Change log:

Version 1.7.73 (built with libpng-1.6.10 and zlib-1.2.8)
  Restored calls to png_set_crc_action() which had been removed from
    version 1.7.72 for some testing and inadvertently not restored. 
  Changed "fix" internal variable name to "salvage" (still set with "-fix")
  Added code to fix/salvage PNG with "bad adaptive filter value" error.
  Avoid calculating CRC during compression trials except for the last trial,
    when the output is actually written.
  Fixed a bug with reducing 16-bit images to 8-bit using "-reduce" option.
    
Version 1.7.72 (built with libpng-1.6.10 and zlib-1.2.8)

Version 1.7.71 (built with libpng-1.6.9 and zlib-1.2.8)
  Built the Windows binaries using -DTOO_FAR=32767; neglected to do this
    in versions 1.7.42 through 1.7.70, which caused the Windows binaries
    to produce different (usually a few bytes larger) results than Linux.
    Thanks to F. Kayser for reporting the discrepancy.

Version 1.7.70 (built with libpng-1.6.8 and zlib-1.2.8)

Version 1.7.69 (built with libpng-1.6.6 and zlib-1.2.8)
  Updated libpng to version 1.6.6.

Version 1.7.68 (built with libpng-1.6.4 and zlib-1.2.8)
  Check for NULL return from malloc().
  Undefine CLOCKS_PER_SECOND "1000" found in some version of MinGW.
  Replaced most "atoi(argv[++i])" with "pngcrush_get_long" which does
    "BUMP_I; strtol(argv[i],ptr,10)" and added pngcrush_check_long macro
    to detect malformed or missing parameters (debian bug 716149).
  Added global_things_have_changed=1 when reading -bkgd.
  The "-bit_depth N" option did not work reliably and has been removed.

Version 1.7.67 (built with libpng-1.5.17 and zlib-1.2.8)
  Fixed handling of "-text" and "-ztext" options for text input. They had been
    reduced to "-t" and "-z" with an incorrect argument (3 instead of 2) in
    version 1.7.62. Bug report and patch from Tsukasa Oi.

Version 1.7.66 (built with libpng-1.5.17 and zlib-1.2.8)
  Revised pngcrush_examine_pixels_fn() to fix some incorrect reductions.

Version 1.7.65 (built with libpng-1.5.17 and zlib-1.2.8)
  Do not allow any colortype or depth reductions if acTL is present.
  Added warnings to explain why any requested reductions were not allowed.

Version 1.7.64 (built with libpng-1.5.17 and zlib-1.2.8)

Version 1.7.63 (built with libpng-1.5.16 and zlib-1.2.8)
  Add "int dowildcard=-1;" in an attempt to get wildcard arguments working
    in the cross-compiled MinGW executables.

Version 1.7.62 (built with libpng-1.5.16 and zlib-1.2.8)
  Remove old filename before renaming, when using the "-ow" option on
    any Windows platform, not just CYGWIN (see log entry for pngcrush-1.7.43).
  Reverted error with handling single-character options like "-v", introduced
    in 1.7.61.

Version 1.7.61 (built with libpng-1.5.16 and zlib-1.2.8)
  Check sBIT chunk data to see if reduction to gray or to 8-bit is permitted,
    i.e., the RGB sBIT values are equal to each other or the sBIT values are
    not greater than 8, respectively.
  Do not try to make_opaque if the tRNS chunk is found.
  Added warning when ignoring an invalid commandline option.
  Improved brute_force handling with specified level, filter, or strategy.

Version 1.7.60 (built with libpng-1.5.16 and zlib-1.2.8)
  Revise -reduce so reducing from color-type 6 to grayscale works.
  Issue a warning if reducing bit depth or color type would violate various
    chunk dependencies, and do not perform the action:
    Do not reduce to grayscale if a color bKGD chunk, sBIT or iCCP chunk
       is present.
    Do not reduce bit depth if bKGD or sBIT chunk is present.
    Do not reduce palette length if the hIST chunk is present.
  Set "found_iCCP" flag to zero to avoid re-reading a bad iCCP chunk.

Version 1.7.59 (built with libpng-1.5.16 and zlib-1.2.8)
  Show the acTL chunk in the chunk list in verbose output.
  Fixed several bugs reported by pornel at users.sf.net:
    Do not call png_set_benign_errors when PNG_BENIGN_ERRORS_SUPPORTED
      is not defined
    Renamed PNG_UNUSED() macro PNGCRUSH_UNUSED().
    Moved a closing bracket inside the PNGCRUSH_LOCO block.
    Moved the declaration of "new_mng" outside a PNGCRUSH_LOCO block.
    Put reference to "input_format" inside a PNGCRUSH_LOCO block.
    Moved declarations of mng_out and mngname inside a PNGCRUSH_LOGO block.

Version 1.7.58 (built with libpng-1.5.15 and zlib-1.2.7-1)
  Do not enable reduce_palette by default for "-reduce", "-new", or "-old".
    It still is failing for some files.

