Imager release history.  Older releases can be found in Changes.old

Imager 1.005 - 16 Apr 2016

It's now been ten years since I switched to the new Changes file in
release 0.55.

 - revert the ivdformat probes, they don't work as is and trying to
   fix them is too much work for now.

Imager 1.004_004 - 15 Apr 2016

 - test that the ivdformat from Config is correct and look for a valid
   one if it isn't.
   For the strange Win32 failures.

 - fix a copy and paste error in pod in samples/samp-form.cgi

Imager 1.004_003 - 23 Mar 2016

 - add some extra error reporting to the I/O layers tests, this might
   help catch a failure seen on Win32.

Imager 1.004_002 - 20 Mar 2016

 - don't use the seek() method on opened() handles in
   t/200-file/400-basic.t.  In older versions of perl such handles are
   only IO::Handle objects, not IO::File, and don't have a seek()

Imager 1.004_001 - 16 Mar 2016 (Birthday release - but not my birthday)

 - re-work the t/200-file/400-basic.t to correctly handle failures
   It's custom ok() function didn't have a prototype and didn't use scalar().
   This caused ok() to use the note instead of the value being tested when
   the method called returned an empty list.  For an example of the problem
   caused see:

 - remove some noise from when Imager tried to work with bugs in old
   versions of giflib.  I can't do much about the new bugs.

 - the new autolevels filter (Imager 0.99) used an integer for the
   sample scaling factor which caused the top output level to be too
   low (depending on the scaling required.)  It now uses a double.

 - the new autolevels filter had off-by-one errors calculating the
   minimum and maximum luminance from the histogram.  This slightly
   reduced the contrast of the output image.

 - the new autolevels filter now uses a lookup table for 8-bit images
   to avoid a floating point multiply for each sample.

 - fixed several memory leaks detected by valgrind:

   - addcolors() leaked the temporary array of colors supplied to the
     internal API

   - make_palette() leaked the temporary image array (not the images
     themselves) passed to the internal API.

   - combine()/i_combine() leaked its working row buffers

   - getcolorusage()/i_get_anonymous_color_histo() leaked the sample
     buffer if there were too many colors

   - getcolorcount()/i_count_colors() leaked the sample buffer if
     there were too many colors.

   - the nearest_color filter (undocumented until I find a use for it)
     leaked both temporaries passed to the API and internal buffers

   - the internal process of upgrading a paletted image to a direct
     color image would leak a context object reference count.

   - a write failure when writing to a GIF file could leak memory.

   - failing to write to a 1-bit/pixel ICO image could leak memory.

 - Imager no longer deliberately leaks the context object from the
   initial thread.  This was done to ensure there was always a context
   object available, but the code that needed that now handles the
   lack correctly,

 - fixed some uninitialized memory usage detected by valgrind:

   - rotate()/i_rotate_exact()/i_rotate_exact_bg()/i_matrix_transform()/
     i_matrix_transform_bg() didn't initialize a working color value
     if there was zero pixel coverage.

   - i_ft2_cp() didn't initialize the complete color when producing an
     intermediate image.  This caused uninitialized value usage when
     logging the color.

Imager 1.004 - 8 Nov 2015

 - Imager::Color::Table is now pre-loaded by the preload() method.

 - fix an assertion triggered under perl 5.23.4.
   Thanks to A. Sinan Unur for the report and the patch.

 - Imager->new can now be used to read raw image files.
   Thanks to Richard Kelsch for reporting this.

 - deal with output changes from Pod::Spell

Imager 1.003 - 12 May 2015

 - update 1.002 release notes to include the center change for filled
   circle drawing.

 - flood_fill() would escape beyond a 4-connected space under some
   Added many more flood_fill() tests.

Imager 1.002 - 3 Apr 2015

 - drawing anti-aliased filled circles is now 10 to 50 times faster
   depending on the size of the circle.
   This also changed the center from being the center of the pixel to
   being the top left of the pixel to match the filled arcs drawn by

 - enhancements to polygon filling:

   - added a mode parameter to control how overlapping regions behave

   - added a polypolygon() method to fill more than one polygon at a

   - polygon filling is now exposed through the API.

 - added colormodel(), alphachannel() and colorchannels() methods.
   These were added for two reasons:

    - a future version of Imager may allow the number of channels in
      an image to not directly represent the color model of an image.
      eg. a greyscale TIFF image with multiple alpha channels.

    - a future version of Imager may allow an image to be read without
      translation, for example a TIFF file that contains measurements
      from an instrument.  Currently Imager transforms the samples into
      the range 0.0 ... 1.0 which may means the user has to translates
      the value back.

      An untranslated image would be unusable as image data, so
      colormodel() would return "unknown" in this case.

      Similarly a CMYK image might be returned as an "unknown" color
      model image, if the caller chooses to disable translation to

Imager 1.001 - 2 Jan 2015

 - both Imager and perl 5.21.3 define my_strerror(), prevent a conflict
   Thanks to Slaven Rezic for the report and the patch.

