.h and algorithm.txt, distributed
2913with zlib) for details on the compression levels.
2914
2915    #include zlib.h
2916
2917    /* Set the zlib compression level */
2918    png_set_compression_level(png_ptr,
2919        Z_BEST_COMPRESSION);
2920
2921    /* Set other zlib parameters for compressing IDAT */
2922    png_set_compression_mem_level(png_ptr, 8);
2923    png_set_compression_strategy(png_ptr,
2924        Z_DEFAULT_STRATEGY);
2925    png_set_compression_window_bits(png_ptr, 15);
2926    png_set_compression_method(png_ptr, 8);
2927    png_set_compression_buffer_size(png_ptr, 8192)
2928
2929    /* Set zlib parameters for text compression
2930     * If you don't call these, the parameters
2931     * fall back on those defined for IDAT chunks
2932     */
2933    png_set_text_compression_mem_level(png_ptr, 8);
2934    png_set_text_compression_strategy(png_ptr,
2935        Z_DEFAULT_STRATEGY);
2936    png_set_text_compression_window_bits(png_ptr, 15);
2937    png_set_text_compression_method(png_ptr, 8);
2938
2939Setting the contents of info for output
2940
2941You now need to fill in the png_info structure with all the data you
2942wish to write before the actual image.  Note that the only thing you
2943are allowed to write after the image is the text chunks and the time
2944chunk (as of PNG Specification 1.2, anyway).  See png_write_end() and
2945the latest PNG specification for more information on that.  If you
2946wish to write them before the image, fill them in now, and flag that
2947data as being valid.  If you want to wait until after the data, don't
2948fill them until png_write_end().  For all the fields in png_info and
2949their data types, see png.h.  For explanations of what the fields
2950contain, see the PNG specification.
2951
2952Some of the more important parts of the png_info are:
2953
2954    png_set_IHDR(png_ptr, info_ptr, width, height,
2955       bit_depth, color_type, interlace_type,
2956       compression_type, filter_method)
2957
2958    width          - holds the width of the image
2959                     in pixels (up to 2^31).
2960
2961    height         - holds the height of the image
2962                     in pixels (up to 2^31).
2963
2964    bit_depth      - holds the bit depth of one of the
2965                     image channels.
2966                     (valid values are 1, 2, 4, 8, 16
2967                     and depend also on the
2968                     color_type.  See also significant
2969                     bits (sBIT) below).
2970
2971    color_type     - describes which color/alpha
2972                     channels are present.
2973                     PNG_COLOR_TYPE_GRAY
2974                        (bit depths 1, 2, 4, 8, 16)
2975                     PNG_COLOR_TYPE_GRAY_ALPHA
2976                        (bit depths 8, 16)
2977                     PNG_COLOR_TYPE_PALETTE
2978                        (bit depths 1, 2, 4, 8)
2979                     PNG_COLOR_TYPE_RGB
2980                        (bit_depths 8, 16)
2981                     PNG_COLOR_TYPE_RGB_ALPHA
2982                        (bit_depths 8, 16)
2983
2984                     PNG_COLOR_MASK_PALETTE
2985                     PNG_COLOR_MASK_COLOR
2986                     PNG_COLOR_MASK_ALPHA
2987
2988    interlace_type - PNG_INTERLACE_NONE or
2989                     PNG_INTERLACE_ADAM7
2990
2991    compression_type - (must be
2992                     PNG_COMPRESSION_TYPE_DEFAULT)
2993
2994    filter_method  - (must be PNG_FILTER_TYPE_DEFAULT
2995                     or, if you are writing a PNG to
2996                     be embedded in a MNG datastream,
2997                     can also be
2998                     PNG_INTRAPIXEL_DIFFERENCING)
2999
3000If you call png_set_IHDR(), the call must appear before any of the
3001other png_set_*() functions, because they might require access to some of
3002the IHDR settings.  The remaining png_set_*() functions can be called
3003in any order.
3004
3005If you wish, you can reset the compression_type, interlace_type, or
3006filter_method later by calling png_set_IHDR() again; if you do this, the
3007width, height, bit_depth, and color_type must be the same in each call.
3008
3009    png_set_PLTE(png_ptr, info_ptr, palette,
3010       num_palette);
3011
3012    palette        - the palette for the file
3013                     (array of png_color)
3014    num_palette    - number of entries in the palette
3015
3016
3017    png_set_gAMA(png_ptr, info_ptr, file_gamma);
3018    png_set_gAMA_fixed(png_ptr, info_ptr, int_file_gamma);
3019
3020    file_gamma     - the gamma at which the image was
3021                     created (PNG_INFO_gAMA)
3022
3023    int_file_gamma - 100,000 times the gamma at which
3024                     the image was created
3025
3026    png_set_cHRM(png_ptr, info_ptr,  white_x, white_y, red_x, red_y,
3027                     green_x, green_y, blue_x, blue_y)
3028    png_set_cHRM_XYZ(png_ptr, info_ptr, red_X, red_Y, red_Z, green_X,
3029                     green_Y, green_Z, blue_X, blue_Y, blue_Z)
3030    png_set_cHRM_fixed(png_ptr, info_ptr, int_white_x, int_white_y,
3031                     int_red_x, int_red_y, int_green_x, int_green_y,
3032                     int_blue_x, int_blue_y)
3033    png_set_cHRM_XYZ_fixed(png_ptr, info_ptr, int_red_X, int_red_Y,
3034                     int_red_Z, int_green_X, int_green_Y, int_green_Z,
3035                     int_blue_X, int_blue_Y, int_blue_Z)
3036
3037    {white,red,green,blue}_{x,y}
3038                     A color space encoding specified using the chromaticities
3039                     of the end points and the white point.
3040
3041    {red,green,blue}_{X,Y,Z}
3042                     A color space encoding specified using the encoding end
3043                     points - the CIE tristimulus specification of the intended
3044                     color of the red, green and blue channels in the PNG RGB
3045                     data.  The white point is simply the sum of the three end
3046                     points.
3047
3048    png_set_sRGB(png_ptr, info_ptr, srgb_intent);
3049
3050    srgb_intent    - the rendering intent
3051                     (PNG_INFO_sRGB) The presence of
3052                     the sRGB chunk means that the pixel
3053                     data is in the sRGB color space.
3054                     This chunk also implies specific
3055                     values of gAMA and cHRM.  Rendering
3056                     intent is the CSS-1 property that
3057                     has been defined by the International
3058                     Color Consortium
3059                     (http://www.color.org).
3060                     It can be one of
3061                     PNG_sRGB_INTENT_SATURATION,
3062                     PNG_sRGB_INTENT_PERCEPTUAL,
3063                     PNG_sRGB_INTENT_ABSOLUTE, or
3064                     PNG_sRGB_INTENT_RELATIVE.
3065
3066
3067    png_set_sRGB_gAMA_and_cHRM(png_ptr, info_ptr,
3068       srgb_intent);
3069
3070    srgb_intent    - the rendering intent
3071                     (PNG_INFO_sRGB) The presence of the
3072                     sRGB chunk means that the pixel
3073                     data is in the sRGB color space.
3074                     This function also causes gAMA and
3075                     cHRM chunks with the specific values
3076                     that are consistent with sRGB to be
3077                     written.
3078
3079    png_set_iCCP(png_ptr, info_ptr, name, compression_type,
3080                       profile, proflen);
3081
3082    name             - The profile name.
3083
3084    compression_type - The compression type; always
3085                       PNG_COMPRESSION_TYPE_BASE for PNG 1.0.
3086                       You may give NULL to this argument to
3087                       ignore it.
3088
3089    profile          - International Color Consortium color
3090                       profile data. May contain NULs.
3091
3092    proflen          - length of profile data in bytes.
3093
3094    png_set_sBIT(png_ptr, info_ptr, sig_bit);
3095
3096    sig_bit        - the number of significant bits for
3097                     (PNG_INFO_sBIT) each of the gray, red,
3098                     green, and blue channels, whichever are
3099                     appropriate for the given color type
3100                     (png_color_16)
3101
3102    png_set_tRNS(png_ptr, info_ptr, trans_alpha,
3103       num_trans, trans_color);
3104
3105    trans_alpha    - array of alpha (transparency)
3106                     entries for palette (PNG_INFO_tRNS)
3107
3108    num_trans      - number of transparent entries
3109                     (PNG_INFO_tRNS)
3110
3111    trans_color    - graylevel or color sample values
3112                     (in order red, green, blue) of the
3113                     single transparent color for
3114                     non-paletted images (PNG_INFO_tRNS)
3115
3116    png_set_eXIf_1(png_ptr, info_ptr, num_exif, exif);
3117
3118    exif           - Exif profile (array of
3119                     png_byte) (PNG_INFO_eXIf)
3120
3121    png_set_hIST(png_ptr, info_ptr, hist);
3122
3123    hist           - histogram of palette (array of
3124                     png_uint_16) (PNG_INFO_hIST)
3125
3126    png_set_tIME(png_ptr, info_ptr, mod_time);
3127
3128    mod_time       - time image was last modified
3129                     (PNG_VALID_tIME)
3130
3131    png_set_bKGD(png_ptr, info_ptr, background);
3132
3133    background     - background color (of type
3134                     png_color_16p) (PNG_VALID_bKGD)
3135
3136    png_set_text(png_ptr, info_ptr, text_ptr, num_text);
3137
3138    text_ptr       - array of png_text holding image
3139                     comments
3140
3141    text_ptr[i].compression - type of compression used
3142                 on "text" PNG_TEXT_COMPRESSION_NONE
3143                           PNG_TEXT_COMPRESSION_zTXt
3144                           PNG_ITXT_COMPRESSION_NONE
3145                           PNG_ITXT_COMPRESSION_zTXt
3146    text_ptr[i].key   - keyword for comment.  Must contain
3147                 1-79 characters.
3148    text_ptr[i].text  - text comments for current
3149                         keyword.  Can be NULL or empty.
3150    text_ptr[i].text_length - length of text string,
3151                 after decompression, 0 for iTXt
3152    text_ptr[i].itxt_length - length of itxt string,
3153                 after decompression, 0 for tEXt/zTXt
3154    text_ptr[i].lang  - language of comment (NULL or
3155                         empty for unknown).
3156    text_ptr[i].translated_keyword  - keyword in UTF-8 (NULL
3157                         or empty for unknown).
3158
3159    Note that the itxt_length, lang, and lang_key
3160    members of the text_ptr structure only exist when the
3161    library is built with iTXt chunk support.  Prior to
3162    libpng-1.4.0 the library was built by default without
3163    iTXt support. Also note that when iTXt is supported,
3164    they contain NULL pointers when the "compression"
3165    field contains PNG_TEXT_COMPRESSION_NONE or
3166    PNG_TEXT_COMPRESSION_zTXt.
3167
3168    num_text       - number of comments
3169
3170    png_set_sPLT(png_ptr, info_ptr, &palette_ptr,
3171       num_spalettes);
3172
3173    palette_ptr    - array of png_sPLT_struct structures
3174                     to be added to the list of palettes
3175                     in the info structure.
3176    num_spalettes  - number of palette structures to be
3177                     added.
3178
3179    png_set_oFFs(png_ptr, info_ptr, offset_x, offset_y,
3180        unit_type);
3181
3182    offset_x  - positive offset from the left
3183                     edge of the screen
3184
3185    offset_y  - positive offset from the top
3186                     edge of the screen
3187
3188    unit_type - PNG_OFFSET_PIXEL, PNG_OFFSET_MICROMETER
3189
3190    png_set_pHYs(png_ptr, info_ptr, res_x, res_y,
3191        unit_type);
3192
3193    res_x       - pixels/unit physical resolution
3194                  in x direction
3195
3196    res_y       - pixels/unit physical resolution
3197                  in y direction
3198
3199    unit_type   - PNG_RESOLUTION_UNKNOWN,
3200                  PNG_RESOLUTION_METER
3201
3202    png_set_sCAL(png_ptr, info_ptr, unit, width, height)
3203
3204    unit        - physical scale units (an integer)
3205
3206    width       - width of a pixel in physical scale units
3207
3208    height      - height of a pixel in physical scale units
3209                  (width and height are doubles)
3210
3211    png_set_sCAL_s(png_ptr, info_ptr, unit, width, height)
3212
3213    unit        - physical scale units (an integer)
3214
3215    width       - width of a pixel in physical scale units
3216                  expressed as a string
3217
3218    height      - height of a pixel in physical scale units
3219                 (width and height are strings like "2.54")
3220
3221    png_set_unknown_chunks(png_ptr, info_ptr, &unknowns,
3222       num_unknowns)
3223
3224    unknowns          - array of png_unknown_chunk
3225                        structures holding unknown chunks
3226    unknowns[i].name  - name of unknown chunk
3227    unknowns[i].data  - data of unknown chunk
3228    unknowns[i].size  - size of unknown chunk's data
3229    unknowns[i].location - position to write chunk in file
3230                           0: do not write chunk
3231                           PNG_HAVE_IHDR: before PLTE
3232                           PNG_HAVE_PLTE: before IDAT
3233                           PNG_AFTER_IDAT: after IDAT
3234
3235The "location" member is set automatically according to
3236what part of the output file has already been written.
3237You can change its value after calling png_set_unknown_chunks()
3238as demonstrated in pngtest.c.  Within each of the "locations",
3239the chunks are sequenced according to their position in the
3240structure (that is, the value of "i", which is the order in which
3241the chunk was either read from the input file or defined with
3242png_set_unknown_chunks).
3243
3244A quick word about text and num_text.  text is an array of png_text
3245structures.  num_text is the number of valid structures in the array.
3246Each png_text structure holds a language code, a keyword, a text value,
3247and a compression type.
3248
3249The compression types have the same valid numbers as the compression
3250types of the image data.  Currently, the only valid number is zero.
3251However, you can store text either compressed or uncompressed, unlike
3252images, which always have to be compressed.  So if you don't want the
3253text compressed, set the compression type to PNG_TEXT_COMPRESSION_NONE.
3254Because tEXt and zTXt chunks don't have a language field, if you
3255specify PNG_TEXT_COMPRESSION_NONE or PNG_TEXT_COMPRESSION_zTXt
3256any language code or translated keyword will not be written out.