Version 1.7.57 (built with libpng-1.5.15 and zlib-1.2.7-1)
  Added "-new" option that turns on "-reduce" and "-force" which will be
    the default settings for version 1.8.0 and beyond.
  Added "-old" option that turns off "-reduce" and "-force" which are the
    current default settings.
  Updated copyright year for zlib-1.2.7-1.
  Reverted to libpng-1.5.15 to be able to read old PNG files with TOO FAR
    errors.  This will of course only work with the embedded libpng.

Version 1.7.56 (built with libpng-1.6.1 and zlib-1.2.7-1)
  Only use pngcrush_debug_malloc() and pngcrush_debug_free() if the result
    is going to be shown.
  Added PNG_PASS_ROWS, PNG_UNUSED, and other macro definitions, when building
    with libpng-1.4.x and older libpng versions.
  Multiplied rowbytes by 8/bit_depth when using the system library because
    we do not call png_read_transform_info(). This prevents a crash when
    reading sub-8-bit input files.

Version 1.7.55 (built with libpng-1.6.1 and zlib-1.2.7-1)

Version 1.7.54 (built with libpng-1.6.1rc01 and zlib-1.2.7-1)

Version 1.7.53 (built with libpng-1.6.1rc01 and zlib-1.2.7)
  Removed plte_len stuff from the "To do" list because it is done.
  Shorten the indexed-PNG tRNS chunk length if it has more entries than the PLTE chunk.

Version 1.7.52 (built with libpng-1.6.1beta06 and zlib-1.2.7)
  Added license info for cexcept.h, libpng, and zlib.
  Added consideration of "zopfli" compression to the "To do" list.
  Fixed a typo that caused a cHRM chunk to be "found" if an iCCP chunk
    were present.
  Reset best_byte_count before trial loop.
  Revise global png_set_keep_unknown_chunks() calls to avoid a libpng16
    warning.
  Reset "intent" to "specified_intent" before trial loop.
  Reset "plte_len" to "specified_plte_len" before trial loop.
  Initialize length of each trial to 0x7fffffff so any untried method
    is not the "best method".
 
Version 1.7.51 (built with libpng-1.6.0 and zlib-1.2.7)
  Added "-noreduce" option, in preparation for "-reduce" becoming the
    default behaviour in version 1.8.0.  This turns off lossless bit depth,
    color type and palette reduction, and opaque alpha channel removal.
  Zero out the high byte of transparent color for color-type 0 and 2,
    when reducing from 16 bits to 8.
  Undefined a bunch of stuff in pngcrush.h that we do not use, saves about
    100 kbytes of executable file size in addition to about 50k saved by
    undefining the simplified API.
  Fixed double-underscore typo in an #ifdef in png.c
  If "-reduce" is on and the background index is larger than the reduced
    palette_length+1, reduce it to the palette_length+1.
  Increased required_window_size if necessary to account for slightly larger
    size of interlaced files due to additional filter bytes and padding.

Version 1.7.50 (built with libpng-1.6.0 and zlib-1.2.7)
  Removed completed items from the "To do" list.
  Ignore the argument of the "plte_len" argument and just set the
    "reduce_palette" flag.

Version 1.7.49 (built with libpng-1.5.14 and zlib-1.2.7)
  Use png_set_benign_errors() to allow certain errors in the input file
    to be handled as warnings.
  Skip PNG_ABORT redefinition when using libpng-1.4.0 and later.
  Implemented "-reduce" option to identify and reduce all-gray images,
    all-opaque images, unused PLTE entries, and 16-bit images that can be
    reduced losslessly to 8-bit.

Version 1.7.48 (built with libpng-1.5.14 and zlib-1.2.7)
  Reserved method==0 for examining the pixels during trial 0, if necessary.
    Changed blacken_fn() to separate pngcrush_examine_pixels_fn() and
    pngcrush_transform_pixels_fn() callback functions.  The "examine"
    function is only done during trial 0; it sets a flag whether
    any fully transparent pixels were found, and pngcrush only runs
    pngcrush_transform_pixels_fn() if necessary.
  This is in preparation for future versions, which will examine other
    conditions such as if the image is opaque or gray or can be losslessly
    reduced in bit depth, set flags in trial 0 and accomplish the
    transformations in the remaining trials (see the To do list starting
    about line 200 in the pngcrush.c source).
  Removed "PNGCRUSH_COUNT_COLORS" blocks again.

Version 1.7.47 (built with libpng-1.5.13 and zlib-1.2.7)
  Do not do the heuristic trials of the first 10 methods when -brute is
    specified, because it did not save time as I hoped.
  Fixed a mistake in 1.7.45 and 1.7.46 that caused the output file to
    not be written.

Version 1.7.46 (built with libpng-1.5.13 and zlib-1.2.7)
  Moved the new level 0 methods to the end of the trial list (methods 137-148)

Version 1.7.45 (built with libpng-1.5.13 and zlib-1.2.7)
  Added method 0 (uncompressed). "-m 0" now simply turns on method 0.
  Added "-try10" option that has the same effect that "-m 0" previously did,
    namely to try only the first ten methods.
  Inserted new methods 17 through 21 with zlib level 0.
  Do the heuristic trials of the first 10 methods when -brute is specified,
    to get quickly to a small solution, so we can bail out of most of the
    remaining trials early. Previously these 10 methods were skipped during
    a -brute run.
  Removed the "-reduce" line from the help screen when PNGCRUSH_COUNT_COLORS
    is disabled.