 - GIF: clean-up a test file if the test for the giflib 4.2.0 file
   version bug fails.

 - fix Imager::Matrix2d::rotate()'s centre point handling

 - ICO: don't apply the icon mask to images with an alpha channel by

 - make verbose probing output more verbose

 - use Imager::Probe to probe for freetype 1.x

 - The --enable and --disable parameters to the top-level Makefile.PL
   work again.

 - update the bundled/modified Devel::CheckLib to handle the gcc-4
   symlink on Cygwin

Imager 1.000 - 28 Jul 2014

There's nothing special about Imager 1.000, it's just another release.

 - fix the skip check for the iolayer qr// buffer test

 - improve error reporting for the iolayer test failing on a small
   number of Win32 CPAN testers

Imager 0.99_02 - 21 Jul 2014

 - Imager::Filter::Mandelbrot (and dynfilt/mandelbrot.c) - initialize
   the blue channel in the generated palette, and allow each color
   component to be in the range 100..255 instead of just 100..254.
   Thanks to Ralf Neubauer.

 - revert "improved the XS for i_io_read() and i_io_raw_read()"
   This caused problems with older perls and didn't provide much of a
   performance improvement.

 - support Inline 0.57 and later.
   Inline 0.57 changed the "with" interface.

 - don't define our own MAXINT macro, it conflicts with windows header
   files and in a few places it was the wrong value to use anyway.

Imager 0.99_01 - 29 Jun 2014

 - GIF: add support for giflib 5.1.0, which added error code pointer
   parameters to EGifCloseFile() and DGifCloseFile().

 - GIF: avoid a double-free when do_write() fails.

 - fix SV type probing to work on perl before 5.12.  Broken in 0.99.

 - PNG: skip the benign error test before libpng 1.6.0, since the
   error we're testing didn't exist before 1.6.0. (continued)

Imager 0.99 - 25 Jun 2014

 - Imager::IO->new_buffer() (and hence Imager->read()'s data
   parameter) now accepts a reference to a scalar as well as just a
   plain scalar.

 - Imager::IO->new_buffer() now always makes a copy of the passed in
   buffer to avoid problems with temporary objects used for the return
   value of SvPVbyte().

 - improved the XS for i_io_read() and i_io_raw_read()

 - load plugins from absolute paths on Android
   Thanks to Brian Fraser.

 - added the jpeg_optimize parameter for writing JPEG files.  This can
   significantly reduce file sizes, but uses more memory and time.

 - the autolevels filter now works on the luminosity of the image
   rather then working per channel.  The old autolevels filter is
   still available as "autolevels_skew".

 - Imager::File::PNG now supports libpng 1.6.10.
   1.6.10 changed CRC errors from benign errors to normal errors,
   which broke the test which used CRC errors to check for benign
   error support.  Switched to using a 1-bit grey-scale image with a
   palette to test for benign errors.

Imager 0.98 - 3 Jan 2014

Incompatible changes:

 - the return value of setpixel() has changed.

   Previously the return values for undrawable pixels vs caller errors
   changed depending on whether you used the multiple pixel calling
   mechanism, or the single pixel mechanism.


    - for an invalid parameter, such as an unknown colour, or missing
      parameter, an empty list (or undef in scalar context) is
      returned, and errstr() is set,

    - otherwise the number of pixels drawn is returned.  If none of
      the pixels could be drawn (they were all outside the image), "0
      but true" is returned.

Other changes:

 - when drawing on an image with an alpha channel where the source
   minimum is greater than zero, Imager would read from beyond the end
   of a malloc() allocated buffer.  In rare circumstances this could
   lead to some of the source image not being written to the target
   image, or possibly to a segmentation fault.
   I don't believe this has any security concerns beyond that.

 - if the first text drawn with Imager::Font::T1 is not anti-aliased,
   text drawn would be nonsense.  This would also read beyond the end
   of a malloced buffer.

 - non-AA text drawn with an Imager::Font::TT (Truetype 1) font would be
   truncated early.

 - Imager::Font::Wrap no longer requires the image parameter.

 - a documentation typo fix in Imager::Transformations
   Thanks to Adrian Yee.

 - Image::Probe, Imager's internal module for probing library
   locations now searches the directories specified by LD_RUN_PATH,
   and the corresponing s/\blib$/include/ directories for header

Imager 0.97 - 15 Jul 2013

No changes from 0.96_02.

Imager 0.96_02 - 8 Jul 2013

 - PNG: treat a version mismatch between headers and library as a
   probe failure.

 - PNG: the check for benign error support is more complex than a
   simple library version check, check for the appropriate macro.
   Thanks for Slaven Rezic for following up on my CPAN Testers result

 - W32: add a missing head1 AUTHORS to the old

Imager 0.96_01 - 1 Jul 2013

 - TIFF: handle SampleFormat = 2 by translating the signed integer
   values to unsigned by flipping their sign bits.

   Handle SampleFormat = 3 where possible.