3257
3258Until text gets around a few hundred bytes, it is not worth compressing it.
3259After the text has been written out to the file, the compression type
3260is set to PNG_TEXT_COMPRESSION_NONE_WR or PNG_TEXT_COMPRESSION_zTXt_WR,
3261so that it isn't written out again at the end (in case you are calling
3262png_write_end() with the same struct).
3263
3264The keywords that are given in the PNG Specification are:
3265
3266    Title            Short (one line) title or
3267                     caption for image
3268
3269    Author           Name of image's creator
3270
3271    Description      Description of image (possibly long)
3272
3273    Copyright        Copyright notice
3274
3275    Creation Time    Time of original image creation
3276                     (usually RFC 1123 format, see below)
3277
3278    Software         Software used to create the image
3279
3280    Disclaimer       Legal disclaimer
3281
3282    Warning          Warning of nature of content
3283
3284    Source           Device used to create the image
3285
3286    Comment          Miscellaneous comment; conversion
3287                     from other image format
3288
3289The keyword-text pairs work like this.  Keywords should be short
3290simple descriptions of what the comment is about.  Some typical
3291keywords are found in the PNG specification, as is some recommendations
3292on keywords.  You can repeat keywords in a file.  You can even write
3293some text before the image and some after.  For example, you may want
3294to put a description of the image before the image, but leave the
3295disclaimer until after, so viewers working over modem connections
3296don't have to wait for the disclaimer to go over the modem before
3297they start seeing the image.  Finally, keywords should be full
3298words, not abbreviations.  Keywords and text are in the ISO 8859-1
3299(Latin-1) character set (a superset of regular ASCII) and can not
3300contain NUL characters, and should not contain control or other
3301unprintable characters.  To make the comments widely readable, stick
3302with basic ASCII, and avoid machine specific character set extensions
3303like the IBM-PC character set.  The keyword must be present, but
3304you can leave off the text string on non-compressed pairs.
3305Compressed pairs must have a text string, as only the text string
3306is compressed anyway, so the compression would be meaningless.
3307
3308PNG supports modification time via the png_time structure.  Two
3309conversion routines are provided, png_convert_from_time_t() for
3310time_t and png_convert_from_struct_tm() for struct tm.  The
3311time_t routine uses gmtime().  You don't have to use either of
3312these, but if you wish to fill in the png_time structure directly,
3313you should provide the time in universal time (GMT) if possible
3314instead of your local time.  Note that the year number is the full
3315year (e.g. 1998, rather than 98 - PNG is year 2000 compliant!), and
3316that months start with 1.
3317
3318If you want to store the time of the original image creation, you should
3319use a plain tEXt chunk with the "Creation Time" keyword.  This is
3320necessary because the "creation time" of a PNG image is somewhat vague,
3321depending on whether you mean the PNG file, the time the image was
3322created in a non-PNG format, a still photo from which the image was
3323scanned, or possibly the subject matter itself.  In order to facilitate
3324machine-readable dates, it is recommended that the "Creation Time"
3325tEXt chunk use RFC 1123 format dates (e.g. "22 May 1997 18:07:10 GMT"),
3326although this isn't a requirement.  Unlike the tIME chunk, the
3327"Creation Time" tEXt chunk is not expected to be automatically changed
3328by the software.  To facilitate the use of RFC 1123 dates, a function
3329png_convert_to_rfc1123_buffer(buffer, png_timep) is provided to
3330convert from PNG time to an RFC 1123 format string.  The caller must provide
3331a writeable buffer of at least 29 bytes.
3332
3333Writing unknown chunks
3334
3335You can use the png_set_unknown_chunks function to queue up private chunks
3336for writing.  You give it a chunk name, location, raw data, and a size.  You
3337also must use png_set_keep_unknown_chunks() to ensure that libpng will
3338handle them.  That's all there is to it.  The chunks will be written by the
3339next following png_write_info_before_PLTE, png_write_info, or png_write_end
3340function, depending upon the specified location.  Any chunks previously
3341read into the info structure's unknown-chunk list will also be written out
3342in a sequence that satisfies the PNG specification's ordering rules.
3343
3344Here is an example of writing two private chunks, prVt and miNE:
3345
3346    #ifdef PNG_WRITE_UNKNOWN_CHUNKS_SUPPORTED
3347    /* Set unknown chunk data */
3348    png_unknown_chunk unk_chunk[2];
3349    strcpy((char *) unk_chunk[0].name, "prVt";
3350    unk_chunk[0].data = (unsigned char *) "PRIVATE DATA";
3351    unk_chunk[0].size = strlen(unk_chunk[0].data)+1;
3352    unk_chunk[0].location = PNG_HAVE_IHDR;
3353    strcpy((char *) unk_chunk[1].name, "miNE";
3354    unk_chunk[1].data = (unsigned char *) "MY CHUNK DATA";
3355    unk_chunk[1].size = strlen(unk_chunk[0].data)+1;
3356    unk_chunk[1].location = PNG_AFTER_IDAT;
3357    png_set_unknown_chunks(write_ptr, write_info_ptr,
3358        unk_chunk, 2);
3359    /* Needed because miNE is not safe-to-copy */
3360    png_set_keep_unknown_chunks(png, PNG_HANDLE_CHUNK_ALWAYS,
3361       (png_bytep) "miNE", 1);
3362    # if PNG_LIBPNG_VER < 10600
3363      /* Deal with unknown chunk location bug in 1.5.x and earlier */
3364      png_set_unknown_chunk_location(png, info, 0, PNG_HAVE_IHDR);
3365      png_set_unknown_chunk_location(png, info, 1, PNG_AFTER_IDAT);
3366    # endif
3367    # if PNG_LIBPNG_VER < 10500
3368      /* PNG_AFTER_IDAT writes two copies of the chunk prior to libpng-1.5.0,
3369       * one before IDAT and another after IDAT, so don't use it; only use
3370       * PNG_HAVE_IHDR location.  This call resets the location previously
3371       * set by assignment and png_set_unknown_chunk_location() for chunk 1.
3372       */
3373      png_set_unknown_chunk_location(png, info, 1, PNG_HAVE_IHDR);
3374    # endif
3375    #endif
3376
3377The high-level write interface
3378
3379At this point there are two ways to proceed; through the high-level
3380write interface, or through a sequence of low-level write operations.
3381You can use the high-level interface if your image data is present
3382in the info structure.  All defined output
3383transformations are permitted, enabled by the following masks.
3384
3385    PNG_TRANSFORM_IDENTITY      No transformation
3386    PNG_TRANSFORM_PACKING       Pack 1, 2 and 4-bit samples
3387    PNG_TRANSFORM_PACKSWAP      Change order of packed
3388                                pixels to LSB first
3389    PNG_TRANSFORM_INVERT_MONO   Invert monochrome images
3390    PNG_TRANSFORM_SHIFT         Normalize pixels to the
3391                                sBIT depth
3392    PNG_TRANSFORM_BGR           Flip RGB to BGR, RGBA
3393                                to BGRA
3394    PNG_TRANSFORM_SWAP_ALPHA    Flip RGBA to ARGB or GA
3395                                to AG
3396    PNG_TRANSFORM_INVERT_ALPHA  Change alpha from opacity
3397                                to transparency
3398    PNG_TRANSFORM_SWAP_ENDIAN   Byte-swap 16-bit samples
3399    PNG_TRANSFORM_STRIP_FILLER        Strip out filler
3400                                      bytes (deprecated).
3401    PNG_TRANSFORM_STRIP_FILLER_BEFORE Strip out leading
3402                                      filler bytes
3403    PNG_TRANSFORM_STRIP_FILLER_AFTER  Strip out trailing
3404                                      filler bytes
3405
3406If you have valid image data in the info structure (you can use
3407png_set_rows() to put image data in the info structure), simply do this:
3408
3409    png_write_png(png_ptr, info_ptr, png_transforms, NULL)
3410
3411where png_transforms is an integer containing the bitwise OR of some set of
3412transformation flags.  This call is equivalent to png_write_info(),
3413followed the set of transformations indicated by the transform mask,
3414then png_write_image(), and finally png_write_end().
3415
3416(The final parameter of this call is not yet used.  Someday it might point
3417to transformation parameters required by some future output transform.)
3418
3419You must use png_transforms and not call any png_set_transform() functions
3420when you use png_write_png().
3421
3422The low-level write interface
3423
3424If you are going the low-level route instead, you are now ready to
3425write all the file information up to the actual image data.  You do
3426this with a call to png_write_info().
3427
3428    png_write_info(png_ptr, info_ptr);
3429
3430Note that there is one transformation you may need to do before
3431png_write_info().  In PNG files, the alpha channel in an image is the
3432level of opacity.  If your data is supplied as a level of transparency,
3433you can invert the alpha channel before you write it, so that 0 is
3434fully transparent and 255 (in 8-bit or paletted images) or 65535
3435(in 16-bit images) is fully opaque, with
3436
3437    png_set_invert_alpha(png_ptr);
3438
3439This must appear before png_write_info() instead of later with the
3440other transformations because in the case of paletted images the tRNS
3441chunk data has to be inverted before the tRNS chunk is written.  If
3442your image is not a paletted image, the tRNS data (which in such cases
3443represents a single color to be rendered as transparent) won't need to
3444be changed, and you can safely do this transformation after your
3445png_write_info() call.
3446
3447If you need to write a private chunk that you want to appear before
3448the PLTE chunk when PLTE is present, you can write the PNG info in
3449two steps, and insert code to write your own chunk between them:
3450
3451    png_write_info_before_PLTE(png_ptr, info_ptr);
3452    png_set_unknown_chunks(png_ptr, info_ptr, ...);
3453    png_write_info(png_ptr, info_ptr);
3454
3455After you've written the file information, you can set up the library
3456to handle any special transformations of the image data.  The various
3457ways to transform the data will be described in the order that they
3458should occur.  This is important, as some of these change the color
3459type and/or bit depth of the data, and some others only work on
3460certain color types and bit depths.  Even though each transformation
3461checks to see if it has data that it can do something with, you should
3462make sure to only enable a transformation if it will be valid for the
3463data.  For example, don't swap red and blue on grayscale data.
3464
3465PNG files store RGB pixels packed into 3 or 6 bytes.  This code tells
3466the library to strip input data that has 4 or 8 bytes per pixel down
3467to 3 or 6 bytes (or strip 2 or 4-byte grayscale+filler data to 1 or 2
3468bytes per pixel).
3469
3470    png_set_filler(png_ptr, 0, PNG_FILLER_BEFORE);
3471
3472where the 0 is unused, and the location is either PNG_FILLER_BEFORE or
3473PNG_FILLER_AFTER, depending upon whether the filler byte in the pixel
3474is stored XRGB or RGBX.
3475
3476PNG files pack pixels of bit depths 1, 2, and 4 into bytes as small as
3477they can, resulting in, for example, 8 pixels per byte for 1 bit files.
3478If the data is supplied at 1 pixel per byte, use this code, which will
3479correctly pack the pixels into a single byte:
3480
3481    png_set_packing(png_ptr);
3482
3483PNG files reduce possible bit depths to 1, 2, 4, 8, and 16.  If your
3484data is of another bit depth, you can write an sBIT chunk into the
3485file so that decoders can recover the original data if desired.
3486
3487    /* Set the true bit depth of the image data */
3488    if (color_type & PNG_COLOR_MASK_COLOR)
3489    {
3490       sig_bit.red = true_bit_depth;
3491       sig_bit.green = true_bit_depth;
3492       sig_bit.blue = true_bit_depth;
3493    }
3494
3495    else
3496    {
3497       sig_bit.gray = true_bit_depth;
3498    }
3499
3500    if (color_type & PNG_COLOR_MASK_ALPHA)
3501    {
3502       sig_bit.alpha = true_bit_depth;
3503    }
3504
3505    png_set_sBIT(png_ptr, info_ptr, &sig_bit);
3506
3507If the data is stored in the row buffer in a bit depth other than
3508one supported by PNG (e.g. 3 bit data in the range 0-7 for a 4-bit PNG),
3509this will scale the values to appear to be the correct bit depth as
3510is required by PNG.
3511
3512    png_set_shift(png_ptr, &sig_bit);
3513
3514PNG files store 16-bit pixels in network byte order (big-endian,
3515ie. most significant bits first).  This code would be used if they are
3516supplied the other way (little-endian, i.e. least significant bits
3517first, the way PCs store them):
3518
3519    if (bit_depth > 8)
3520       png_set_swap(png_ptr);
3521
3522If you are using packed-pixel images (1, 2, or 4 bits/pixel), and you
3523need to change the order the pixels are packed into bytes, you can use:
3524
3525    if (bit_depth < 8)
3526       png_set_packswap(png_ptr);
3527
3528PNG files store 3 color pixels in red, green, blue order.  This code
3529would be used if they are supplied as blue, green, red:
3530
3531    png_set_bgr(png_ptr);
3532
3533PNG files describe monochrome as black being zero and white being
3534one. This code would be used if the pixels are supplied with this reversed
3535(black being one and white being zero):
3536
3537    png_set_invert_mono(png_ptr);
3538
3539Finally, you can write your own transformation function if none of
3540the existing ones meets your needs.  This is done by setting a callback
3541with
3542
3543    png_set_write_user_transform_fn(png_ptr,
3544       write_transform_fn);
3545
3546You must supply the function
3547
3548    void write_transform_fn(png_structp png_ptr, png_row_infop
3549       row_info, png_bytep data)
3550
3551See pngtest.c for a working example.  Your function will be called
3552before any of the other transformations are processed.  If supported
3553libpng also supplies an information routine that may be called from
3554your callback:
3555
3556   png_get_current_row_number(png_ptr);
3557   png_get_current_pass_number(png_ptr);
3558
3559This returns the current row passed to the transform.  With interlaced
3560images the value returned is the row in the input sub-image image.  Use
3561PNG_ROW_FROM_PASS_ROW(row, pass) and PNG_COL_FROM_PASS_COL(col, pass) to
3562find the output pixel (x,y) given an interlaced sub-image pixel (row,col,pass).