Version 1.7.44 (built with libpng-1.5.14 and zlib-1.2.7)

Version 1.7.43 (built with libpng-1.5.13 and zlib-1.2.7)
  Added "remove(inname)" before "rename(outname, inname)" when using the "-ow"
    option on CYGWIN/MinGW because "rename()" does not work if the target file
    exists.
  Use the bundled "zlib.h" when PNGCRUSH_H is defined, otherwise use the
    system .

Version 1.7.42 (built with libpng-1.5.13 and zlib-1.2.7)
  Use malloc() and free() instead of png_malloc_default() and
    png_free_default().  This will be required to run with libpng-1.7.x.
  Revised the PNG_ABORT definition in pngcrush.h to work with libpng-1.7.x.
  Revised zutil.h to avoid redefining ptrdiff_t on MinGW/CYGWIN platforms.

Version 1.7.41 (built with libpng-1.5.13 and zlib-1.2.7)
  Reverted to version 1.7.38.  Versions 1.7.39 and 1.7.40 failed to
    open an output file.

Version 1.7.40 (built with libpng-1.5.13 and zlib-1.2.7)
  Revised the "To do" list.

Version 1.7.39 (built with libpng-1.5.13 and zlib-1.2.7)
  Removed "PNGCRUSH_COUNT_COLORS" blocks which I no longer intend to
    implement because that feature is already available in ImageMagick.  Kept
    "reduce_to_gray" and "it_is_opaque" flags which I do hope to implement
    soon.
  Changed NULL to pngcrush_default_read_data in png_set_read_fn() calls, to fix
    an insignificant error introduced in pngcrush-1.7.14, that caused most
    reads to not go through the alternate read function.  Also always set this
    function, instead of depending on STDIO_SUPPORTED.

Version 1.7.38 (built with libpng-1.5.13 and zlib-1.2.7)
  Bail out of a trial if byte count exceeds best byte count so far.  This
    avoids wasting CPU time on trial compressions of trials that exceed the
    best compression found so far.
  Added -bail and -nobail options.  Use -nobail to get a complete report
    of filesizes; otherwise the report just says ">N" for any trial
    that exceeds size N where N is the best size achieved so far.
  Added -blacken option, to enable changing the color samples of any
    fully-transparent pixels to zero in PNG files with color-type 4 or 6,
    potentially improving their compressibility. Note that this is an
    irreversible lossy change: the underlying colors of all fully transparent
    pixels are lost, if they were not already black.

Version 1.7.37 (built with libpng-1.5.12 and zlib-1.2.7)
  Reverted pngcrush.c back to 1.7.35 and fixed the bug with PLTE handling.

Version 1.7.36 (built with libpng-1.5.12 and zlib-1.2.7)
  Reverted pngcrush.c to version 1.7.34 because pngcrush is failing with
    some paletted PNGs.
  Separated CFLAGS and CPPFLAGS in the makefile (with "-I" and "-DZ_SOLO"
    in CPPFLAGS)

Version 1.7.35 (built with libpng-1.5.12 and zlib-1.2.7)
  Removed FOPEN of fpout except for the last trial.  The open files caused
    "pngcrush -brute -e _ext.png *.png" to fail on the 10th file (about the
    1024th compression trial) due to being unable to open the output file.

Version 1.7.34 (built with libpng-1.5.12 and zlib-1.2.7)
  Compute and report sum of critical chunk lengths IHDR, PLTE, IDAT, and IEND,
    plus the 8-byte PNG signature instead of just the total IDAT data length.
    Simplify finding the lengths from the trial compressions, by replacing
    the write function with one that simply counts the bytes that would have
    been written to a trial PNG, instead of actually writing a PNG, reading it
    back, and counting the IDAT bytes.
  Removed comments about the system library having to be libpng14 or earlier.
    This restriction was fixed in version 1.7.20.

Version 1.7.33  (built with libpng-1.5.12 and zlib-1.2.7)
  Ignore all ancillary chunks except during the final trial.  This can be
    significantly faster when large ancillary chunks such as iCCP and zTXt
    are present.

Version 1.7.32  (built with libpng-1.5.12 and zlib-1.2.7)
  Fixed bug introduced in 1.7.30: Do not call png_set_check_for_invalid_index()
    when nosave != 0 (otherwise pngcrush crashes with the "-n" option).

Version 1.7.31  (built with libpng-1.5.11 and zlib-1.2.7)
  Dropped *.tar.bz2 from distribution.
  Added a comma that was missing from one of the "usage" strings (error
    introduced in version 1.7.29).

Version 1.7.30  (built with libpng-1.5.11 and zlib-1.2.7)
  Only run the new (in libpng-1.5.10) test of palette indexes during the
    first trial.