   SampleFormat is ignored for paletted images.

   Mixed SampleFormat are handled incorrectly, since libtiff returns
   only the first SampleFormat value, and an image with both an alpha
   channel and SampleFormat = 2 for color channels probably has a
   different SampleFormat for the alpha channel.

 - XS clean up:

   - Imager::Color's rgba() method now returns its values as integers
     instead of floating point. (IV instead of NV).

   - The XS for i_map() and i_matrix_transform() now use the AV *
     typemap instead of rolling their own.

   - The XS for DSO_call() now uses the HV * typemap instead of
     rolling it's own.

   - The XS for i_poly_aa(), i_poly_aa_cfill(), i_transform() and
     i_gradgen() now use a new T_AVARRAY typemap that greatly
     simplifies the XS code.

   - many other minor XS changes

   - some XS code formatting

 - TT (Freetype 1.x) and FT2 (Freetype 2.x) non-antialiased rendering
   now renders in alpha-combining mode, to match antialiased output.

 - add sample code and cookbook notes for drawing a blurred drop-shadow.

 - add support for libpng 1.6, which changed the defaults for
   reporting some types of error.

 - FT2: use gsamp/psamp() to transfer pixels from the work image to
   the output image instead of gpix/ppix.

 - eliminate various GCC warnings, mostly initialized but otherwise
   unused variables.

Imager 0.96 - 19 May 2013

 - rearrange Imager's test files.

 - Imager::Probe::_gcc_lib_paths now forces C locale.
   A sufficiently recent gcc prints localized keys for the search
   paths, which avoids that localization.

 - fix a pod error detected by Pod::Simple 3.27 and laterm and skip
   and true.
   reported by Andreas König

 - add a test for unclosed POD in C source, and fix the errors it

 - fix some spelling errors detected by the newer aspell in Debian

 - remove a trailing ' from lib/Imager/Font/'s POD
   thanks to gregor herrmann <>

 - add a test for standard =head1 commands and fix the errors found

Imager 0.95 - 19 Apr 2013

 - handle the SVf_UTF8 flag correctly on magic strings passed to
   Imager::IO's write() and raw_write() methods.
   This may misbehave in perl 5.6.x since the UTF8 flag may not be
   visible in the main SV after SvGETMAGIC().

 - document that bounding_box() ignores the transformation matrix
   supplied to tranform() for fonts.

Imager 0.94_02 - 5 Apr 2013

 - enable debug logging for the standard font tests

 - skip the overloaded UTF-8 text output checks on 5.6.x and earlier,
   due to a bug in that version of perl.

 - don't test for read() failing on a write-only handle in 5.8.x and
   earlier, this was fixed in change 862083f7e4 in perl.

 - report the version of Inline found during testing (an attempt to
   diagnose some intermittent failures.)

Imager 0.94_01 - 2 Mar 2013

 - NOTE: possibly backward incompatible:
   support reading from/writing to perl filehandes that aren't raw

   This allows Imager's I/O to honour handles with layers such as
   gzip, scalar file handles or tied file handles.

   This is backward incompatible in that previous Imager would simply
   use fileno() to retrieve the fd for the file and call write(2) etc
   on it directly.

 - moved most of README to lib/Imager/Install.pod which should make it
   more accessible to the "web" generation, also significantly updated
   and re-worked it.

 - updated README's for the separately distributed modules to refer to
   Imager::Install, and that they need -dev versions of packages

 - the JPEG test code now reports the compile-time library version

 - avoid a possible compiler optimization issue on CentOS 5.9 i386 by
   rearranging (and mildly optimizing) some code.

 - fix a POD error in Imager::Fill (detected on new Pod-Simple)

 - fix a broken link to Graphics::Magick

 - drawing text to a channel with FT2 would draw with random coverage
   due to an uninitialized alpha channel.

 - marked the function pointer underlying the mm_log() API with the
   correct gcc magic and fixed the resulting warnings.

 - fixed some other compiler warnings

 - Imager::Font::W32 now properly reports text drawing errors

 - handle the SVf_UTF8 flag correctly on magic (eg. overloaded)
   strings passed as text to draw(), bounding_box(), has_chars(),
   glyph_names() (where supported) in each of the font drivers.
   This may misbehave in perl 5.6.x since the UTF8 flag may not be
   visible in the main SV after SvGETMAGIC(). (partial)

Imager 0.94 - 15 Dec 2012

Variations on some of these changes were included in development

 - improved thread safety
   - the internal error stack and log file handle are now in a per-thread
     context object
   - JPEG now captures IPTC information in a thread-safe way
   - avoid globals where possible for warning capture in libtiff
   - use a mutex to avoid re-entering thread-unsafe giflib
   - use a mutex to avoid re-entering thread-unsafe tifflib
   - use a mutex to avoid re-entering thread-unsafe T1Lib
   - use a library handle per thread for freetype 2.
   - use an engine handle per thread for freetype 1.x.
   - originally these changes broke ABI compatibility, this has been

 - clarify the return value of getpixel();

 - fixed a race condition in parallel testing for T1

 - fixed a bug in handling yoff for untransformed image-based fills

 - documentation improvements for Imager::Fill

 - FT2: report the library version while testing.