3563
3564The discussion of interlace handling above contains more information on how to
3565use these values.
3566
3567You can also set up a pointer to a user structure for use by your
3568callback function.
3569
3570    png_set_user_transform_info(png_ptr, user_ptr, 0, 0);
3571
3572The user_channels and user_depth parameters of this function are ignored
3573when writing; you can set them to zero as shown.
3574
3575You can retrieve the pointer via the function png_get_user_transform_ptr().
3576For example:
3577
3578    voidp write_user_transform_ptr =
3579       png_get_user_transform_ptr(png_ptr);
3580
3581It is possible to have libpng flush any pending output, either manually,
3582or automatically after a certain number of lines have been written.  To
3583flush the output stream a single time call:
3584
3585    png_write_flush(png_ptr);
3586
3587and to have libpng flush the output stream periodically after a certain
3588number of scanlines have been written, call:
3589
3590    png_set_flush(png_ptr, nrows);
3591
3592Note that the distance between rows is from the last time png_write_flush()
3593was called, or the first row of the image if it has never been called.
3594So if you write 50 lines, and then png_set_flush 25, it will flush the
3595output on the next scanline, and every 25 lines thereafter, unless
3596png_write_flush() is called before 25 more lines have been written.
3597If nrows is too small (less than about 10 lines for a 640 pixel wide
3598RGB image) the image compression may decrease noticeably (although this
3599may be acceptable for real-time applications).  Infrequent flushing will
3600only degrade the compression performance by a few percent over images
3601that do not use flushing.
3602
3603Writing the image data
3604
3605That's it for the transformations.  Now you can write the image data.
3606The simplest way to do this is in one function call.  If you have the
3607whole image in memory, you can just call png_write_image() and libpng
3608will write the image.  You will need to pass in an array of pointers to
3609each row.  This function automatically handles interlacing, so you don't
3610need to call png_set_interlace_handling() or call this function multiple
3611times, or any of that other stuff necessary with png_write_rows().
3612
3613    png_write_image(png_ptr, row_pointers);
3614
3615where row_pointers is:
3616
3617    png_byte *row_pointers[height];
3618
3619You can point to void or char or whatever you use for pixels.
3620
3621If you don't want to write the whole image at once, you can
3622use png_write_rows() instead.  If the file is not interlaced,
3623this is simple:
3624
3625    png_write_rows(png_ptr, row_pointers,
3626       number_of_rows);
3627
3628row_pointers is the same as in the png_write_image() call.
3629
3630If you are just writing one row at a time, you can do this with
3631a single row_pointer instead of an array of row_pointers:
3632
3633    png_bytep row_pointer = row;
3634
3635    png_write_row(png_ptr, row_pointer);
3636
3637When the file is interlaced, things can get a good deal more complicated.
3638The only currently (as of the PNG Specification version 1.2, dated July
36391999) defined interlacing scheme for PNG files is the "Adam7" interlace
3640scheme, that breaks down an image into seven smaller images of varying
3641size.  libpng will build these images for you, or you can do them
3642yourself.  If you want to build them yourself, see the PNG specification
3643for details of which pixels to write when.
3644
3645If you don't want libpng to handle the interlacing details, just
3646use png_set_interlace_handling() and call png_write_rows() the
3647correct number of times to write all the sub-images
3648(png_set_interlace_handling() returns the number of sub-images.)
3649
3650If you want libpng to build the sub-images, call this before you start
3651writing any rows:
3652
3653    number_of_passes = png_set_interlace_handling(png_ptr);
3654
3655This will return the number of passes needed.  Currently, this is seven,
3656but may change if another interlace type is added.
3657
3658Then write the complete image number_of_passes times.
3659
3660    png_write_rows(png_ptr, row_pointers, number_of_rows);
3661
3662Think carefully before you write an interlaced image.  Typically code that
3663reads such images reads all the image data into memory, uncompressed, before
3664doing any processing.  Only code that can display an image on the fly can
3665take advantage of the interlacing and even then the image has to be exactly
3666the correct size for the output device, because scaling an image requires
3667adjacent pixels and these are not available until all the passes have been
3668read.
3669
3670If you do write an interlaced image you will hardly ever need to handle
3671the interlacing yourself.  Call png_set_interlace_handling() and use the
3672approach described above.
3673
3674The only time it is conceivable that you will really need to write an
3675interlaced image pass-by-pass is when you have read one pass by pass and
3676made some pixel-by-pixel transformation to it, as described in the read
3677code above.  In this case use the PNG_PASS_ROWS and PNG_PASS_COLS macros
3678to determine the size of each sub-image in turn and simply write the rows
3679you obtained from the read code.
3680
3681Finishing a sequential write
3682
3683After you are finished writing the image, you should finish writing
3684the file.  If you are interested in writing comments or time, you should
3685pass an appropriately filled png_info pointer.  If you are not interested,
3686you can pass NULL.
3687
3688    png_write_end(png_ptr, info_ptr);
3689
3690When you are done, you can free all memory used by libpng like this:
3691
3692    png_destroy_write_struct(&png_ptr, &info_ptr);
3693
3694It is also possible to individually free the info_ptr members that
3695point to libpng-allocated storage with the following function:
3696
3697    png_free_data(png_ptr, info_ptr, mask, seq)
3698
3699    mask  - identifies data to be freed, a mask
3700            containing the bitwise OR of one or
3701            more of
3702              PNG_FREE_PLTE, PNG_FREE_TRNS,
3703              PNG_FREE_HIST, PNG_FREE_ICCP,
3704              PNG_FREE_PCAL, PNG_FREE_ROWS,
3705              PNG_FREE_SCAL, PNG_FREE_SPLT,
3706              PNG_FREE_TEXT, PNG_FREE_UNKN,
3707            or simply PNG_FREE_ALL
3708
3709    seq   - sequence number of item to be freed
3710            (-1 for all items)
3711
3712This function may be safely called when the relevant storage has
3713already been freed, or has not yet been allocated, or was allocated
3714by the user  and not by libpng,  and will in those cases do nothing.
3715The "seq" parameter is ignored if only one item of the selected data
3716type, such as PLTE, is allowed.  If "seq" is not -1, and multiple items
3717are allowed for the data type identified in the mask, such as text or
3718sPLT, only the n'th item in the structure is freed, where n is "seq".
3719
3720If you allocated data such as a palette that you passed in to libpng
3721with png_set_*, you must not free it until just before the call to
3722png_destroy_write_struct().
3723
3724The default behavior is only to free data that was allocated internally
3725by libpng.  This can be changed, so that libpng will not free the data,
3726or so that it will free data that was allocated by the user with png_malloc()
3727or png_calloc() and passed in via a png_set_*() function, with
3728
3729    png_data_freer(png_ptr, info_ptr, freer, mask)
3730
3731    freer  - one of
3732               PNG_DESTROY_WILL_FREE_DATA
3733               PNG_SET_WILL_FREE_DATA
3734               PNG_USER_WILL_FREE_DATA
3735
3736    mask   - which data elements are affected
3737             same choices as in png_free_data()
3738
3739For example, to transfer responsibility for some data from a read structure
3740to a write structure, you could use
3741
3742    png_data_freer(read_ptr, read_info_ptr,
3743       PNG_USER_WILL_FREE_DATA,
3744       PNG_FREE_PLTE|PNG_FREE_tRNS|PNG_FREE_hIST)
3745
3746    png_data_freer(write_ptr, write_info_ptr,
3747       PNG_DESTROY_WILL_FREE_DATA,
3748       PNG_FREE_PLTE|PNG_FREE_tRNS|PNG_FREE_hIST)
3749
3750thereby briefly reassigning responsibility for freeing to the user but
3751immediately afterwards reassigning it once more to the write_destroy
3752function.  Having done this, it would then be safe to destroy the read
3753structure and continue to use the PLTE, tRNS, and hIST data in the write
3754structure.
3755
3756This function only affects data that has already been allocated.
3757You can call this function before calling after the png_set_*() functions
3758to control whether the user or png_destroy_*() is supposed to free the data.
3759When the user assumes responsibility for libpng-allocated data, the
3760application must use
3761png_free() to free it, and when the user transfers responsibility to libpng
3762for data that the user has allocated, the user must have used png_malloc()
3763or png_calloc() to allocate it.
3764
3765If you allocated text_ptr.text, text_ptr.lang, and text_ptr.translated_keyword
3766separately, do not transfer responsibility for freeing text_ptr to libpng,
3767because when libpng fills a png_text structure it combines these members with
3768the key member, and png_free_data() will free only text_ptr.key.  Similarly,
3769if you transfer responsibility for free'ing text_ptr from libpng to your
3770application, your application must not separately free those members.
3771For a more compact example of writing a PNG image, see the file example.c.
3772
3773V. Simplified API
3774
3775The simplified API, which became available in libpng-1.6.0, hides the details
3776of both libpng and the PNG file format itself.
3777It allows PNG files to be read into a very limited number of
3778in-memory bitmap formats or to be written from the same formats.  If these
3779formats do not accommodate your needs then you can, and should, use the more
3780sophisticated APIs above - these support a wide variety of in-memory formats
3781and a wide variety of sophisticated transformations to those formats as well
3782as a wide variety of APIs to manipulate ancilliary information.
3783
3784To read a PNG file using the simplified API:
3785
3786  1) Declare a 'png_image' structure (see below) on the stack, set the
3787     version field to PNG_IMAGE_VERSION and the 'opaque' pointer to NULL
3788     (this is REQUIRED, your program may crash if you don't do it.)
3789
3790  2) Call the appropriate png_image_begin_read... function.
3791
3792  3) Set the png_image 'format' member to the required sample format.
3793
3794  4) Allocate a buffer for the image and, if required, the color-map.
3795
3796  5) Call png_image_finish_read to read the image and, if required, the
3797     color-map into your buffers.
3798
3799There are no restrictions on the format of the PNG input itself; all valid
3800color types, bit depths, and interlace methods are acceptable, and the
3801input image is transformed as necessary to the requested in-memory format
3802during the png_image_finish_read() step.  The only caveat is that if you
3803request a color-mapped image from a PNG that is full-color or makes
3804complex use of an alpha channel the transformation is extremely lossy and the
3805result may look terrible.
3806
3807To write a PNG file using the simplified API:
3808
3809  1) Declare a 'png_image' structure on the stack and memset()
3810     it to all zero.
3811
3812  2) Initialize the members of the structure that describe the
3813     image, setting the 'format' member to the format of the
3814     image samples.
3815
3816  3) Call the appropriate png_image_write... function with a
3817     pointer to the image and, if necessary, the color-map to write
3818     the PNG data.
3819
3820png_image is a structure that describes the in-memory format of an image
3821when it is being read or defines the in-memory format of an image that you
3822need to write.  The "png_image" structure contains the following members:
3823
3824   png_controlp opaque  Initialize to NULL, free with png_image_free
3825   png_uint_32  version Set to PNG_IMAGE_VERSION
3826   png_uint_32  width   Image width in pixels (columns)
3827   png_uint_32  height  Image height in pixels (rows)
3828   png_uint_32  format  Image format as defined below
3829   png_uint_32  flags   A bit mask containing informational flags
3830   png_uint_32  colormap_entries; Number of entries in the color-map
3831   png_uint_32  warning_or_error;
3832   char         message[64];
3833
3834In the event of an error or warning the "warning_or_error"
3835field will be set to a non-zero value and the 'message' field will contain
3836a '\0' terminated string with the libpng error or warning message.  If both
3837warnings and an error were encountered, only the error is recorded.  If there
3838are multiple warnings, only the first one is recorded.
3839
3840The upper 30 bits of the "warning_or_error" value are reserved; the low two
3841bits contain a two bit code such that a value more than 1 indicates a failure
3842in the API just called:
3843
3844   0 - no warning or error
3845   1 - warning
3846   2 - error
3847   3 - error preceded by warning
3848
3849The pixels (samples) of the image have one to four channels whose components
3850have original values in the range 0 to 1.0:
3851
3852  1: A single gray or luminance channel (G).
3853  2: A gray/luminance channel and an alpha channel (GA).
3854  3: Three red, green, blue color channels (RGB).
3855  4: Three color channels and an alpha channel (RGBA).
3856
3857The channels are encoded in one of two ways:
3858
3859  a) As a small integer, value 0..255, contained in a single byte.  For the
3860alpha channel the original value is simply value/255.  For the color or
3861luminance channels the value is encoded according to the sRGB specification
3862and matches the 8-bit format expected by typical display devices.
3863
3864The color/gray channels are not scaled (pre-multiplied) by the alpha
3865channel and are suitable for passing to color management software.
3866
3867  b) As a value in the range 0..65535, contained in a 2-byte integer, in
3868the native byte order of the platform on which the application is running.
3869All channels can be converted to the original value by dividing by 65535; all
3870channels are linear.  Color channels use the RGB encoding (RGB end-points) of
3871the sRGB specification.  This encoding is identified by the
3872PNG_FORMAT_FLAG_LINEAR flag below.
3873
3874When the simplified API needs to convert between sRGB and linear colorspaces,
3875the actual sRGB transfer curve defined in the sRGB specification (see the
3876article at https://en.wikipedia.org/wiki/SRGB) is used, not the gamma=1/2.2
3877approximation used elsewhere in libpng.
3878
3879When an alpha channel is present it is expected to denote pixel coverage
3880of the color or luminance channels and is returned as an associated alpha
3881channel: the color/gray channels are scaled (pre-multiplied) by the alpha
3882value.
3883
3884The samples are either contained directly in the image data, between 1 and 8
3885bytes per pixel according to the encoding, or are held in a color-map indexed
3886by bytes in the image data.  In the case of a color-map the color-map entries
3887are individual samples, encoded as above, and the image data has one byte per
3888pixel to select the relevant sample from the color-map.
3889
3890PNG_FORMAT_*
3891
3892The #defines to be used in png_image::format.  Each #define identifies a
3893particular layout of channel data and, if present, alpha values.  There are
3894separate defines for each of the two component encodings.
3895
3896A format is built up using single bit flag values.  All combinations are
3897valid.  Formats can be built up from the flag values or you can use one of
3898the predefined values below.  When testing formats always use the FORMAT_FLAG
3899macros to test for individual features - future versions of the library may
3900add new flags.