Version 1.7.29  (built with libpng-1.5.10 and zlib-1.2.7)
  Set "things_have_changed" flag when adding text chunks, so the "-force"
    option is no longer necessary when adding text to an already-compressed
    file.
  Direct usage message and error messages to stderr instead of stdout. If
    anyone is still using DOS they may have to change the "if 0" at line
    990 to "if 1".  If you need to have the messages on standard output
    as in the past, use 2>&1 to redirect them.
  Added "pngcrush -n -v files.png" to the usage message.

Version 1.7.28  (built with libpng-1.5.10 and zlib-1.2.7)
  Write proper copyright year for zlib, depending upon ZLIB_VERNUM

Version 1.7.27  (built with libpng-1.5.10 and zlib-1.2.6)
  Increased row_buf malloc to row_bytes+64 instead of row_bytes+16, to
    match the size of big_row_buf in pngrutil.c (it is 48 in libpng14, 15, 16,
    and 64 in libpng10, 12.  Otherwise there is a double-free crash when the
    row_buf is destroyed.

Version 1.7.26  (built with libpng-1.5.10 and zlib-1.2.6)
  Increased the text_text buffer from 2048 to 10*2048 (Ralph Giles), and
    changed an incorrect test for keyword length "< 180" to "< 80".  The
    text_text buffer was inadvertently reduced from 20480 to 2048 in
    pngcrush-1.7.9.
  Added -DZ_SOLO to CFLAGS, needed to compile zlib-1.2.6.
  Changed user limits to width and height max 500000, malloc max 2MB,
    cache max 500.
  Added -nolimits option which sets the user limits to the default
    unlimited values.

Version 1.7.25  (built with libpng-1.5.9 and zlib-1.2.5)

Version 1.7.24  (built with libpng-1.5.7 and zlib-1.2.5)
  Do not append a slash to the directory name if it already has one.

Version 1.7.23  (built with libpng-1.5.7 and zlib-1.2.5)
  Ignore any attempt to use "-ow" with the "-d" or "-e" options, with warning.
  Include zlib.h if ZLIB_H is not defined (instead of checking the libpng
    version; see entry below for pngcrush-1.7.14), and include string.h
    if _STRING_H_ is not defined (because libpng-1.6 does not include string.h)
  Define SLASH = backslash on Windows platforms so the "-d" option will work..

Version 1.7.22  (built with libpng-1.5.6 and zlib-1.2.5)
  Added "-ow" (overwrite) option.  The input file is overwritten and the
    output file is just used temporarily and removed after it is copied
    over the input file..  If you do not specify an output file, "pngout.png"
    is used as the temporary file. Caution: the temporary file must be on
    the same filesystem as the input file.  Contributed by a group of students
    of the University of Paris who were taking the "Understanding of Programs"
    course and wished to gain familiarity with an open-source program.

Version 1.7.21  (built with libpng-1.5.6 and zlib-1.2.5)
  Defined TOO_FAR=32767 in Makefile (instead of in pngcrush.h)

Version 1.7.20  (built with libpng-1.5.5 and zlib-1.2.5)
  Removed the call to png_read_transform_info() when the system libpng
    is being used, so it can be built with a system libpng.

Version 1.7.19  (built with libpng-1.5.5 and zlib-1.2.5)
  pngcrush-1.7.18 failed to read interlaced PNGs.  Reverted the change
    from calling png_read_transform_info() to png_read_update_info().
    Since png_read_transform_info() is not exported we again cannot build
    with the system libpng15.

Version 1.7.18  (built with libpng-1.5.5 and zlib-1.2.5)
  This version will work with either a "system" libpng14 or libpng15, or with
    the embedded libpng15.  The deprecated usage of libpng png_struct members
    and unexported functions has been removed.
  Fixing "too far back" errors does not work with libpng15.
  Revised the format of the time report (all on one line so you can get
    a nice compact report by piping the output to "grep coding").

Version 1.7.17  (built with libpng-1.5.5beta08 and zlib-1.2.5)
  Changed "#if !defined(PNG_NO_STDIO)" to "#ifdef PNG_STDIO_SUPPORTED"
    as recommended in the libpng documentation.
  Added PNG_UINT_32_NAME macro and used it to simplify chunk_type integer
    definitions.

Version 1.7.16  (built with libpng-1.5.4 and zlib-1.2.5)
  Only report best method==0 if pngcrush cannot match the input filesize.
    Otherwise, if there is no improvement, report the first matching method.

Version 1.7.15  (built with libpng-1.5.2rc02 and zlib-1.2.5)
  Force bit_depth to 1, 2, or 4 when -plte_len is <=2, <=4, or <=16 and
    the -bit_depth option is not present, to avoid writing invalid palette
    indexes.

Version 1.7.14  (built with libpng-1.5.1beta08 and zlib-1.2.5)
  Removed WIN32_WCE support (libpng has dropped it already)
  Include zlib.h and define png_memcpy, etc., and revise the
    png_get_iCCP() and png_set_iCCP() calls to be able to build
    with bundled libpng-1.5.x.  Pngcrush cannot be built yet with
    a system libpng-1.5.x.
  Dropped most of pngcrush.h, that eliminates various parts of libpng.