Imager 0.93 - 15 Oct 2012

Bug fixes:

 - previously a transparency enabled GIF write (the default) would
   always use a GIF89a header even if none of the images were

 - previously the probe step for freetype-config would fail on cygwin

 - catch an invalid matrix parameter to convert() instead of crashing

 - remove the 16-bit/sample limitation from the documentation for
   setsamples(), it hasn't applied for many releases.

 - don't copy setsamples() data parameter, it may be a large scalar

 - clean up .dSYM directories generated performing probes on OS X
   Mountain Lion.

 - pass the --verbose command-line option through to Imager::Probe in
   sub-module's Makefile.PLs


 - support for giflib 5.0. (5.0.1 and later)
   The giflib API Imager uses doesn't have a mechanism to set the
   header version in 5.0.0.

 - update the GIF library probe code to check for the new giflib 5.0.1
   EGifSetGifVersion() function, and to check for the giflib 4.2.0
   uninitialized gif89 bug.

 - avoid static variables when capturing IPTC data from JPEG files

 - match Imager::Font;:T1's error message translations to those from
   later versions of T1Lib.

 - for libtiff versions that support extended warning handlers (3.8.0
   or later), use them to avoid some global variables.

 - note Image::ExifTool for better metadata support.

Imager 0.92 - 14 Aug 2012

 - giflib 4.2 eliminates the GIF_LIB_VERSION macro, handle that
   correctly for both probing and runtime.

 - allow building JPEG/imexif.c on C89 compilers.

 - allow building GIF/imgif.c on C89 compilers.

Imager 0.91 - 4 Jun 2012

Bug fixes:

 - The size of rotated images is no longer rounded up so aggressively.
   Added rounding to the linear interpolation done for rotate() and
   matrix_transform() for 1 and 3 channel 8-bit images.
   Adjusted the two tranlate matrices used to build the final rotation
   matrix to use the distance between the outlier pixels rather than
   the pixel dimensions (ie. dimension-1).
   With all of this the output of rotate(degrees => 270) on an 8-bit
   image exactly matches the output of rotate(right => 270).

Other changes:

 - eliminate the old IIM_new(), IIM_DESTROY() names from Imager's
   internals, those names only matter for the XS interface.

 - improve error reporting when PERL_INITIALIZE_IMAGER_CALLBACKS finds
   the API level compiled into a loadable module such as
   Imager::File::GIF doesn't match that of Imager.  Previously it
   could be difficult to determine exactly which module was failing to

 - added Imager->check_file_limits() as an interface to the
   i_int_check_image_file_limits() API.

 - Imager->set_file_limits(reset => 1) now resets the limits to the
   defaults rather than setting them all to unlimited.

 - wrote a brief security note as Imager::Security

Imager 0.90 - 30 Apr 2012

Bug fixes:

 - Imager::Test::is_imaged() attempted to process an "epsilon"
   parameter but the prototype didn't allow for the extra parameter.
   Corrected the prototype.

 - when downconverting samples (eg. from floating point to 8 bit)
   Imager now rounds the sample value rather than attempting to
   allocate input values over output values equally.  The old
   behaviour had the probably surprising effect of converting a
   floating point 2.1/255.0 into an 8-bit 3 rather than the expected
   This may result in slightly different 8 or 16-bit output images.

 - BI_BITFIELD BMP images were handled incorrectly, both in
   incorrectly calculating the space required for the masks and in
   processing the image data itself.

 - some odd width BMP BI_RLE4 images use run lengths that run one off
   the edge of the image.  Imager now allows this, discarding the
   extra column.

 - odd length RLE4 runs in BMP images were decoded incorrectly.

 - pkg-config could sometimes return a library that was in a directory
   EU::MM / $Config{libpth} doesn't know about, but the compiler did.
   If no -L is included in the pkg-config library information check if
   EU::MM can find the library, and if not, search our configured
   directories and insert that into the library flags.

 - Imager::Probe can now probe for libraries with dependent libraries,
   common for static linking, eg. libpng requires libz.

 - libpng 1.5 specific probes were looking for libpng 1.4 filenames.

 - added alternative probe configurations that try to link libz, to
   handle a statically linked libpng.

 - if a probe includes testcode, Imager::Probe now checks that code as
   part of the process of checking each configuration rather than as a
   post test of the found configuration.  This allows alternate
   configurations to be checked if a matching but non-working
   configuration is found.

Other changes:

 - when reading or writing images via callbacks, the default callbacks
   now push an error message indicating that a required callback was
   called but not supplied.

 - clarify which callbacks are required for reading and writing TIFF

 - improve logging for creation of callback I/O layers.

 - a little more documentation for Imager::Probe.