3901
3902When reading or writing color-mapped images the format should be set to the
3903format of the entries in the color-map then png_image_{read,write}_colormap
3904called to read or write the color-map and set the format correctly for the
3905image data.  Do not set the PNG_FORMAT_FLAG_COLORMAP bit directly!
3906
3907NOTE: libpng can be built with particular features disabled. If you see
3908compiler errors because the definition of one of the following flags has been
3909compiled out it is because libpng does not have the required support.  It is
3910possible, however, for the libpng configuration to enable the format on just
3911read or just write; in that case you may see an error at run time.
3912You can guard against this by checking for the definition of the
3913appropriate "_SUPPORTED" macro, one of:
3914
3915   PNG_SIMPLIFIED_{READ,WRITE}_{BGR,AFIRST}_SUPPORTED
3916
3917   PNG_FORMAT_FLAG_ALPHA    format with an alpha channel
3918   PNG_FORMAT_FLAG_COLOR    color format: otherwise grayscale
3919   PNG_FORMAT_FLAG_LINEAR   2-byte channels else 1-byte
3920   PNG_FORMAT_FLAG_COLORMAP image data is color-mapped
3921   PNG_FORMAT_FLAG_BGR      BGR colors, else order is RGB
3922   PNG_FORMAT_FLAG_AFIRST   alpha channel comes first
3923
3924Supported formats are as follows.  Future versions of libpng may support more
3925formats; for compatibility with older versions simply check if the format
3926macro is defined using #ifdef.  These defines describe the in-memory layout
3927of the components of the pixels of the image.
3928
3929First the single byte (sRGB) formats:
3930
3931   PNG_FORMAT_GRAY
3932   PNG_FORMAT_GA
3933   PNG_FORMAT_AG
3934   PNG_FORMAT_RGB
3935   PNG_FORMAT_BGR
3936   PNG_FORMAT_RGBA
3937   PNG_FORMAT_ARGB
3938   PNG_FORMAT_BGRA
3939   PNG_FORMAT_ABGR
3940
3941Then the linear 2-byte formats.  When naming these "Y" is used to
3942indicate a luminance (gray) channel.  The component order within the pixel
3943is always the same - there is no provision for swapping the order of the
3944components in the linear format.  The components are 16-bit integers in
3945the native byte order for your platform, and there is no provision for
3946swapping the bytes to a different endian condition.
3947
3948   PNG_FORMAT_LINEAR_Y
3949   PNG_FORMAT_LINEAR_Y_ALPHA
3950   PNG_FORMAT_LINEAR_RGB
3951   PNG_FORMAT_LINEAR_RGB_ALPHA
3952
3953With color-mapped formats the image data is one byte for each pixel. The byte
3954is an index into the color-map which is formatted as above.  To obtain a
3955color-mapped format it is sufficient just to add the PNG_FOMAT_FLAG_COLORMAP
3956to one of the above definitions, or you can use one of the definitions below.
3957
3958   PNG_FORMAT_RGB_COLORMAP
3959   PNG_FORMAT_BGR_COLORMAP
3960   PNG_FORMAT_RGBA_COLORMAP
3961   PNG_FORMAT_ARGB_COLORMAP
3962   PNG_FORMAT_BGRA_COLORMAP
3963   PNG_FORMAT_ABGR_COLORMAP
3964
3965PNG_IMAGE macros
3966
3967These are convenience macros to derive information from a png_image
3968structure.  The PNG_IMAGE_SAMPLE_ macros return values appropriate to the
3969actual image sample values - either the entries in the color-map or the
3970pixels in the image.  The PNG_IMAGE_PIXEL_ macros return corresponding values
3971for the pixels and will always return 1 for color-mapped formats.  The
3972remaining macros return information about the rows in the image and the
3973complete image.
3974
3975NOTE: All the macros that take a png_image::format parameter are compile time
3976constants if the format parameter is, itself, a constant.  Therefore these
3977macros can be used in array declarations and case labels where required.
3978Similarly the macros are also pre-processor constants (sizeof is not used) so
3979they can be used in #if tests.
3980
3981  PNG_IMAGE_SAMPLE_CHANNELS(fmt)
3982    Returns the total number of channels in a given format: 1..4
3983
3984  PNG_IMAGE_SAMPLE_COMPONENT_SIZE(fmt)
3985    Returns the size in bytes of a single component of a pixel or color-map
3986    entry (as appropriate) in the image: 1 or 2.
3987
3988  PNG_IMAGE_SAMPLE_SIZE(fmt)
3989    This is the size of the sample data for one sample.  If the image is
3990    color-mapped it is the size of one color-map entry (and image pixels are
3991    one byte in size), otherwise it is the size of one image pixel.
3992
3993  PNG_IMAGE_MAXIMUM_COLORMAP_COMPONENTS(fmt)
3994    The maximum size of the color-map required by the format expressed in a
3995    count of components.  This can be used to compile-time allocate a
3996    color-map:
3997
3998    png_uint_16 colormap[PNG_IMAGE_MAXIMUM_COLORMAP_COMPONENTS(linear_fmt)];
3999
4000    png_byte colormap[PNG_IMAGE_MAXIMUM_COLORMAP_COMPONENTS(sRGB_fmt)];
4001
4002    Alternatively use the PNG_IMAGE_COLORMAP_SIZE macro below to use the
4003    information from one of the png_image_begin_read_ APIs and dynamically
4004    allocate the required memory.
4005
4006  PNG_IMAGE_COLORMAP_SIZE(fmt)
4007   The size of the color-map required by the format; this is the size of the
4008   color-map buffer passed to the png_image_{read,write}_colormap APIs. It is
4009   a fixed number determined by the format so can easily be allocated on the
4010   stack if necessary.
4011
4012Corresponding information about the pixels
4013
4014  PNG_IMAGE_PIXEL_CHANNELS(fmt)
4015   The number of separate channels (components) in a pixel; 1 for a
4016   color-mapped image.
4017
4018  PNG_IMAGE_PIXEL_COMPONENT_SIZE(fmt)\
4019   The size, in bytes, of each component in a pixel; 1 for a color-mapped
4020   image.
4021
4022  PNG_IMAGE_PIXEL_SIZE(fmt)
4023   The size, in bytes, of a complete pixel; 1 for a color-mapped image.
4024
4025Information about the whole row, or whole image
4026
4027  PNG_IMAGE_ROW_STRIDE(image)
4028   Returns the total number of components in a single row of the image; this
4029   is the minimum 'row stride', the minimum count of components between each
4030   row.  For a color-mapped image this is the minimum number of bytes in a
4031   row.
4032
4033   If you need the stride measured in bytes, row_stride_bytes is
4034   PNG_IMAGE_ROW_STRIDE(image) * PNG_IMAGE_PIXEL_COMPONENT_SIZE(fmt)
4035   plus any padding bytes that your application might need, for example
4036   to start the next row on a 4-byte boundary.
4037
4038  PNG_IMAGE_BUFFER_SIZE(image, row_stride)
4039   Return the size, in bytes, of an image buffer given a png_image and a row
4040   stride - the number of components to leave space for in each row.
4041
4042  PNG_IMAGE_SIZE(image)
4043   Return the size, in bytes, of the image in memory given just a png_image;
4044   the row stride is the minimum stride required for the image.
4045
4046  PNG_IMAGE_COLORMAP_SIZE(image)
4047   Return the size, in bytes, of the color-map of this image.  If the image
4048   format is not a color-map format this will return a size sufficient for
4049   256 entries in the given format; check PNG_FORMAT_FLAG_COLORMAP if
4050   you don't want to allocate a color-map in this case.
4051
4052PNG_IMAGE_FLAG_*
4053
4054Flags containing additional information about the image are held in
4055the 'flags' field of png_image.
4056
4057  PNG_IMAGE_FLAG_COLORSPACE_NOT_sRGB == 0x01
4058    This indicates the the RGB values of the in-memory bitmap do not
4059    correspond to the red, green and blue end-points defined by sRGB.
4060
4061  PNG_IMAGE_FLAG_FAST == 0x02
4062   On write emphasise speed over compression; the resultant PNG file will be
4063   larger but will be produced significantly faster, particular for large
4064   images.  Do not use this option for images which will be distributed, only
4065   used it when producing intermediate files that will be read back in
4066   repeatedly.  For a typical 24-bit image the option will double the read
4067   speed at the cost of increasing the image size by 25%, however for many
4068   more compressible images the PNG file can be 10 times larger with only a
4069   slight speed gain.
4070
4071  PNG_IMAGE_FLAG_16BIT_sRGB == 0x04
4072    On read if the image is a 16-bit per component image and there is no gAMA
4073    or sRGB chunk assume that the components are sRGB encoded.  Notice that
4074    images output by the simplified API always have gamma information; setting
4075    this flag only affects the interpretation of 16-bit images from an
4076    external source.  It is recommended that the application expose this flag
4077    to the user; the user can normally easily recognize the difference between
4078    linear and sRGB encoding.  This flag has no effect on write - the data
4079    passed to the write APIs must have the correct encoding (as defined
4080    above.)
4081
4082    If the flag is not set (the default) input 16-bit per component data is
4083    assumed to be linear.
4084
4085    NOTE: the flag can only be set after the png_image_begin_read_ call,
4086    because that call initializes the 'flags' field.
4087
4088READ APIs
4089
4090   The png_image passed to the read APIs must have been initialized by setting
4091   the png_controlp field 'opaque' to NULL (or, better, memset the whole thing.)
4092
4093   int png_image_begin_read_from_file( png_imagep image,
4094     const char *file_name)
4095
4096     The named file is opened for read and the image header
4097     is filled in from the PNG header in the file.
4098
4099   int png_image_begin_read_from_stdio (png_imagep image,
4100     FILE* file)
4101
4102      The PNG header is read from the stdio FILE object.
4103
4104   int png_image_begin_read_from_memory(png_imagep image,
4105      png_const_voidp memory, png_size_t size)
4106
4107      The PNG header is read from the given memory buffer.
4108
4109   int png_image_finish_read(png_imagep image,
4110      png_colorp background, void *buffer,
4111      png_int_32 row_stride, void *colormap));
4112
4113      Finish reading the image into the supplied buffer and
4114      clean up the png_image structure.
4115
4116      row_stride is the step, in png_byte or png_uint_16 units
4117      as appropriate, between adjacent rows.  A positive stride
4118      indicates that the top-most row is first in the buffer -
4119      the normal top-down arrangement.  A negative stride
4120      indicates that the bottom-most row is first in the buffer.
4121
4122      background need only be supplied if an alpha channel must
4123      be removed from a png_byte format and the removal is to be
4124      done by compositing on a solid color; otherwise it may be
4125      NULL and any composition will be done directly onto the
4126      buffer.  The value is an sRGB color to use for the
4127      background, for grayscale output the green channel is used.
4128
4129      For linear output removing the alpha channel is always done
4130      by compositing on black.
4131
4132   void png_image_free(png_imagep image)
4133
4134      Free any data allocated by libpng in image->opaque,
4135      setting the pointer to NULL.  May be called at any time
4136      after the structure is initialized.
4137
4138When the simplified API needs to convert between sRGB and linear colorspaces,
4139the actual sRGB transfer curve defined in the sRGB specification (see the
4140article at https://en.wikipedia.org/wiki/SRGB) is used, not the gamma=1/2.2
4141approximation used elsewhere in libpng.
4142
4143WRITE APIS
4144
4145For write you must initialize a png_image structure to describe the image to
4146be written:
4147
4148   version: must be set to PNG_IMAGE_VERSION
4149   opaque: must be initialized to NULL
4150   width: image width in pixels
4151   height: image height in rows
4152   format: the format of the data you wish to write
4153   flags: set to 0 unless one of the defined flags applies; set
4154      PNG_IMAGE_FLAG_COLORSPACE_NOT_sRGB for color format images
4155      where the RGB values do not correspond to the colors in sRGB.
4156   colormap_entries: set to the number of entries in the color-map (0 to 256)
4157
4158   int png_image_write_to_file, (png_imagep image,
4159      const char *file, int convert_to_8bit, const void *buffer,
4160      png_int_32 row_stride, const void *colormap));
4161
4162      Write the image to the named file.
4163
4164   int png_image_write_to_memory (png_imagep image, void *memory,
4165      png_alloc_size_t * PNG_RESTRICT memory_bytes,
4166      int convert_to_8_bit, const void *buffer, ptrdiff_t row_stride,
4167      const void *colormap));
4168
4169      Write the image to memory.
4170
4171   int png_image_write_to_stdio(png_imagep image, FILE *file,
4172      int convert_to_8_bit, const void *buffer,
4173      png_int_32 row_stride, const void *colormap)
4174
4175      Write the image to the given (FILE*).
4176
4177With all write APIs if image is in one of the linear formats with
4178(png_uint_16) data then setting convert_to_8_bit will cause the output to be
4179a (png_byte) PNG gamma encoded according to the sRGB specification, otherwise
4180a 16-bit linear encoded PNG file is written.
4181
4182With all APIs row_stride is handled as in the read APIs - it is the spacing
4183from one row to the next in component sized units (float) and if negative
4184indicates a bottom-up row layout in the buffer.  If you pass zero, libpng will
4185calculate the row_stride for you from the width and number of channels.
4186
4187Note that the write API does not support interlacing, sub-8-bit pixels,
4188indexed (paletted) images, or most ancillary chunks.
4189
4190VI. Modifying/Customizing libpng
4191
4192There are two issues here.  The first is changing how libpng does
4193standard things like memory allocation, input/output, and error handling.
4194The second deals with more complicated things like adding new chunks,
4195adding new transformations, and generally changing how libpng works.
4196Both of those are compile-time issues; that is, they are generally
4197determined at the time the code is written, and there is rarely a need
4198to provide the user with a means of changing them.
4199
4200Memory allocation, input/output, and error handling
4201
4202All of the memory allocation, input/output, and error handling in libpng
4203goes through callbacks that are user-settable.  The default routines are
4204in pngmem.c, pngrio.c, pngwio.c, and pngerror.c, respectively.  To change
4205these functions, call the appropriate png_set_*_fn() function.