Version 1.7.13  (built with libpng-1.4.5 and zlib-1.2.5)

Version 1.7.12  (built with libpng-1.4.4beta05 and zlib-1.2.5)

Version 1.7.11  (built with libpng-1.4.2 and zlib-1.2.5)

Version 1.7.10  (built with libpng-1.4.1 and zlib-1.2.3.9)
  Added missing "(...)" in png_get_uint_32().
  Only compile png_get_uint_32(), etc., when PNG_LIBPNG_VER < 1.2.9
  Revised help info for "-zitxt".

Version 1.7.9  (built with libpng-1.4.1 and zlib-1.2.3.9)
  Defined TOO_FAR == 32767 in pngcrush.h (instead of in deflate.c)
  Revised the "nolib" Makefiles to remove reference to gzio.c and
    pnggccrd.c
  Imposed user limits of chunk_malloc_max=4000000 and chunk_cache_max=500.

Version 1.7.8  (built with libpng-1.4.0 and zlib-1.2.3.5)
  Removed gzio.c

Version 1.7.7  (built with libpng-1.4.0 and zlib-1.2.3.4)
  Updated bundled libpng to version 1.4.0.
  Check the "-plte_len n" option for out-of-range value of n.
  Changed local variable "write" to "z_write" in inffast.c (zlib-1.2.3.4)
    to avoid shadowed declaration warning.

Version 1.7.6  (built with libpng-1.4.0rc02 and zlib-1.2.3.2)
  Change some "#if defined(X)" to "#ifdef X" according to libpng coding style.
  Added some defines to suppress pedantic warnings from libpng-1.2.41beta15
    and later.  A warning about deprecated access to png_ptr->zstream is
    otherwise unavoidable.  When building the embedded libpng, a warning
    about png_default_error() returning is also otherwise unavoidable.
  Write premultiplied alpha if output extension is .ppng and
    PNG_READ_PREMULTIPLIED_ALPHA_SUPPORTED is set (needs libpng-1.5.0).
  Check the "-m method" option for out-of-range method value.

Version 1.7.5  (built with libpng-1.2.41beta14 and zlib-1.2.3.2)

Version 1.7.4  (built with libpng-1.2.40rc01 and zlib-1.2.3.2)
  Use unmodified pngconf.h from libpng-1.2.41beta05 or later.

Version 1.7.3  (built with libpng-1.2.40 and zlib-1.2.3.2)
  Print contents of text chunks after IDAT, even when the -n option
    is used.  This requires a slight modification of pngconf.h,
    when libpng-1.2.x is used.

Version 1.7.2  (built with libpng-1.2.40 and zlib-1.2.3.2)
  Added check for "verbose" on some printf statements.

Version 1.7.1  (built with libpng-1.2.39 and zlib-1.2.3.2)
  Revised some prototypes to eliminate "Shadowed Declaration" warnings.
  Moved warning about discarding APNG chunks to the end.
  Replaced *.tar.lzma with *.tar.xz in the distribution.

Version 1.7.0  (built with libpng-1.2.38 and zlib-1.2.3.2)
  Save (but do not recompress) APNG chunks if the output file has the
    ".apng" extension and the color_type and bit_depth are not changed.

Version 1.6.20 (built with libpng-1.2.38 and zlib-1.2.3.2)
  Changed local variable "write" to "wwrite" in inffast.c (zlib) to avoid
    shadowed declaration warning.

Version 1.6.19 (built with libpng-1.2.37 and zlib-1.2.3.2)
  Added missing braces that cause an incorrect png_error() to be issued.

Version 1.6.18 (built with libpng-1.2.37 and zlib-1.2.3.2)
  Removed extra FCLOSE(fpin) and FCLOSE(fpout) in the first Catch{} block,
    since they get removed anyway right after that (hanno boeck).
  Define PNG_NO_READ|WRITE_cHRM and PNG_NO_READ_|WRITEiCCP in pngcrush.h
    and reordered pngcrush.h

Version 1.6.17 (built with libpng-1.2.36 and zlib-1.2.3.2)
  Defined TOO_FAR == 32767 in deflate.c (again).  The definition
    has continually been inadvertently omitted during zlib updates
    since pngcrush version 1.6.4.
  Revised handling of xcode files so at least we can get printout
    of IHDR values with "pngcrush -fix -n -v xcode.png".
  Moved ChangeLog.txt back into pngcrush.c so it does not get lost.
  Removed single quotes from the ChangeLog.

Version 1.6.16 (built with libpng-1.2.35 and zlib-1.2.3.2)
  Added -newtimestamp and -oldtimestamp options and changed
    default condition to timestamping the output file with
    the current time (i.e., -newtimestamp is default)
  If the -oldtimestamp option is used then the output file
    has the same timestamp as the input file.
  Added CgBI chunk detection.

Version 1.6.15 (built with libpng-1.2.35 and zlib-1.2.3.2)
  Fixes some missing typecasts on png_malloc() calls, patch from
    an anonymous reporter to the SourceForge bug tracker.
  Added -time_stamp option to change time stamping from default
    condition.

Version 1.6.14 (built with libpng-1.2.35 and zlib-1.2.3.2)
  Avoids CVE-2009-0040.