 - the i_get_file_background() and i_get_file_backgroundf() APIs now
   return int to indicate whether the i_background tag was found.

 - PNG rework
   - improve error reporting
   - add png_interlace, png_bits tags
   - read paletted images as paletted images, including transparency
   - read 1 bit greyscale images as a type suitable for other file
     handlers to write as bilevel
   - read 16 bit/sample PNG as 16-bit/sample Imager images
   - write "bilevel" paletted images as 1 bit grayscale images
   - write paletted images as paletted images
   - write 16-bit (or higher)/sample images as 16-bit/sample PNG
   - improved metadata support

Imager 0.89 - 18 Mar 2012

Bug fixes:

 - getpixel(..., type => "float") and the API i_gpixf() have been
   broken on paletted images since they were implemented.

Other changes:

 - links in the METHOD INDEX now point at the method documentation
   rather than the heading you can find them under.

 - Imager (and the bundled dynamic modules) no longer fallback to
   using DynaLoader if loading via XSLoader fails.
   For the bundled modules this could hide useful error messages.

 - IM_DEBUG_MALLOC mymalloc() no longer sn?printfs() a string to a
   buffer in the array of allocations, but just stores the filename
   pointer and line number.

Imager 0.88 - 22 Feb 2012

 - describe how to build libgif etc on OS X in such a way as to be
   compatible with a fat binary perl (such as the system perl), in

 - update the change notes for 0.77_01 to indicate that libungif
   support was removed.

 - add some other imaging modules to SEE ALSO

 - note that the generator of the apparently non-DFSG-free postscript
   in MMOne.pfb is a Debian package.

 - setsamples() is now a true complement to getsamples() - it can
   write 8-bit or floating point samples from a scalar or array
   reference.  This adds i_psamp() and i_psampf() to the C level API.

 - the XS interfaces to i_gsamp(), i_gsampf() and i_gsamp_bits() have
   changed to make better use of the typemap, but these aren't part of
   the perl level API anyway.  There were no changes to the C level
   interfaces to these functions.

 - getpixel() and setpixel() now accept a mix of scalar and array
   references for the x and y parameters, and unequal array lengths is
   no longer an error.

Bug fixes:

 - correctly calculate the size of a rotated image

 - fix incorrect rounding of color values for matrix_transform() and

 - Win32 text output is now done in normal combine mode, the alpha
   component of the color is now significant.

 - remove long unused gif case from read() (partial)

 - the getsamples() target parameter was being treated as a hashref
   when it's meant to be an array reference.

 - getpixel() and setpixel() now returns an empty list when invalid
   parameters are supplied.
   Invalid values for type now result in an error for getpixel().

Imager 0.87 - 03 Jan 2012

 - document the return value of the filter() method.

 - document i_gsamp_bits() and i_psamp_bits().

 - properly increment the Imager::Matrix2d $VERSION.

 - actually include the Imager::Test tests in the dist

 - correctly read and write 256x256 pixel ICO files

 - make the error message from read() or read_multi() when they can't
   identify the file type match reality.

 - read() now uses $FORMATGUESS if it can't determine the file type
   based on the file header, to match read_multi().

 - re-work and add tests for def_guess_type().  def_guess_type() no
   longer returns any random file extension as the file type.

 - add gray4, gray16 and gray as presets for make_colors.

 - add make_palette() method that produces a palette from one or more

 - fix the Imager dependency for the separately distributed font

 - fix i_render_color() to properly draw in "normal" mode - ie. when
   writing to a 1 or 3 channel image the second or fourth channel of
   the source color was being ignored, it is now significant.

Imager 0.86 - 31 Oct 2011

 - improve error reporting for W32 tests

Imager 0.85_02 - 24 Oct 2011

Bug fixes:

 - eliminate unused i_gif_opts type (clean-up)

 - fix combine=0 fill color anti-aliasing on the double/sample path

 - make default text color non-transparent

 - apply the last of the Debian unforwarded spelling fixes

 - the log() method used its message parameter as a C level format

 - provide our own STRLEN typemap entry for older perls.

 - add extra ppport.h configuration to support older perls.

 - depend on Scalar::Util, since we use it and older perls don't have

 - add overloaded eq to Imager::Matrix2d, since older perls don't seem
   to synthesize it from overloaded "".

  - use T1_StrError() for error messages on modern libt1

 - actually load the font rather than just adding it to the catalog on

 - Imager::Font->new now produces better error messages for the T1

 - the font has_chars() method now returns perl's true and false
   values in list context rather than integers, which should be more

 - the btm data structure used by the flood_fill code is now
   initialized more efficiently.

 - updated the Thanks list in README

 - check there's at least one coefficient for the convolution filter

 - make the APIRef synopsis ordering consistent, older versions of
   perl could order it differently.

 - we rely on's d_vsnprintf as to whether we use
   vsnprintf/snprintf, which is defined in the Win32 even
   though it only has _ prefixed versions of these.  Define our own
   prefixed names on Win32.