4206
4207Memory allocation is done through the functions png_malloc(), png_calloc(),
4208and png_free().  The png_malloc() and png_free() functions currently just
4209call the standard C functions and png_calloc() calls png_malloc() and then
4210clears the newly allocated memory to zero; note that png_calloc(png_ptr, size)
4211is not the same as the calloc(number, size) function provided by stdlib.h.
4212There is limited support for certain systems with segmented memory
4213architectures and the types of pointers declared by png.h match this; you
4214will have to use appropriate pointers in your application.  If you prefer
4215to use a different method of allocating and freeing data, you can use
4216png_create_read_struct_2() or png_create_write_struct_2() to register your
4217own functions as described above.  These functions also provide a void
4218pointer that can be retrieved via
4219
4220    mem_ptr=png_get_mem_ptr(png_ptr);
4221
4222Your replacement memory functions must have prototypes as follows:
4223
4224    png_voidp malloc_fn(png_structp png_ptr,
4225       png_alloc_size_t size);
4226
4227    void free_fn(png_structp png_ptr, png_voidp ptr);
4228
4229Your malloc_fn() must return NULL in case of failure.  The png_malloc()
4230function will normally call png_error() if it receives a NULL from the
4231system memory allocator or from your replacement malloc_fn().
4232
4233Your free_fn() will never be called with a NULL ptr, since libpng's
4234png_free() checks for NULL before calling free_fn().
4235
4236Input/Output in libpng is done through png_read() and png_write(),
4237which currently just call fread() and fwrite().  The FILE * is stored in
4238png_struct and is initialized via png_init_io().  If you wish to change
4239the method of I/O, the library supplies callbacks that you can set
4240through the function png_set_read_fn() and png_set_write_fn() at run
4241time, instead of calling the png_init_io() function.  These functions
4242also provide a void pointer that can be retrieved via the function
4243png_get_io_ptr().  For example:
4244
4245    png_set_read_fn(png_structp read_ptr,
4246        voidp read_io_ptr, png_rw_ptr read_data_fn)
4247
4248    png_set_write_fn(png_structp write_ptr,
4249        voidp write_io_ptr, png_rw_ptr write_data_fn,
4250        png_flush_ptr output_flush_fn);
4251
4252    voidp read_io_ptr = png_get_io_ptr(read_ptr);
4253    voidp write_io_ptr = png_get_io_ptr(write_ptr);
4254
4255The replacement I/O functions must have prototypes as follows:
4256
4257    void user_read_data(png_structp png_ptr,
4258        png_bytep data, png_size_t length);
4259
4260    void user_write_data(png_structp png_ptr,
4261        png_bytep data, png_size_t length);
4262
4263    void user_flush_data(png_structp png_ptr);
4264
4265The user_read_data() function is responsible for detecting and
4266handling end-of-data errors.
4267
4268Supplying NULL for the read, write, or flush functions sets them back
4269to using the default C stream functions, which expect the io_ptr to
4270point to a standard *FILE structure.  It is probably a mistake
4271to use NULL for one of write_data_fn and output_flush_fn but not both
4272of them, unless you have built libpng with PNG_NO_WRITE_FLUSH defined.
4273It is an error to read from a write stream, and vice versa.
4274
4275Error handling in libpng is done through png_error() and png_warning().
4276Errors handled through png_error() are fatal, meaning that png_error()
4277should never return to its caller.  Currently, this is handled via
4278setjmp() and longjmp() (unless you have compiled libpng with
4279PNG_NO_SETJMP, in which case it is handled via PNG_ABORT()),
4280but you could change this to do things like exit() if you should wish,
4281as long as your function does not return.
4282
4283On non-fatal errors, png_warning() is called
4284to print a warning message, and then control returns to the calling code.
4285By default png_error() and png_warning() print a message on stderr via
4286fprintf() unless the library is compiled with PNG_NO_CONSOLE_IO defined
4287(because you don't want the messages) or PNG_NO_STDIO defined (because
4288fprintf() isn't available).  If you wish to change the behavior of the error
4289functions, you will need to set up your own message callbacks.  These
4290functions are normally supplied at the time that the png_struct is created.
4291It is also possible to redirect errors and warnings to your own replacement
4292functions after png_create_*_struct() has been called by calling:
4293
4294    png_set_error_fn(png_structp png_ptr,
4295        png_voidp error_ptr, png_error_ptr error_fn,
4296        png_error_ptr warning_fn);
4297
4298If NULL is supplied for either error_fn or warning_fn, then the libpng
4299default function will be used, calling fprintf() and/or longjmp() if a
4300problem is encountered.  The replacement error functions should have
4301parameters as follows:
4302
4303    void user_error_fn(png_structp png_ptr,
4304        png_const_charp error_msg);
4305
4306    void user_warning_fn(png_structp png_ptr,
4307        png_const_charp warning_msg);
4308
4309Then, within your user_error_fn or user_warning_fn, you can retrieve
4310the error_ptr if you need it, by calling
4311
4312    png_voidp error_ptr = png_get_error_ptr(png_ptr);
4313
4314The motivation behind using setjmp() and longjmp() is the C++ throw and
4315catch exception handling methods.  This makes the code much easier to write,
4316as there is no need to check every return code of every function call.
4317However, there are some uncertainties about the status of local variables
4318after a longjmp, so the user may want to be careful about doing anything
4319after setjmp returns non-zero besides returning itself.  Consult your
4320compiler documentation for more details.  For an alternative approach, you
4321may wish to use the "cexcept" facility (see https://cexcept.sourceforge.io/),
4322which is illustrated in pngvalid.c and in contrib/visupng.
4323
4324Beginning in libpng-1.4.0, the png_set_benign_errors() API became available.
4325You can use this to handle certain errors (normally handled as errors)
4326as warnings.
4327
4328    png_set_benign_errors (png_ptr, int allowed);
4329
4330    allowed: 0: treat png_benign_error() as an error.
4331             1: treat png_benign_error() as a warning.
4332
4333As of libpng-1.6.0, the default condition is to treat benign errors as
4334warnings while reading and as errors while writing.
4335
4336Custom chunks
4337
4338If you need to read or write custom chunks, you may need to get deeper
4339into the libpng code.  The library now has mechanisms for storing
4340and writing chunks of unknown type; you can even declare callbacks
4341for custom chunks.  However, this may not be good enough if the
4342library code itself needs to know about interactions between your
4343chunk and existing `intrinsic' chunks.
4344
4345If you need to write a new intrinsic chunk, first read the PNG
4346specification. Acquire a first level of understanding of how it works.
4347Pay particular attention to the sections that describe chunk names,
4348and look at how other chunks were designed, so you can do things
4349similarly.  Second, check out the sections of libpng that read and
4350write chunks.  Try to find a chunk that is similar to yours and use
4351it as a template.  More details can be found in the comments inside
4352the code.  It is best to handle private or unknown chunks in a generic method,
4353via callback functions, instead of by modifying libpng functions. This
4354is illustrated in pngtest.c, which uses a callback function to handle a
4355private "vpAg" chunk and the new "sTER" chunk, which are both unknown to
4356libpng.
4357
4358If you wish to write your own transformation for the data, look through
4359the part of the code that does the transformations, and check out some of
4360the simpler ones to get an idea of how they work.  Try to find a similar
4361transformation to the one you want to add and copy off of it.  More details
4362can be found in the comments inside the code itself.
4363
4364Configuring for gui/windowing platforms:
4365
4366You will need to write new error and warning functions that use the GUI
4367interface, as described previously, and set them to be the error and
4368warning functions at the time that png_create_*_struct() is called,
4369in order to have them available during the structure initialization.
4370They can be changed later via png_set_error_fn().  On some compilers,
4371you may also have to change the memory allocators (png_malloc, etc.).
4372
4373Configuring zlib:
4374
4375There are special functions to configure the compression.  Perhaps the
4376most useful one changes the compression level, which currently uses
4377input compression values in the range 0 - 9.  The library normally
4378uses the default compression level (Z_DEFAULT_COMPRESSION = 6).  Tests
4379have shown that for a large majority of images, compression values in
4380the range 3-6 compress nearly as well as higher levels, and do so much
4381faster.  For online applications it may be desirable to have maximum speed
4382(Z_BEST_SPEED = 1).  With versions of zlib after v0.99, you can also
4383specify no compression (Z_NO_COMPRESSION = 0), but this would create
4384files larger than just storing the raw bitmap.  You can specify the
4385compression level by calling:
4386
4387    #include zlib.h
4388    png_set_compression_level(png_ptr, level);
4389
4390Another useful one is to reduce the memory level used by the library.
4391The memory level defaults to 8, but it can be lowered if you are
4392short on memory (running DOS, for example, where you only have 640K).
4393Note that the memory level does have an effect on compression; among
4394other things, lower levels will result in sections of incompressible
4395data being emitted in smaller stored blocks, with a correspondingly
4396larger relative overhead of up to 15% in the worst case.
4397
4398    #include zlib.h
4399    png_set_compression_mem_level(png_ptr, level);
4400
4401The other functions are for configuring zlib.  They are not recommended
4402for normal use and may result in writing an invalid PNG file.  See
4403zlib.h for more information on what these mean.
4404
4405    #include zlib.h
4406    png_set_compression_strategy(png_ptr,
4407        strategy);
4408
4409    png_set_compression_window_bits(png_ptr,
4410        window_bits);
4411
4412    png_set_compression_method(png_ptr, method);
4413
4414This controls the size of the IDAT chunks (default 8192):
4415
4416    png_set_compression_buffer_size(png_ptr, size);
4417
4418As of libpng version 1.5.4, additional APIs became
4419available to set these separately for non-IDAT
4420compressed chunks such as zTXt, iTXt, and iCCP:
4421
4422    #include zlib.h
4423    #if PNG_LIBPNG_VER >= 10504
4424    png_set_text_compression_level(png_ptr, level);
4425
4426    png_set_text_compression_mem_level(png_ptr, level);
4427
4428    png_set_text_compression_strategy(png_ptr,
4429        strategy);
4430
4431    png_set_text_compression_window_bits(png_ptr,
4432        window_bits);
4433
4434    png_set_text_compression_method(png_ptr, method);
4435    #endif
4436
4437Controlling row filtering
4438
4439If you want to control whether libpng uses filtering or not, which
4440filters are used, and how it goes about picking row filters, you
4441can call one of these functions.  The selection and configuration
4442of row filters can have a significant impact on the size and
4443encoding speed and a somewhat lesser impact on the decoding speed
4444of an image.  Filtering is enabled by default for RGB and grayscale
4445images (with and without alpha), but not for paletted images nor
4446for any images with bit depths less than 8 bits/pixel.
4447
4448The 'method' parameter sets the main filtering method, which is
4449currently only '0' in the PNG 1.2 specification.  The 'filters'
4450parameter sets which filter(s), if any, should be used for each
4451scanline.  Possible values are PNG_ALL_FILTERS, PNG_NO_FILTERS,
4452or PNG_FAST_FILTERS to turn filtering on and off, or to turn on
4453just the fast-decoding subset of filters, respectively.
4454
4455Individual filter types are PNG_FILTER_NONE, PNG_FILTER_SUB,
4456PNG_FILTER_UP, PNG_FILTER_AVG, PNG_FILTER_PAETH, which can be bitwise
4457ORed together with '|' to specify one or more filters to use.
4458These filters are described in more detail in the PNG specification.
4459If you intend to change the filter type during the course of writing
4460the image, you should start with flags set for all of the filters
4461you intend to use so that libpng can initialize its internal
4462structures appropriately for all of the filter types.  (Note that this
4463means the first row must always be adaptively filtered, because libpng
4464currently does not allocate the filter buffers until png_write_row()
4465is called for the first time.)
4466
4467    filters = PNG_NO_FILTERS;
4468    filters = PNG_ALL_FILTERS;
4469    filters = PNG_FAST_FILTERS;
4470
4471    or
4472
4473    filters = PNG_FILTER_NONE | PNG_FILTER_SUB |
4474              PNG_FILTER_UP | PNG_FILTER_AVG |
4475              PNG_FILTER_PAETH;
4476
4477    png_set_filter(png_ptr, PNG_FILTER_TYPE_BASE,
4478       filters);
4479
4480              The second parameter can also be
4481              PNG_INTRAPIXEL_DIFFERENCING if you are
4482              writing a PNG to be embedded in a MNG
4483              datastream.  This parameter must be the
4484              same as the value of filter_method used
4485              in png_set_IHDR().
4486
4487Requesting debug printout
4488
4489The macro definition PNG_DEBUG can be used to request debugging
4490printout.  Set it to an integer value in the range 0 to 3.  Higher
4491numbers result in increasing amounts of debugging information.  The
4492information is printed to the "stderr" file, unless another file
4493name is specified in the PNG_DEBUG_FILE macro definition.
4494
4495When PNG_DEBUG > 0, the following functions (macros) become available:
4496
4497   png_debug(level, message)
4498   png_debug1(level, message, p1)
4499   png_debug2(level, message, p1, p2)
4500
4501in which "level" is compared to PNG_DEBUG to decide whether to print
4502the message, "message" is the formatted string to be printed,
4503and p1 and p2 are parameters that are to be embedded in the string
4504according to printf-style formatting directives.  For example,
4505
4506   png_debug1(2, "foo=%d", foo);
4507
4508is expanded to
4509
4510   if (PNG_DEBUG > 2)
4511      fprintf(PNG_DEBUG_FILE, "foo=%d\n", foo);
4512
4513When PNG_DEBUG is defined but is zero, the macros aren't defined, but you
4514can still use PNG_DEBUG to control your own debugging:
4515
4516   #ifdef PNG_DEBUG
4517       fprintf(stderr, ...
4518   #endif
4519
4520When PNG_DEBUG = 1, the macros are defined, but only png_debug statements
4521having level = 0 will be printed.  There aren't any such statements in
4522this version of libpng, but if you insert some they will be printed.
4523
4524VII.  MNG support
4525
4526The MNG specification (available at http://www.libpng.org/pub/mng) allows
4527certain extensions to PNG for PNG images that are embedded in MNG datastreams.