Version 1.6.12 (built with libpng-1.2.34 and zlib-1.2.3.2)

Version 1.6.11 (built with libpng-1.2.33 and zlib-1.2.3.2)
  Eliminated a memory leak in libpng with writing bad tEXt chunks.

Version 1.6.10 (built with libpng-1.2.31 and zlib-1.2.3.2)
  Add sTER chunk support.

Version 1.6.9 (built with libpng-1.2.31 and zlib-1.2.3.2)
  Updated cexcept.h to version 2.0.1
  Add missing curly brackets.

Version 1.6.8 (built with libpng-1.2.29 and zlib-1.2.3.2)
  Fixed bug with handling of -z and -zi options.

Version 1.6.7 (built with libpng-1.2.29 and zlib-1.2.3.2)
  Moved PNG_UINT_CHNK and some other defines from pngcrush.h to pngcrush.c
  Reject invalid color_type or bit_depth.

Version 1.6.6 (built with libpng-1.2.29 and zlib-1.2.3.2)
  Added dSIG support.  Pngcrush will not rewrite an image containing
  a dSIG chunk immediately following the IHDR chunk, unless the
  dSIG is explicitly removed with "-rem dSIG" or explicitly kept
  with "-keep dSIG".  In the latter case the saved dSIG chunks will
  become invalid if any changes are made to the datastream.

  Fixed bug in writing unknown chunks from the end_info_ptr.

Version 1.6.5 (built with libpng-1.2.29 and zlib-1.2.3.2)
  Discontinued adding a new gAMA chunk when writing sRGB chunk.

Version 1.6.4 (built with libpng-1.2.9rc1 and zlib-1.2.3)
  Fixed bug in handling of undocumented -trns_a option (Michal Politowski).
  Fixed bug with "nosave" handling of unknown chunks.

Version 1.6.3 (built with libpng-1.2.9beta11 and zlib-1.2.3)

  Fixed documentation of iTXt input (Shlomi Tal).
  Removed #define PNG_INTERNAL and provided prototypes for some
  internal libpng functions that are duplicated in pngcrush.c

Version 1.6.2 (built with libpng-1.2.8 and zlib-1.2.3)

  Fixed bug with "PNG_ROWBYTES" usage, introduced in version 1.6.0.
  The bug could cause a crash and only affects the "nolib" builds.

  Converted C++ style (// ...) comments to C style (/* ... */).

  Defined TOO_FAR == 32767 in deflate.c (again).  The definition was
  omitted from version 1.6.0 when zlib was upgraded to version 1.2.3.

Version 1.6.1 (distributed as 1.6.0, built with libpng-1.2.8 and zlib-1.2.3)

  Copied non-exported libpng functions from libpng into pngcrush, to make
  pngcrush play more nicely with shared libpng.  These are not compiled
  when a static library is being built with the bundled libpng and
  pngcrush.h is included.

Version 1.6.0-grr (built with libpng-1.2.4 and zlib-1.1.4pc or zlib-1.2.2)

  Moved ChangeLog out of pngcrush.c comments and into a separate file.

  Filtered pngcrush.c through "indent -kr" and "expand" for readability.

  Moved 550 lines of usage/help/copyright/license/version info to separate
  function(s) and cleaned up significantly.

  Added some comments for ease of navigation and readability.

  Stripped out a bunch of ancient-libpng compatibility stuff.

  Defined PNG_UINT_* macros (pngcrush.h for now).

  Fixed unknown-chunk handling ("-rem alla" and "-rem gifx" now work).

  Created modified version of makefile that supports external zlib.

  Added support for methods using Z_RLE zlib strategy (zlib 1.2.x only).

  Documented -huffman option in usage screen.

  Added IDAT statistics to final per-file summary.

  Added utime() support to give output files same timestamps as input files.

Version 1.5.10 (built with libpng-1.2.4 and zlib-1.1.4pc)

  Fixed bug, introduced in 1.5.9, that caused defaults for method 0 to
  be used instead of copying the original image, when the original was
  already smallest.

Version 1.5.9 (built with libpng-1.2.4beta3 and zlib-1.1.4pc)

  Work around CPU timer wraparound at 2G microseconds.

  Upgraded zlib from 1.1.3 to 1.1.4.  Pngcrush is believed not to
  be vulnerable to the zlib-1.1.3 buffer-overflow bug.

  Choose the first instance of smallest IDAT instead of the last,
  for faster final recompression, suggested by TSamuel.

Version 1.5.8 (built with libpng-1.2.1)

  Added -trns_a option for entering a tRNS array.

Version 1.5.7 (built with libpng-1.2.0)

  Added setargv.obj to Makefile.msc to expand wildcards, e.g., *.png

  Use constant string "pngcrush" instead of argv[0] when appropriate.

  Only check stats for infile==outfile once per input file, or not at all
  if "-nofilecheck" option is present or if a directory was created.

  Fixed bugs with changing bit_depth of grayscale images.

Version 1.5.6 (built with libpng-1.0.12)

  Eliminated extra "Removed the cHNK chunk" messages generated by version
  1.5.5 when "-rem alla" or "-rem allb" is used.

  All unknown chunks including safe-to-copy chunks are now removed in
  response to the "-rem alla" or "-rem allb" options.