 - fix library detection with MSVC

 - search a few more library directories, so EU::MM doesn't discard
   them.  Hopefully fixes:

Imager 0.85_01 - 10 Oct 2011

 - add simple tests for the Imager::Test test_image generators

 - io_glue I/O buffering re-work:

   - reorganize io_glue to do it's own buffering by default

   - the unbuffered functions are available as i_io_raw_read() (or
     raw_read() from perl) etc but are not recommended for typical

   - use the new i_io_peekn() when checking for file magic to avoid
     seek, allowing Imager to detect the file type and read the file
     from an unseekable stream (for formats that don't use random

   - added several new I/O layer API functions.

   - fix the TGA performance problem, most noticably on Win32

   - TIFF now uses wrapper functions of the correct types to avoid casts

   - the callback IO object did its own buffering, controlled by the
     maxbuffer parameter supplied to the read() and write() methods.
     This buffering has been removed, to avoid redundancy with the
     common io_glue buffering.  This also avoids a bug in that code
     which could rarely pass a zero length to the read callback and
     then panic about the result.

   - the callback IO object now tests the result of calling the close
     callback, which should return true for success.

   - the PNM reader did its own buffering.  This buffering has been
     removed to avoid redundancy with the common io_glue buffering.

   - previously the read handlers for fd and callback I/O layers would
     call the underlying interface (POSIX read or your supplied callback)
     until it filled the buffer.  It now only makes one call.

   - added public constructors for I/O layer objects (see Imager::IO)

   - all core file handlers now use the i_io_foo() wrappers to gain
     access to buffered I/O rather than calling the I/O layer
     callbacks directly.

   - all core file handlers now check for error on close.

   - Backward compatibility: if you hava custom file handlers, you can
     use i_io_write() etc since they're available as macros in older
     versions of Imager.

   - eliminate the final remnants of io_glue_commit_types().

   - bump IMAGER_API_VERSION, since the above may break assumptions.

 - removed the long unused i_gen_reader() and i_gen_writer() utility

Imager 0.85 - 29 Aug 2011

The main changes in the release versus 0.84 are:

 - on 64-bit systems, 64-bit types are consistently used for image
   dimensions and co-ordinated, and for memory block sizes.

 - handle IFD loops in TIFF files correctly.  Previously this would
   lead to an infinite loop.

Bug fixes:

 - fix the link in the getheight() entry in the method index

Imager 0.84_02 - 22 Aug 2011

Development release, this will become 0.85 if CPAN testers is

Bug fixes:

 - the image file limits set by set_file_limits() weren't being
   checked when reading TIFF files.

 - Provide more information about file format module load errors on a
   failed image file read() or write().

 - use TIFFReadDirectory() instead of TIFFSetDirectory() to iterate
   through TIFF images, since it checks for IFD loops.

 - don't leak memory when out of range palette indexes are supplied to

 - require a later version of CPAN::Meta to ensure JSON::PP and
   CPAN::Meta::YAML are available.

 - hoist the per-line calculations for the flines implementations, and
   modernize the tests a bit.

 - detect snprintf()/vsnprintf() (cheat by using and use
   them when available.

 - if t1lib failed to reinitialize it would be left as marked

 - update the bundled (and still modified) Devel::CheckLib

Imager 0.84_01 - 8 Aug 2011

Development release as a sanity check for the types re-work.

Massive types re-work:

 - the type used internally for pixel co-ordinates and image sizes is
   now 64-bit on 64-bit platforms (at least sane ones).

 - size_t is now used consistently for memory block sizes.

 - since this changes the binary interface, the Imager API version has
   been incremented.  Any module that uses the API will need to be
   rebuilt.  In most cases that will be enough, but calls to any APIs
   that take a pointer to image sizes may need source changes.

 - you should be able to create very large images on 64-bit systems,
   if you have enough memory.  Successfully created a 32768 x 49192 x
   3 channel image, filled with a tiled image and scaled it.  The
   unscaled image was also successfully saved to a JPEG.

 - check the image size before attempting to write BMP, GIF, JPEG,
   PNG, SGI, TGA, TIFF images.

 - correctly handle reading TGA images over 32767 pixels wide or tall.

Incidental changes:

 - the gif_left and gif_top tags are now clamped to non-negative
   values when writing a GIF image.

 - removed dead callback read/write code

The default maximum memory size when reading files is now 1G.

Imager 0.84 - 20 Jun 2011

 - Imager no longer inherits from Exporter (unless you're running an
   old, old perl.

 - Imager can now write progressive JPEGs (it could always read them).

Bug fixes:

 - re-work, document and test Imager's logging facility.

 - fix META.yml testing and the generated META.yml

 - test and add error reporting to to_*() family methods

 - add to_rgb_double() method.

 - Imager no longer exports anything by default

 - convert colors to grayscale if the supplied (or generated) palette
   contains only grays when performing error diffusion color

 - writing a paletted image to GIF wouldn't always use the colors
   supplied (or generated, eg. via make_colors => "mono"), which was
   confusing at best.