4528Libpng can support some of these extensions.  To enable them, use the
4529png_permit_mng_features() function:
4530
4531   feature_set = png_permit_mng_features(png_ptr, mask)
4532
4533   mask is a png_uint_32 containing the bitwise OR of the
4534        features you want to enable.  These include
4535        PNG_FLAG_MNG_EMPTY_PLTE
4536        PNG_FLAG_MNG_FILTER_64
4537        PNG_ALL_MNG_FEATURES
4538
4539   feature_set is a png_uint_32 that is the bitwise AND of
4540      your mask with the set of MNG features that is
4541      supported by the version of libpng that you are using.
4542
4543It is an error to use this function when reading or writing a standalone
4544PNG file with the PNG 8-byte signature.  The PNG datastream must be wrapped
4545in a MNG datastream.  As a minimum, it must have the MNG 8-byte signature
4546and the MHDR and MEND chunks.  Libpng does not provide support for these
4547or any other MNG chunks; your application must provide its own support for
4548them.  You may wish to consider using libmng (available at
4549https://www.libmng.com/) instead.
4550
4551VIII.  Changes to Libpng from version 0.88
4552
4553It should be noted that versions of libpng later than 0.96 are not
4554distributed by the original libpng author, Guy Schalnat, nor by
4555Andreas Dilger, who had taken over from Guy during 1996 and 1997, and
4556distributed versions 0.89 through 0.96, but rather by another member
4557of the original PNG Group, Glenn Randers-Pehrson.  Guy and Andreas are
4558still alive and well, but they have moved on to other things.
4559
4560The old libpng functions png_read_init(), png_write_init(),
4561png_info_init(), png_read_destroy(), and png_write_destroy() have been
4562moved to PNG_INTERNAL in version 0.95 to discourage their use.  These
4563functions will be removed from libpng version 1.4.0.
4564
4565The preferred method of creating and initializing the libpng structures is
4566via the png_create_read_struct(), png_create_write_struct(), and
4567png_create_info_struct() because they isolate the size of the structures
4568from the application, allow version error checking, and also allow the
4569use of custom error handling routines during the initialization, which
4570the old functions do not.  The functions png_read_destroy() and
4571png_write_destroy() do not actually free the memory that libpng
4572allocated for these structs, but just reset the data structures, so they
4573can be used instead of png_destroy_read_struct() and
4574png_destroy_write_struct() if you feel there is too much system overhead
4575allocating and freeing the png_struct for each image read.
4576
4577Setting the error callbacks via png_set_message_fn() before
4578png_read_init() as was suggested in libpng-0.88 is no longer supported
4579because this caused applications that do not use custom error functions
4580to fail if the png_ptr was not initialized to zero.  It is still possible
4581to set the error callbacks AFTER png_read_init(), or to change them with
4582png_set_error_fn(), which is essentially the same function, but with a new
4583name to force compilation errors with applications that try to use the old
4584method.
4585
4586Support for the sCAL, iCCP, iTXt, and sPLT chunks was added at libpng-1.0.6;
4587however, iTXt support was not enabled by default.
4588
4589Starting with version 1.0.7, you can find out which version of the library
4590you are using at run-time:
4591
4592   png_uint_32 libpng_vn = png_access_version_number();
4593
4594The number libpng_vn is constructed from the major version, minor
4595version with leading zero, and release number with leading zero,
4596(e.g., libpng_vn for version 1.0.7 is 10007).
4597
4598Note that this function does not take a png_ptr, so you can call it
4599before you've created one.
4600
4601You can also check which version of png.h you used when compiling your
4602application:
4603
4604   png_uint_32 application_vn = PNG_LIBPNG_VER;
4605
4606IX.  Changes to Libpng from version 1.0.x to 1.2.x
4607
4608Support for user memory management was enabled by default.  To
4609accomplish this, the functions png_create_read_struct_2(),
4610png_create_write_struct_2(), png_set_mem_fn(), png_get_mem_ptr(),
4611png_malloc_default(), and png_free_default() were added.
4612
4613Support for the iTXt chunk has been enabled by default as of
4614version 1.2.41.
4615
4616Support for certain MNG features was enabled.
4617
4618Support for numbered error messages was added.  However, we never got
4619around to actually numbering the error messages.  The function
4620png_set_strip_error_numbers() was added (Note: the prototype for this
4621function was inadvertently removed from png.h in PNG_NO_ASSEMBLER_CODE
4622builds of libpng-1.2.15.  It was restored in libpng-1.2.36).
4623
4624The png_malloc_warn() function was added at libpng-1.2.3.  This issues
4625a png_warning and returns NULL instead of aborting when it fails to
4626acquire the requested memory allocation.
4627
4628Support for setting user limits on image width and height was enabled
4629by default.  The functions png_set_user_limits(), png_get_user_width_max(),
4630and png_get_user_height_max() were added at libpng-1.2.6.
4631
4632The png_set_add_alpha() function was added at libpng-1.2.7.
4633
4634The function png_set_expand_gray_1_2_4_to_8() was added at libpng-1.2.9.
4635Unlike png_set_gray_1_2_4_to_8(), the new function does not expand the
4636tRNS chunk to alpha. The png_set_gray_1_2_4_to_8() function is
4637deprecated.
4638
4639A number of macro definitions in support of runtime selection of
4640assembler code features (especially Intel MMX code support) were
4641added at libpng-1.2.0:
4642
4643    PNG_ASM_FLAG_MMX_SUPPORT_COMPILED
4644    PNG_ASM_FLAG_MMX_SUPPORT_IN_CPU
4645    PNG_ASM_FLAG_MMX_READ_COMBINE_ROW
4646    PNG_ASM_FLAG_MMX_READ_INTERLACE
4647    PNG_ASM_FLAG_MMX_READ_FILTER_SUB
4648    PNG_ASM_FLAG_MMX_READ_FILTER_UP
4649    PNG_ASM_FLAG_MMX_READ_FILTER_AVG
4650    PNG_ASM_FLAG_MMX_READ_FILTER_PAETH
4651    PNG_ASM_FLAGS_INITIALIZED
4652    PNG_MMX_READ_FLAGS
4653    PNG_MMX_FLAGS
4654    PNG_MMX_WRITE_FLAGS
4655    PNG_MMX_FLAGS
4656
4657We added the following functions in support of runtime
4658selection of assembler code features:
4659
4660    png_get_mmx_flagmask()
4661    png_set_mmx_thresholds()
4662    png_get_asm_flags()
4663    png_get_mmx_bitdepth_threshold()
4664    png_get_mmx_rowbytes_threshold()
4665    png_set_asm_flags()
4666
4667We replaced all of these functions with simple stubs in libpng-1.2.20,
4668when the Intel assembler code was removed due to a licensing issue.
4669
4670These macros are deprecated:
4671
4672    PNG_READ_TRANSFORMS_NOT_SUPPORTED
4673    PNG_PROGRESSIVE_READ_NOT_SUPPORTED
4674    PNG_NO_SEQUENTIAL_READ_SUPPORTED
4675    PNG_WRITE_TRANSFORMS_NOT_SUPPORTED
4676    PNG_READ_ANCILLARY_CHUNKS_NOT_SUPPORTED
4677    PNG_WRITE_ANCILLARY_CHUNKS_NOT_SUPPORTED
4678
4679They have been replaced, respectively, by:
4680
4681    PNG_NO_READ_TRANSFORMS
4682    PNG_NO_PROGRESSIVE_READ
4683    PNG_NO_SEQUENTIAL_READ
4684    PNG_NO_WRITE_TRANSFORMS
4685    PNG_NO_READ_ANCILLARY_CHUNKS
4686    PNG_NO_WRITE_ANCILLARY_CHUNKS
4687
4688PNG_MAX_UINT was replaced with PNG_UINT_31_MAX.  It has been
4689deprecated since libpng-1.0.16 and libpng-1.2.6.
4690
4691The function
4692    png_check_sig(sig, num)
4693was replaced with
4694    !png_sig_cmp(sig, 0, num)
4695It has been deprecated since libpng-0.90.
4696
4697The function
4698    png_set_gray_1_2_4_to_8()
4699which also expands tRNS to alpha was replaced with
4700    png_set_expand_gray_1_2_4_to_8()
4701which does not. It has been deprecated since libpng-1.0.18 and 1.2.9.
4702
4703X.  Changes to Libpng from version 1.0.x/1.2.x to 1.4.x
4704
4705Private libpng prototypes and macro definitions were moved from
4706png.h and pngconf.h into a new pngpriv.h header file.
4707
4708Functions png_set_benign_errors(), png_benign_error(), and
4709png_chunk_benign_error() were added.
4710
4711Support for setting the maximum amount of memory that the application
4712will allocate for reading chunks was added, as a security measure.
4713The functions png_set_chunk_cache_max() and png_get_chunk_cache_max()
4714were added to the library.
4715
4716We implemented support for I/O states by adding png_ptr member io_state
4717and functions png_get_io_chunk_name() and png_get_io_state() in pngget.c
4718
4719We added PNG_TRANSFORM_GRAY_TO_RGB to the available high-level
4720input transforms.
4721
4722Checking for and reporting of errors in the IHDR chunk is more thorough.
4723
4724Support for global arrays was removed, to improve thread safety.
4725
4726Some obsolete/deprecated macros and functions have been removed.
4727
4728Typecasted NULL definitions such as
4729   #define png_voidp_NULL            (png_voidp)NULL
4730were eliminated.  If you used these in your application, just use
4731NULL instead.
4732
4733The png_struct and info_struct members "trans" and "trans_values" were
4734changed to "trans_alpha" and "trans_color", respectively.
4735
4736The obsolete, unused pnggccrd.c and pngvcrd.c files and related makefiles
4737were removed.
4738
4739The PNG_1_0_X and PNG_1_2_X macros were eliminated.
4740
4741The PNG_LEGACY_SUPPORTED macro was eliminated.
4742
4743Many WIN32_WCE #ifdefs were removed.
4744
4745The functions png_read_init(info_ptr), png_write_init(info_ptr),
4746png_info_init(info_ptr), png_read_destroy(), and png_write_destroy()
4747have been removed.  They have been deprecated since libpng-0.95.
4748
4749The png_permit_empty_plte() was removed. It has been deprecated
4750since libpng-1.0.9.  Use png_permit_mng_features() instead.
4751
4752We removed the obsolete stub functions png_get_mmx_flagmask(),
4753png_set_mmx_thresholds(), png_get_asm_flags(),
4754png_get_mmx_bitdepth_threshold(), png_get_mmx_rowbytes_threshold(),
4755png_set_asm_flags(), and png_mmx_supported()
4756
4757We removed the obsolete png_check_sig(), png_memcpy_check(), and
4758png_memset_check() functions.  Instead use !png_sig_cmp(), memcpy(),
4759and memset(), respectively.
4760
4761The function png_set_gray_1_2_4_to_8() was removed. It has been
4762deprecated since libpng-1.0.18 and 1.2.9, when it was replaced with
4763png_set_expand_gray_1_2_4_to_8() because the former function also
4764expanded any tRNS chunk to an alpha channel.
4765
4766Macros for png_get_uint_16, png_get_uint_32, and png_get_int_32
4767were added and are used by default instead of the corresponding
4768functions. Unfortunately,
4769from libpng-1.4.0 until 1.4.4, the png_get_uint_16 macro (but not the
4770function) incorrectly returned a value of type png_uint_32.
4771
4772We changed the prototype for png_malloc() from
4773    png_malloc(png_structp png_ptr, png_uint_32 size)
4774to
4775    png_malloc(png_structp png_ptr, png_alloc_size_t size)
4776
4777This also applies to the prototype for the user replacement malloc_fn().
4778
4779The png_calloc() function was added and is used in place of
4780of "png_malloc(); memset();" except in the case in png_read_png()
4781where the array consists of pointers; in this case a "for" loop is used
4782after the png_malloc() to set the pointers to NULL, to give robust.
4783behavior in case the application runs out of memory part-way through
4784the process.
4785
4786We changed the prototypes of png_get_compression_buffer_size() and
4787png_set_compression_buffer_size() to work with png_size_t instead of
4788png_uint_32.
4789
4790Support for numbered error messages was removed by default, since we
4791never got around to actually numbering the error messages. The function
4792png_set_strip_error_numbers() was removed from the library by default.
4793
4794The png_zalloc() and png_zfree() functions are no longer exported.
4795The png_zalloc() function no longer zeroes out the memory that it
4796allocates.  Applications that called png_zalloc(png_ptr, number, size)
4797can call png_calloc(png_ptr, number*size) instead, and can call
4798png_free() instead of png_zfree().
4799
4800Support for dithering was disabled by default in libpng-1.4.0, because
4801it has not been well tested and doesn't actually "dither".
4802The code was not
4803removed, however, and could be enabled by building libpng with
4804PNG_READ_DITHER_SUPPORTED defined.  In libpng-1.4.2, this support
4805was re-enabled, but the function was renamed png_set_quantize() to
4806reflect more accurately what it actually does.  At the same time,
4807the PNG_DITHER_[RED,GREEN_BLUE]_BITS macros were also renamed to
4808PNG_QUANTIZE_[RED,GREEN,BLUE]_BITS, and PNG_READ_DITHER_SUPPORTED
4809was renamed to PNG_READ_QUANTIZE_SUPPORTED.
4810
4811We removed the trailing '.' from the warning and error messages.
4812
4813XI.  Changes to Libpng from version 1.4.x to 1.5.x
4814
4815From libpng-1.4.0 until 1.4.4, the png_get_uint_16 macro (but not the
4816function) incorrectly returned a value of type png_uint_32.
4817The incorrect macro was removed from libpng-1.4.5.
4818
4819Checking for invalid palette index on write was added at libpng
48201.5.10.  If a pixel contains an invalid (out-of-range) index libpng issues
4821a benign error.  This is enabled by default because this condition is an
4822error according to the PNG specification, Clause 11.3.2, but the error can
4823be ignored in each png_ptr with
4824
4825   png_set_check_for_invalid_index(png_ptr, allowed);
4826
4827      allowed  - one of
4828                 0: disable benign error (accept the
4829                    invalid data without warning).