  Issue a warning if the user tries "-cc" option when it is not supported.

Version 1.5.5 (built with libpng-1.0.12)

  Reset reduce_to_gray and it_is_opaque flags prior to processing each
  image.

  Enable removal of safe-to-copy chunks that are being handled as unknown
  e.g., "-rem time".

Version 1.5.4 (built with libpng-1.0.11)

  Added 262 to the length of uncompressed data when calculating
  required_window_size, to account for zlib/deflate implementation.

  Added "-bit_depth n" to the help screen.

  Call png_set_packing() when increasing bit_depth to 2 or 4.

  Added warning about not overwriting an existing tRNS chunk.

  Reduced the memory usage

  Write 500K IDAT chunks even when system libpng is being used.

  Ignore all-zero cHRM chunks, with a warning.

Version 1.5.3 (built with libpng-1.0.9beta5)

  Added "-loco" option (writes MNG files with filter_method 64)

  "-dir" and "-ext" options are no longer mutually exclusive, e.g.:
  pngcrush -loco -dir Crushed -ext .mng *.png

Version 1.5.2 (built with libpng-1.0.9beta1)

  Added "-iccp" option.

  Increased the zlib memory level, which improves compression (typically
  about 1.3 percent for photos) at the expense of increased memory usage.

  Enabled the "-max max_idat_size" option, even when max_idat_size
  exceeds the default 1/2 megabyte size.

  Added missing "png_ptr" argument to png_error() call

  Added "-loco" option, to enable the LOCO color transformation
  (R->R-G, G, B->B-G) while writing a MNG with filter_method 64. Undo
  the transformation and write the regular PNG filter_method (0) if the
  MNG filter_method 64 is detected.

  Revised the "-help" output slightly and improved the "-version" output.

  The "-already[_crushed]" option is now ignored if the "-force" option
  is present or if chunks are being added, deleted, or modified.

  Improved "things_have_changed" behavior (now, when set in a particular
  file, it is not set for all remaining files)

Version 1.5.1 (built with libpng-1.0.8)

  Disabled color counting by default and made it controllable with new
  -cc and -no_cc commandline arguments.

  Added some #ifdef PNGCRUSH_COUNT_COLORS around code that needs it.

  Revised count_colors() attempting to avoid stack corruption that has
  been observed on RedHat 6.2

  Added the word "irrevocably" to the license and changed "without fee"
  to "without payment of any fee".

Version 1.5.0 (built with libpng-1.0.8)

  After encountering an image with a bad Photoshop iCCP chunk, pngcrush
  1.4.5 through 1.4.8 write sRGB and gAMA=45455 chunks in all
  remaining PNG files on the command line.  This has been fixed so the
  correction is only applied to the particular bad input file.

Version 1.4.8 (built with libpng-1.0.8rc1)

  Detect and remove all-opaque alpha channel.
  Detect and reduce all-gray truecolor images to grayscale.

Version 1.4.7 (built with libpng-1.0.8rc1)

  Restored the "-ext" option that was inadvertently overridden with
  a new "-exit" option in version 1.4.6 ("-exit" is used to force an
  "exit" instead of a "return" from the main program).

Version 1.4.6 (built with libpng-1.0.8rc1)

  Fixed bug in color-counting of noninterlaced images.

  Added capability of processing multiple rows at a time (disabled by
  default because it turns out to be no faster).

  Replaced "return" statements in main() with "exit" statements.
  Force exit instead of return with "-exit" argument.

  Added the UCITA disclaimers to the help output.

Version 1.4.5 (built with libpng-1.0.7rc2 and cexcept-1.0.0)

  Added color-counting and palette-building capability (enable by
  defining PNGCRUSH_COUNT_COLORS).  In a future version, this will
  give pngcrush the ability to reduce RGBA images to indexed-color
  or grayscale when fewer than 257 RGBA combinations are present,
  and no color is present that requires 16-bit precision.  For now,
  it only reports the frequencies.

  Added "-fix" option, for fixing bad CRCs and other correctable
  conditions.

  Write sBIT.alpha=1 when adding an opaque alpha channel and sBIT
  is present.

  Identify the erroneous 2615-byte sRGB monitor profile being written
  by Photoshop 5.5, which causes many apps to crash, and replace it with
  an sRGB chunk.

  Added a check for input and output on different devices before rejecting
  the output file as being the same as the input file based on inode.

  Added some UCITA language to the disclaimer.

Version 1.4.4 (built with libpng-1.0.6i and cexcept-0.6.3)

  Can be built on RISC OS platforms, thanks to Darren Salt.

Version 1.4.3 (built with libpng-1.0.6h and cexcept-0.6.3)

  Reduced scope of Try/Catch blocks to avoid nesting them, and
  removed returns from within the Try blocks, where they are not
  allowed.

  Removed direct access to the png structure when possible, and isolated
  the remaining direct accesses to the png structure into new
  png_get_compression_buffer_size(), png_set_compression_buffer_size(),
  and png_set_unknown_chunk_location() functions that were installed
  in libpng version 1.0.6g.