 - replace (imager|tony) in the doc, since I don't
   plan to continue receiving mail at that address.

Imager 0.83 - 21 May 2011

Bug fixes:

 - diag() the error message on failure for some TIFF tests that are
   failing on a Solaris smoker.

Imager 0.82_01 - 17 May 2011

Dev release, in case the compose tests are too sensitive.

Bug fixes:

 - Imager::Font::T1 incorrectly checked for absolute filename under
   Win32.  Thanks to kmx for the report and fix.

 - compose() with the target, source or mask position off the top or
   left of the target image didn't clip the source image correctly.

 - compose() now returns a useful error message on a non-positive

 - now at 100% test coverage. (As opposed to, umm, much,
   much less.)

Imager 0.82 - 14 Mar 2011

Bug fixes:

 - eliminate calls to i_has_format() from the test suite, since it's
   no longer a useful way to check for file format support.  Eliminate
   i_has_format() from the functions exposed via XS.

 - eliminate calls to note(), which isn't in the (very old) version of
   Test::More we have as a pre-requisite.  note() is modern enough
   that I don't feel a need to require a Test::More upgrade for it.

 - skip the threads tests on Test::More 2.00_*

 - add an (unshipped) test to check Imager's internal POD links

 - improve the library detection summary

 - increase the version of Imager::Font::Type1 so that upgrades don't
   downgrade the version in this file.

 - if we see an -rpath (or -R) option in $Config{lddlflags} supply
   that option for the directories that would normally go in
   LD_RUN_PATH.  Typically an explicit -rpath overrides LD_RUN_PATH.

Imager 0.81 - 14 Feb 2011

 - added coverage tests for masked images (maskimg.c @100% test coverage)

 - add hsv() method to Imager::Color
   Thanks to Leolo (Philip Gwyn)

 - split libt1 Type 1 font support into a sub-module (partial)

 - add a preload() class method for use in forking servers, and to
   work around limitations in PAR.

Bug fixes:

 - paletted writes to a masked image are now masked correctly.
   Revealed by new coverage tests.

 - update the filter plugin documentation.

 - add the matrix() method to Imager::Matrix2d to allow creation of a
   matrix with specified co-efficients.  You can now multiple an
   Imager::Matrix2d object by a 9 element array ref or a number.

 - really fix loading TTF fonts with FT2 when FT1 isn't available.
   Thanks to Leolo (Philip Gwyn)

 - make sure each test script that needs testout/ creates it.

 - handle a slightly different warning from libtiff 4.x

 - the sat transform2() op returned an incorrect saturation.

Imager 0.80 - 17 Jan 2011

 - added coverage tests for Imager::Fountain and Imager::Color::Float

 - Imager is now maintained in git

Bug fixes:

 - images with an translucent alpha channel were not scaled correctly
   by the default (qtype=normal) scaling method.

 - Imager::Color::Float now translates "#FFFFFF" to white instead of
   just a little darker.

 - make the default color map build algorithm "mediancut".  This
   changes the default color map builder for writing GIFs back to what
   it was in 0.77, reverting a performance regression.

 - handle failure to create a masked image correctly

Imager 0.79 - 10 Dec 2010

 - add Imager::Test to the POD coverage tests and document the missing

 - the convert() method now optimizes the case where all output
   channels are either 0, sourced from a single input channel or 1.
   This significantly speeds up presets like "addalpha", "green".

 - add sample, as suggested by Dan Oppenheim.

 - add the combine() method to combine channels from multiple source
   images into a new image

Bug fixes:

 - treat the co-efficients for convert() as doubles instead of floats.

 - If a higher (earlier) priority font handler failed to load, that
   would crash preventing loading of later font handlers.

 - parse defines from the options returned by pkg-config --cflags

 - a regen of the MANIFEST revealed that GIF and FT2 tests weren't
   included in the tarball.  They are now included.

Imager 0.78 - 4 Oct 2010

Bug fixes:

 - don't access deprecated members of the png_structp.

 - document that using color objects is faster than supplying colors
   by name, etc.

 - Imager::Probe now accepts array references for incpath and libpath.

Imager 0.77_02 - 27 Sep 2010

 - moved Win32, FreeType 2 font support into sub-modules. (partial)
   Uses Imager::Probe now.

 - tested successfully with jpeg-8b

Bug fixes:

 - from _01: look for missing file support test files in the right

 - flood_fill() wouldn't fill the right side of a single scan-line
   fill area.
   Thanks to Nicolas Roggli for reporting this.

 - flood_fill wouldn't fill to the left edge of the image if the
   starting line didn't reach the left edge.
   Thanks to Nicolas Roggli for reporting this.

Imager 0.77_01 - 13 Sep 2010

 - add each library-directory/pkgconfig/ to the pkg-config search path
   in Imager::Probe.
   Thanks to Justin Davis.

 - moved GIF, TIFF, JPEG file handling code into sub-modules in
   preparation for separate distribution. (partial)

   Note: this removed support for libungif from Imager.