4830                 1: enable benign error (treat the
4831                    invalid data as an error or a
4832                    warning).
4833
4834If the error is ignored, or if png_benign_error() treats it as a warning,
4835any invalid pixels are decoded as opaque black by the decoder and written
4836as-is by the encoder.
4837
4838Retrieving the maximum palette index found was added at libpng-1.5.15.
4839This statement must appear after png_read_png() or png_read_image() while
4840reading, and after png_write_png() or png_write_image() while writing.
4841
4842   int max_palette = png_get_palette_max(png_ptr, info_ptr);
4843
4844This will return the maximum palette index found in the image, or "-1" if
4845the palette was not checked, or "0" if no palette was found.  Note that this
4846does not account for any palette index used by ancillary chunks such as the
4847bKGD chunk; you must check those separately to determine the maximum
4848palette index actually used.
4849
4850There are no substantial API changes between the non-deprecated parts of
4851the 1.4.5 API and the 1.5.0 API; however, the ability to directly access
4852members of the main libpng control structures, png_struct and png_info,
4853deprecated in earlier versions of libpng, has been completely removed from
4854libpng 1.5, and new private "pngstruct.h", "pnginfo.h", and "pngdebug.h"
4855header files were created.
4856
4857We no longer include zlib.h in png.h.  The include statement has been moved
4858to pngstruct.h, where it is not accessible by applications. Applications that
4859need access to information in zlib.h will need to add the '#include "zlib.h"'
4860directive.  It does not matter whether this is placed prior to or after
4861the '"#include png.h"' directive.
4862
4863The png_sprintf(), png_strcpy(), and png_strncpy() macros are no longer used
4864and were removed.
4865
4866We moved the png_strlen(), png_memcpy(), png_memset(), and png_memcmp()
4867macros into a private header file (pngpriv.h) that is not accessible to
4868applications.
4869
4870In png_get_iCCP, the type of "profile" was changed from png_charpp
4871to png_bytepp, and in png_set_iCCP, from png_charp to png_const_bytep.
4872
4873There are changes of form in png.h, including new and changed macros to
4874declare parts of the API.  Some API functions with arguments that are
4875pointers to data not modified within the function have been corrected to
4876declare these arguments with PNG_CONST.
4877
4878Much of the internal use of C macros to control the library build has also
4879changed and some of this is visible in the exported header files, in
4880particular the use of macros to control data and API elements visible
4881during application compilation may require significant revision to
4882application code.  (It is extremely rare for an application to do this.)
4883
4884Any program that compiled against libpng 1.4 and did not use deprecated
4885features or access internal library structures should compile and work
4886against libpng 1.5, except for the change in the prototype for
4887png_get_iCCP() and png_set_iCCP() API functions mentioned above.
4888
4889libpng 1.5.0 adds PNG_ PASS macros to help in the reading and writing of
4890interlaced images.  The macros return the number of rows and columns in
4891each pass and information that can be used to de-interlace and (if
4892absolutely necessary) interlace an image.
4893
4894libpng 1.5.0 adds an API png_longjmp(png_ptr, value).  This API calls
4895the application-provided png_longjmp_ptr on the internal, but application
4896initialized, longjmp buffer.  It is provided as a convenience to avoid
4897the need to use the png_jmpbuf macro, which had the unnecessary side
4898effect of resetting the internal png_longjmp_ptr value.
4899
4900libpng 1.5.0 includes a complete fixed point API.  By default this is
4901present along with the corresponding floating point API.  In general the
4902fixed point API is faster and smaller than the floating point one because
4903the PNG file format used fixed point, not floating point.  This applies
4904even if the library uses floating point in internal calculations.  A new
4905macro, PNG_FLOATING_ARITHMETIC_SUPPORTED, reveals whether the library
4906uses floating point arithmetic (the default) or fixed point arithmetic
4907internally for performance critical calculations such as gamma correction.
4908In some cases, the gamma calculations may produce slightly different
4909results.  This has changed the results in png_rgb_to_gray and in alpha
4910composition (png_set_background for example). This applies even if the
4911original image was already linear (gamma == 1.0) and, therefore, it is
4912not necessary to linearize the image.  This is because libpng has *not*
4913been changed to optimize that case correctly, yet.
4914
4915Fixed point support for the sCAL chunk comes with an important caveat;
4916the sCAL specification uses a decimal encoding of floating point values
4917and the accuracy of PNG fixed point values is insufficient for
4918representation of these values. Consequently a "string" API
4919(png_get_sCAL_s and png_set_sCAL_s) is the only reliable way of reading
4920arbitrary sCAL chunks in the absence of either the floating point API or
4921internal floating point calculations.  Starting with libpng-1.5.0, both
4922of these functions are present when PNG_sCAL_SUPPORTED is defined.  Prior
4923to libpng-1.5.0, their presence also depended upon PNG_FIXED_POINT_SUPPORTED
4924being defined and PNG_FLOATING_POINT_SUPPORTED not being defined.
4925
4926Applications no longer need to include the optional distribution header
4927file pngusr.h or define the corresponding macros during application
4928build in order to see the correct variant of the libpng API.  From 1.5.0
4929application code can check for the corresponding _SUPPORTED macro:
4930
4931#ifdef PNG_INCH_CONVERSIONS_SUPPORTED
4932   /* code that uses the inch conversion APIs. */
4933#endif
4934
4935This macro will only be defined if the inch conversion functions have been
4936compiled into libpng.  The full set of macros, and whether or not support
4937has been compiled in, are available in the header file pnglibconf.h.
4938This header file is specific to the libpng build.  Notice that prior to
49391.5.0 the _SUPPORTED macros would always have the default definition unless
4940reset by pngusr.h or by explicit settings on the compiler command line.
4941These settings may produce compiler warnings or errors in 1.5.0 because
4942of macro redefinition.
4943
4944Applications can now choose whether to use these macros or to call the
4945corresponding function by defining PNG_USE_READ_MACROS or
4946PNG_NO_USE_READ_MACROS before including png.h.  Notice that this is
4947only supported from 1.5.0; defining PNG_NO_USE_READ_MACROS prior to 1.5.0
4948will lead to a link failure.
4949
4950Prior to libpng-1.5.4, the zlib compressor used the same set of parameters
4951when compressing the IDAT data and textual data such as zTXt and iCCP.
4952In libpng-1.5.4 we reinitialized the zlib stream for each type of data.
4953We added five png_set_text_*() functions for setting the parameters to
4954use with textual data.
4955
4956Prior to libpng-1.5.4, the PNG_READ_16_TO_8_ACCURATE_SCALE_SUPPORTED
4957option was off by default, and slightly inaccurate scaling occurred.
4958This option can no longer be turned off, and the choice of accurate
4959or inaccurate 16-to-8 scaling is by using the new png_set_scale_16_to_8()
4960API for accurate scaling or the old png_set_strip_16_to_8() API for simple
4961chopping.  In libpng-1.5.4, the PNG_READ_16_TO_8_ACCURATE_SCALE_SUPPORTED
4962macro became PNG_READ_SCALE_16_TO_8_SUPPORTED, and the PNG_READ_16_TO_8
4963macro became PNG_READ_STRIP_16_TO_8_SUPPORTED, to enable the two
4964png_set_*_16_to_8() functions separately.
4965
4966Prior to libpng-1.5.4, the png_set_user_limits() function could only be
4967used to reduce the width and height limits from the value of
4968PNG_USER_WIDTH_MAX and PNG_USER_HEIGHT_MAX, although this document said
4969that it could be used to override them.  Now this function will reduce or
4970increase the limits.
4971
4972Starting in libpng-1.5.22, default user limits were established. These
4973can be overridden by application calls to png_set_user_limits(),
4974png_set_user_chunk_cache_max(), and/or png_set_user_malloc_max().
4975The limits are now
4976                             max possible  default
4977   png_user_width_max        0x7fffffff    1,000,000
4978   png_user_height_max       0x7fffffff    1,000,000
4979   png_user_chunk_cache_max  0 (unlimited) 1000
4980   png_user_chunk_malloc_max 0 (unlimited) 8,000,000
4981
4982The png_set_option() function (and the "options" member of the png struct) was
4983added to libpng-1.5.15, with option PNG_ARM_NEON.
4984
4985The library now supports a complete fixed point implementation and can
4986thus be used on systems that have no floating point support or very
4987limited or slow support.  Previously gamma correction, an essential part
4988of complete PNG support, required reasonably fast floating point.
4989
4990As part of this the choice of internal implementation has been made
4991independent of the choice of fixed versus floating point APIs and all the
4992missing fixed point APIs have been implemented.
4993
4994The exact mechanism used to control attributes of API functions has
4995changed, as described in the INSTALL file.
4996
4997A new test program, pngvalid, is provided in addition to pngtest.
4998pngvalid validates the arithmetic accuracy of the gamma correction
4999calculations and includes a number of validations of the file format.
5000A subset of the full range of tests is run when "make check" is done
5001(in the 'configure' build.)  pngvalid also allows total allocated memory
5002usage to be evaluated and performs additional memory overwrite validation.
5003
5004Many changes to individual feature macros have been made. The following
5005are the changes most likely to be noticed by library builders who
5006configure libpng:
5007
50081) All feature macros now have consistent naming:
5009
5010#define PNG_NO_feature turns the feature off
5011#define PNG_feature_SUPPORTED turns the feature on
5012
5013pnglibconf.h contains one line for each feature macro which is either:
5014
5015#define PNG_feature_SUPPORTED
5016
5017if the feature is supported or:
5018
5019/*#undef PNG_feature_SUPPORTED*/
5020
5021if it is not.  Library code consistently checks for the 'SUPPORTED' macro.
5022It does not, and libpng applications should not, check for the 'NO' macro
5023which will not normally be defined even if the feature is not supported.
5024The 'NO' macros are only used internally for setting or not setting the
5025corresponding 'SUPPORTED' macros.
5026
5027Compatibility with the old names is provided as follows:
5028
5029PNG_INCH_CONVERSIONS turns on PNG_INCH_CONVERSIONS_SUPPORTED
5030
5031And the following definitions disable the corresponding feature:
5032
5033PNG_SETJMP_NOT_SUPPORTED disables SETJMP
5034PNG_READ_TRANSFORMS_NOT_SUPPORTED disables READ_TRANSFORMS
5035PNG_NO_READ_COMPOSITED_NODIV disables READ_COMPOSITE_NODIV
5036PNG_WRITE_TRANSFORMS_NOT_SUPPORTED disables WRITE_TRANSFORMS
5037PNG_READ_ANCILLARY_CHUNKS_NOT_SUPPORTED disables READ_ANCILLARY_CHUNKS
5038PNG_WRITE_ANCILLARY_CHUNKS_NOT_SUPPORTED disables WRITE_ANCILLARY_CHUNKS
5039
5040Library builders should remove use of the above, inconsistent, names.
5041
50422) Warning and error message formatting was previously conditional on
5043the STDIO feature. The library has been changed to use the
5044CONSOLE_IO feature instead. This means that if CONSOLE_IO is disabled
5045the library no longer uses the printf(3) functions, even though the
5046default read/write implementations use (FILE) style stdio.h functions.
5047
50483) Three feature macros now control the fixed/floating point decisions:
5049
5050PNG_FLOATING_POINT_SUPPORTED enables the floating point APIs
5051
5052PNG_FIXED_POINT_SUPPORTED enables the fixed point APIs; however, in
5053practice these are normally required internally anyway (because the PNG
5054file format is fixed point), therefore in most cases PNG_NO_FIXED_POINT
5055merely stops the function from being exported.
5056
5057PNG_FLOATING_ARITHMETIC_SUPPORTED chooses between the internal floating
5058point implementation or the fixed point one.  Typically the fixed point
5059implementation is larger and slower than the floating point implementation
5060on a system that supports floating point; however, it may be faster on a
5061system which lacks floating point hardware and therefore uses a software
5062emulation.
5063
50644) Added PNG_{READ,WRITE}_INT_FUNCTIONS_SUPPORTED.  This allows the
5065functions to read and write ints to be disabled independently of
5066PNG_USE_READ_MACROS, which allows libpng to be built with the functions
5067even though the default is to use the macros - this allows applications
5068to choose at app buildtime whether or not to use macros (previously
5069impossible because the functions weren't in the default build.)
5070
5071XII.  Changes to Libpng from version 1.5.x to 1.6.x
5072
5073A "simplified API" has been added (see documentation in png.h and a simple
5074example in contrib/examples/pngtopng.c).  The new publicly visible API
5075includes the following:
5076
5077   macros:
5078     PNG_FORMAT_*
5079     PNG_IMAGE_*
5080   structures:
5081     png_control
5082     png_image
5083   read functions
5084     png_image_begin_read_from_file()
5085     png_image_begin_read_from_stdio()
5086     png_image_begin_read_from_memory()
5087     png_image_finish_read()
5088     png_image_free()
5089   write functions
5090     png_image_write_to_file()
5091     png_image_write_to_memory()
5092     png_image_write_to_stdio()
5093
5094Starting with libpng-1.6.0, you can configure libpng to prefix all exported
5095symbols, using the PNG_PREFIX macro.
5096
5097We no longer include string.h in png.h.  The include statement has been moved
5098to pngpriv.h, where it is not accessible by applications.  Applications that
5099need access to information in string.h must add an '#include <string.h>'
5100directive.  It does not matter whether this is placed prior to or after
5101the '#include "png.h"' directive.
5102
5103The following API are now DEPRECATED:
5104   png_info_init_3()
5105   png_convert_to_rfc1123() which has been replaced
5106     with png_convert_to_rfc1123_buffer()
5107   png_malloc_default()
5108   png_free_default()
5109   png_reset_zstream()
5110
5111The following have been removed:
5112   png_get_io_chunk_name(), which has been replaced
5113     with png_get_io_chunk_type().  The new
5114     function returns a 32-bit integer instead of
5115     a string.
5116   The png_sizeof(), png_strlen(), png_memcpy(), png_memcmp(), and
5117     png_memset() macros are no longer used in the libpng sources and
5118     have been removed.  These had already been made invisible to applications
5119     (i.e., defined in the private pngpriv.h header file) since libpng-1.5.0.