Version 1.4.2 (built with libpng-1.0.6f and cexcept-0.6.0)

  Removes extra IDAT chunks (such as found in some POV-ray PNGs) with
  a warning instead of bailing out (this feature requires libpng-1.0.6f
  or later, compiled with "#define PNG_ABORT()").

  Removed old setjmp interface entirely.

Version 1.4.1 (built with libpng-1.0.6e and cexcept-0.6.0)

  Uses cexcept.h for error handling instead of the libpng built-in
  setjmp/longjmp mechanism.  See http://cexcept.sf.net/

  Pngcrush.c will now run when compiled with old versions of libpng back
  to version 0.96, although some features will not be available.

Version 1.4.0 (built with libpng-1.0.6 + libpng-1.0.6-patch-a)

Version 1.3.6 (built with libpng-1.0.5v)

  RGB to Grayscale conversion is more accurate (15-bit instead of 8-bit)
  and now uses only integer arithmetic.

  "#ifdefed" out PNG_READ_DITHER

  Changed "Compressed" to "Uncompressed" in help for -itxt.

  Stifled some compiler warnings

Version 1.3.5 (built with libpng-1.0.5s)

  Add test on stat_buf.st_size to verify fpin==fpout, because stat in
  MSVC++6.0 standard version returns stat_buf.st_ino=0 for all files.

  Revised pngcrush.h to make it easier to control PNG_ZBUF_SIZE and
  PNG_NO_FLOATING_POINT_SUPPORTED from a makefile.

  Restored ability to enter "replace_gamma" value as a float even when
  floating point arithmetic is not enabled.

  Enabled removing tEXt, zTXt, or iTXt chunks by chunk type, i.e.,
  "-rem tEXt" only removes tEXt chunks, while "-rem text" removes all
  three types of text chunk.

  Removed definition of TOO_FAR from pngcrush.h

  Uses new libpng error handler; if a file has errors, pngcrush now will
  continue on and compress the remaining files instead of bailing out.

Version 1.3.4 (built with libpng-1.0.5m)

  Do not allow pngcrush to overwrite the input file.

Version 1.3.3 (built with libpng-1.0.5m)

  Restored ability to enter gamma as a float even when floating point
  arithmetic is not enabled.

Version 1.3.2 (built with libpng-1.0.5k)

  Renamed "dirname" to "directory_name" to avoid conflict with "dirname"
  that appears in string.h on some platforms.

  Fixed "PNG_NO_FLOAING_POINT" typo in pngcrush.h

  "#ifdefed" out parts of the help screen for options that are unsupported.

Version 1.3.1 (built with libpng-1.0.5k): Eliminated some spurious warnings
  that were being issued by libpng-1.0.5j.  Added  -itxt, -ztxt, and
  -zitxt descriptions to the help screen.

  Dropped explicit support for pCAL, hIST, sCAL, sPLT, iCCP, tIME, and
  cHRM chunks and handle them as unknown but safe-to-copy instead, using
  new png_handle_as_unknown function available in libpng-1.0.5k.

Version 1.3.0 (built with libpng-1.0.5j): Added support for handling
  unknown chunks.

  pngcrush is now fixed-point only, unless PNG_NO_FLOATING_POINT_SUPPORTED
  is undefined in pngcrush.h.

  Added support for the iCCP, iTXt, sCAL, and sPLT chunks, which
  are now supported by libpng (since libpng-1.0.5j).  None of these have
  been adequately tested.

  "#ifdefed" out more unused code (weighted filters and progressive read;
  this saves about 15k in the size of the executable).

  Moved the special definitions from pngconf.h into a new pngcrush.h

  Disallow 256-byte compression window size when writing, to work around
  an apparent zlib bug.  Either deflate was producing incorrect results in a
  21x21 4-bit image or inflate was decoding it incorrectly; the uncompressed
  stream is 252 bytes, which is uncomfortably close to the resulting
  256-byte compression  window.  This workaround can be removed when zlib
  is fixed.

  The "-m method" can be used any of the 124 methods, without having to
  specify the filter, level, and strategy, instead of just the first 10.

Version 1.2.1 (built with libpng-1.0.5f): Fixed -srgb parameter so it
  really does take an argument, and so it continues to use "0" if an
  integer does not follow the -srgb.

  Added "-plte_len n" argument for truncating the PLTE.  Be sure not to
  truncate it to less than the greatest index actually appearing in IDAT.

Version 1.2.0: Removed registration requirement.  Added open source
  license.  Redefined TOO_FAR=32k in deflate.c.

Changes prior to going "open source":

Version 1.1.8: built with libpng-1.0.5a.  Runs OK with pngvcrd.c.

Version 1.1.7: added ability to add tEXt/zTXt chunks.  Fixed bug with
closing a file that was not opened when using "pngcrush -n".  Fixed
bug with tEXt/zTXt chunks after IDAT not being copied.
Added alpha to the displayed palette table.  Rebuilt with libpng-1.0.5.

Version 1.1.6: fixed bug with one file left open after each image is
processed

Version 1.1.5: Shorten or remove tRNS chunks that are all opaque or have
opaque entries at the end.  Added timing report.

Version 1.1.4: added ability to restrict brute_force to one or more filter
  types, compression levels, or compression strategies.