5120
5121The signatures of many exported functions were changed, such that
5122   png_structp became png_structrp or png_const_structrp
5123   png_infop became png_inforp or png_const_inforp
5124where "rp" indicates a "restricted pointer".
5125
5126Dropped support for 16-bit platforms. The support for FAR/far types has
5127been eliminated and the definition of png_alloc_size_t is now controlled
5128by a flag so that 'small size_t' systems can select it if necessary.
5129
5130Error detection in some chunks has improved; in particular the iCCP chunk
5131reader now does pretty complete validation of the basic format.  Some bad
5132profiles that were previously accepted are now accepted with a warning or
5133rejected, depending upon the png_set_benign_errors() setting, in particular
5134the very old broken Microsoft/HP 3144-byte sRGB profile.  Starting with
5135libpng-1.6.11, recognizing and checking sRGB profiles can be avoided by
5136means of
5137
5138    #if defined(PNG_SKIP_sRGB_CHECK_PROFILE) && \
5139        defined(PNG_SET_OPTION_SUPPORTED)
5140       png_set_option(png_ptr, PNG_SKIP_sRGB_CHECK_PROFILE,
5141           PNG_OPTION_ON);
5142    #endif
5143
5144It's not a good idea to do this if you are using the "simplified API",
5145which needs to be able to recognize sRGB profiles conveyed via the iCCP
5146chunk.
5147
5148The PNG spec requirement that only grayscale profiles may appear in images
5149with color type 0 or 4 and that even if the image only contains gray pixels,
5150only RGB profiles may appear in images with color type 2, 3, or 6, is now
5151enforced.  The sRGB chunk is allowed to appear in images with any color type
5152and is interpreted by libpng to convey a one-tracer-curve gray profile or a
5153three-tracer-curve RGB profile as appropriate.
5154
5155Libpng 1.5.x erroneously used /MD for Debug DLL builds; if you used the debug
5156builds in your app and you changed your app to use /MD you will need to
5157change it back to /MDd for libpng 1.6.x.
5158
5159Prior to libpng-1.6.0 a warning would be issued if the iTXt chunk contained
5160an empty language field or an empty translated keyword.  Both of these
5161are allowed by the PNG specification, so these warnings are no longer issued.
5162
5163The library now issues an error if the application attempts to set a
5164transform after it calls png_read_update_info() or if it attempts to call
5165both png_read_update_info() and png_start_read_image() or to call either
5166of them more than once.
5167
5168The default condition for benign_errors is now to treat benign errors as
5169warnings while reading and as errors while writing.
5170
5171The library now issues a warning if both background processing and RGB to
5172gray are used when gamma correction happens. As with previous versions of
5173the library the results are numerically very incorrect in this case.
5174
5175There are some minor arithmetic changes in some transforms such as
5176png_set_background(), that might be detected by certain regression tests.
5177
5178Unknown chunk handling has been improved internally, without any API change.
5179This adds more correct option control of the unknown handling, corrects
5180a pre-existing bug where the per-chunk 'keep' setting is ignored, and makes
5181it possible to skip IDAT chunks in the sequential reader.
5182
5183The machine-generated configure files are no longer included in branches
5184libpng16 and later of the GIT repository.  They continue to be included
5185in the tarball releases, however.
5186
5187Libpng-1.6.0 through 1.6.2 used the CMF bytes at the beginning of the IDAT
5188stream to set the size of the sliding window for reading instead of using the
5189default 32-kbyte sliding window size.  It was discovered that there are
5190hundreds of PNG files in the wild that have incorrect CMF bytes that caused
5191zlib to issue the "invalid distance too far back" error and reject the file.
5192Libpng-1.6.3 and later calculate their own safe CMF from the image dimensions,
5193provide a way to revert to the libpng-1.5.x behavior (ignoring the CMF bytes
5194and using a 32-kbyte sliding window), by using
5195
5196    png_set_option(png_ptr, PNG_MAXIMUM_INFLATE_WINDOW,
5197        PNG_OPTION_ON);
5198
5199and provide a tool (contrib/tools/pngfix) for rewriting a PNG file while
5200optimizing the CMF bytes in its IDAT chunk correctly.
5201
5202Libpng-1.6.0 and libpng-1.6.1 wrote uncompressed iTXt chunks with the wrong
5203length, which resulted in PNG files that cannot be read beyond the bad iTXt
5204chunk.  This error was fixed in libpng-1.6.3, and a tool (called
5205contrib/tools/png-fix-itxt) has been added to the libpng distribution.
5206
5207Starting with libpng-1.6.17, the PNG_SAFE_LIMITS macro was eliminated
5208and safe limits are used by default (users who need larger limits
5209can still override them at compile time or run time, as described above).
5210
5211The new limits are
5212                                default   spec limit
5213   png_user_width_max         1,000,000  2,147,483,647
5214   png_user_height_max        1,000,000  2,147,483,647
5215   png_user_chunk_cache_max         128  unlimited
5216   png_user_chunk_malloc_max  8,000,000  unlimited
5217
5218Starting with libpng-1.6.18, a PNG_RELEASE_BUILD macro was added, which allows
5219library builders to control compilation for an installed system (a release build).
5220It can be set for testing debug or beta builds to ensure that they will compile
5221when the build type is switched to RC or STABLE. In essence this overrides the
5222PNG_LIBPNG_BUILD_BASE_TYPE definition which is not directly user controllable.
5223
5224Starting with libpng-1.6.19, attempting to set an over-length PLTE chunk
5225is an error. Previously this requirement of the PNG specification was not
5226enforced, and the palette was always limited to 256 entries. An over-length
5227PLTE chunk found in an input PNG is silently truncated.
5228
5229Starting with libpng-1.6.31, the eXIf chunk is supported. Libpng does not
5230attempt to decode the Exif profile; it simply returns a byte array
5231containing the profile to the calling application which must do its own
5232decoding.
5233
5234XIII.  Detecting libpng
5235
5236The png_get_io_ptr() function has been present since libpng-0.88, has never
5237changed, and is unaffected by conditional compilation macros.  It is the
5238best choice for use in configure scripts for detecting the presence of any
5239libpng version since 0.88.  In an autoconf "configure.in" you could use
5240
5241    AC_CHECK_LIB(png, png_get_io_ptr, ...
5242
5243XV. Source code repository
5244
5245Since about February 2009, version 1.2.34, libpng has been under "git" source
5246control.  The git repository was built from old libpng-x.y.z.tar.gz files
5247going back to version 0.70.  You can access the git repository (read only)
5248at
5249
5250    https://github.com/glennrp/libpng or
5251    https://git.code.sf.net/p/libpng/code.git
5252
5253or you can browse it with a web browser at
5254
5255    https://github.com/glennrp/libpng or
5256    https://sourceforge.net/p/libpng/code/ci/libpng16/tree/
5257
5258Patches can be sent to glennrp at users.sourceforge.net or to
5259png-mng-implement at lists.sourceforge.net or you can upload them to
5260the libpng bug tracker at
5261
5262    https://libpng.sourceforge.io/
5263
5264or as a "pull request" to
5265
5266    https://github.com/glennrp/libpng/pulls
5267
5268We also accept patches built from the tar or zip distributions, and
5269simple verbal discriptions of bug fixes, reported either to the
5270SourceForge bug tracker, to the png-mng-implement at lists.sf.net
5271mailing list, as github issues, or directly to glennrp.
5272
5273XV. Coding style
5274
5275Our coding style is similar to the "Allman" style
5276(See https://en.wikipedia.org/wiki/Indent_style#Allman_style), with curly
5277braces on separate lines:
5278
5279    if (condition)
5280    {
5281       action;
5282    }
5283
5284    else if (another condition)
5285    {
5286       another action;
5287    }
5288
5289The braces can be omitted from simple one-line actions:
5290
5291    if (condition)
5292       return (0);
5293
5294We use 3-space indentation, except for continued statements which
5295are usually indented the same as the first line of the statement
5296plus four more spaces.
5297
5298For macro definitions we use 2-space indentation, always leaving the "#"
5299in the first column.
5300
5301    #ifndef PNG_NO_FEATURE
5302    #  ifndef PNG_FEATURE_SUPPORTED
5303    #    define PNG_FEATURE_SUPPORTED
5304    #  endif
5305    #endif
5306
5307Comments appear with the leading "/*" at the same indentation as
5308the statement that follows the comment:
5309
5310    /* Single-line comment */
5311    statement;
5312
5313    /* This is a multiple-line
5314     * comment.
5315     */
5316    statement;
5317
5318Very short comments can be placed after the end of the statement
5319to which they pertain:
5320
5321    statement;    /* comment */
5322
5323We don't use C++ style ("//") comments. We have, however,
5324used them in the past in some now-abandoned MMX assembler
5325code.
5326
5327Functions and their curly braces are not indented, and
5328exported functions are marked with PNGAPI:
5329
5330 /* This is a public function that is visible to
5331  * application programmers. It does thus-and-so.
5332  */
5333 void PNGAPI
5334 png_exported_function(png_ptr, png_info, foo)
5335 {
5336    body;
5337 }
5338
5339The return type and decorations are placed on a separate line
5340ahead of the function name, as illustrated above.
5341
5342The prototypes for all exported functions appear in png.h,
5343above the comment that says
5344
5345    /* Maintainer: Put new public prototypes here ... */
5346
5347We mark all non-exported functions with "/* PRIVATE */"":
5348
5349 void /* PRIVATE */
5350 png_non_exported_function(png_ptr, png_info, foo)
5351 {
5352    body;
5353 }
5354
5355The prototypes for non-exported functions (except for those in
5356pngtest) appear in pngpriv.h above the comment that says
5357
5358  /* Maintainer: Put new private prototypes here ^ */
5359
5360To avoid polluting the global namespace, the names of all exported
5361functions and variables begin with "png_", and all publicly visible C
5362preprocessor macros begin with "PNG".  We request that applications that
5363use libpng *not* begin any of their own symbols with either of these strings.
5364
5365We put a space after the "sizeof" operator and we omit the
5366optional parentheses around its argument when the argument
5367is an expression, not a type name, and we always enclose the
5368sizeof operator, with its argument, in parentheses:
5369
5370  (sizeof (png_uint_32))
5371  (sizeof array)
5372
5373Prior to libpng-1.6.0 we used a "png_sizeof()" macro, formatted as
5374though it were a function.
5375
5376Control keywords if, for, while, and switch are always followed by a space
5377to distinguish them from function calls, which have no trailing space.
5378
5379We put a space after each comma and after each semicolon
5380in "for" statements, and we put spaces before and after each
5381C binary operator and after "for" or "while", and before
5382"?".  We don't put a space between a typecast and the expression
5383being cast, nor do we put one between a function name and the
5384left parenthesis that follows it:
5385
5386    for (i = 2; i > 0; --i)
5387       y[i] = a(x) + (int)b;
5388
5389We prefer #ifdef and #ifndef to #if defined() and #if !defined()
5390when there is only one macro being tested.  We always use parentheses
5391with "defined".
5392
5393We express integer constants that are used as bit masks in hex format,
5394with an even number of lower-case hex digits, and to make them unsigned
5395(e.g., 0x00U, 0xffU, 0x0100U) and long if they are greater than 0x7fff
5396(e.g., 0xffffUL).
5397
5398We prefer to use underscores rather than camelCase in names, except
5399for a few type names that we inherit from zlib.h.
5400
5401We prefer "if (something != 0)" and "if (something == 0)" over
5402"if (something)" and if "(!something)", respectively, and for pointers
5403we prefer "if (some_pointer != NULL)" or "if (some_pointer == NULL)".
5404
5405We do not use the TAB character for indentation in the C sources.
5406
5407Lines do not exceed 80 characters.
5408
5409Other rules can be inferred by inspecting the libpng source.
5410
5411XVI. Y2K Compliance in libpng
5412
5413Since the PNG Development group is an ad-hoc body, we can't make
5414an official declaration.
5415
5416This is your unofficial assurance that libpng from version 0.71 and
5417upward through 1.6.34 are Y2K compliant.  It is my belief that earlier
5418versions were also Y2K compliant.
5419
5420Libpng only has two year fields.  One is a 2-byte unsigned integer
5421that will hold years up to 65535.  The other, which is deprecated,
5422holds the date in text format, and will hold years up to 9999.
5423
5424The integer is
5425    "png_uint_16 year" in png_time_struct.
5426
5427The string is
5428    "char time_buffer[29]" in png_struct.  This is no longer used
5429in libpng-1.6.x and will be removed from libpng-1.7.0.
5430
5431There are seven time-related functions:
5432
5433    png_convert_to_rfc_1123_buffer() in png.c
5434      (formerly png_convert_to_rfc_1152() in error, and
5435      also formerly png_convert_to_rfc_1123())
5436    png_convert_from_struct_tm() in pngwrite.c, called
5437      in pngwrite.c
5438    png_convert_from_time_t() in pngwrite.c
5439    png_get_tIME() in pngget.c
5440    png_handle_tIME() in pngrutil.c, called in pngread.c
5441    png_set_tIME() in pngset.c
5442    png_write_tIME() in pngwutil.c, called in pngwrite.c
5443
5444All appear to handle dates properly in a Y2K environment.  The
5445png_convert_from_time_t() function calls gmtime() to convert from system
5446clock time, which returns (year - 1900), which we properly convert to
5447the full 4-digit year.  There is a possibility that applications using
5448libpng are not passing 4-digit years into the png_convert_to_rfc_1123()
5449function, or that they are incorrectly passing only a 2-digit year
5450instead of "year - 1900" into the png_convert_from_struct_tm() function,
5451but this is not under our control.  The libpng documentation has always
5452stated that it works with 4-digit years, and the APIs have been
5453documented as such.
5454
5455The tIME chunk itself is also Y2K compliant.  It uses a 2-byte unsigned
5456integer to hold the year, and can hold years as large as 65535.
5457
5458zlib, upon which libpng depends, is also Y2K compliant.  It contains
5459no date-related code.
5460
5461
5462   Glenn Randers-Pehrson
5463   libpng maintainer
5464   PNG Development Group