 - optimize filled box drawing (color, not fill)

Bug fixes:

 - Imager::Probe was calling ExtUtils::Liblist to initialize
   LD_RUN_PATH supplying an undefined value rather than the found
   directory.  Thanks to Justin Davis.

 - only prepend ./ to font filenames when passing them to T1Lib and
   then only when it would use its search mechanisms.

 - fix the cache check for the X rgb.txt loader.  This is typically
   used for color translation on Unix-like systems, and the fix
   improves performance of supplying colors by name by about 80 times.
   Test code that managed 3400 10x10 pixel boxes/second sped up to
   25700 boxes/second.

 - clarify that Imager doesn't write EXIF metadata to images.

 - Imager::Probe can now search subdirectories under its include path.
   Used by the PNG Makefile.PL to find headers and libraries when they
   aren't in the base directory, as in cygwin.

Imager 0.77 - 11 Aug 2010

I don't want Imager::File::PNG indexed as part of Imager, but forgot
to update the META.yml before updating the version.

 - don't index Imager::File::PNG as part of Imager

 - add resources to META.yml

Imager 0.76 - not released

Bug fixes:

 - the align_string() method would ignore a string of "0" due to a
   mis-use of C< ||= >.
   Thanks to Maurice Height for reporting this.

Imager 0.75_03 - 09 Aug 2010

Bug fixes:

 - read_types() and write_types() would include png when it wasn't
   available due to a problem with the %formats tie

 - handle dependent libraries correctly (eg -lpng requiring -lz) in
   the code run phase of library probing.

Imager 0.75_02 - 07 Aug 2010

Bug fixes:

 - add file missing from MANIFEST, which was causing TIFF failures.

Imager 0.75_01 - 06 Aug 2010

Test release for the new PNG probe.

 - added the ability to read multiple-image PNM files.
   Note that unlike the pbm/pgm/ppm specification this accepts mixed
   format files and allows white space padding between files.
   Thanks to Philip Gwyn (Leolo) for this patch.

 - moved the PNG file handling code into a sub-module in preparation
   for separate distribution. (partial)
   Also helps avoid complications from -I/-L compile/link options from
   other libraries.


 - Imager->new(data => $data) didn't try to process image file data in

 - t/t50basicoo.t no longer depends on the other tests to generate its 
   input files.
   Also, it wasn't testing pnm (pnm vs ppm mix-up)

 - update the documentation of hardinvert to match the change in 0.62.

 - added hardinvertall filter which also inverts the alpha channel
   (sorry for the mess)

 - when probing for TIFF, set LD_RUN_PATH just as the Makefile does so
   the probe can find the library for the test run.

Imager 0.75 - 20 Jun 2010

 - use PERL_NO_GET_CONTEXT to slightly improve performance on threaded
   perls (not measured)


 - an opacity fill based on a fountain fill would segfault when
   filling an 8-bit/sample image.

 - merge thickline branch polygon fix

Imager 0.74 - 7 May 2010

Bug fixes:

 - read_multi() didn't handle a missing file format library correctly,
   aborting on failing to call i_readgif_multi_wiol() or

 - fix spelling errors patched by Debian

 - add an (unshipped) author test to spellcheck Imager's POD.

 - update the TIFF file format documentation

 - lib/Imager/IO.pod was written almost 4 years ago but never shipped.

Imager 0.73 - 15 Mar 2010

 - implement outline circles, both anti-aliased and not

 - a combine => "none" fill to a 1 or 3 channel image would produce
   the incorrect colour.

Imager 0.72 - 09 Dec 2009

Bump version for release, since 0.71_03 is stable with CPAN testers.

Imager 0.71_03 - 5 Dec 2009

 - further adjust the threads test so it only performs the tests on
   perls where it's expected to work, and only if the threads module
   can be loaded.

Imager 0.71_02 - 1 Dec 2009

 - adjust the way we load the threads module for the threads test so
   it works with non-threaded perls

Imager 0.71_01 - 30 Nov 2009

Bug fixes:

 - use scanline oriented operations to flip images instead of pixel

 - use double/sample operations to flip large sample images instead of
   8-bit sample operations.

 - fix POD nits

 - prevent double-frees when someone creates Imager objects and then
   creates a thread.  Note: this just handles some simple cases,
   Imager doesn't support perl threads, and isn't likely to.

Imager 0.71 - 16 Nov 2009

 - add the opacity fill type - an adaptor that modifies the opacity of
   another fill.

Bug fixes:

 - the conv filter now enforces that the sum of the coefficients is
   non-zero.  Also, rather than skipping pixels off the edge off the
   edge of the image, the closest edge pixel is used.  Previously
   dividing by the zero sum of coefficients could cause invalid
   results or runtime exceptions.
   Thanks to David Cantrell's Alpha-NetBSD CPAN test box for revealing
   this bug.

Imager 0.70 - 21 Sep 2009

Bug fixes:

 - release image row and comments memory on all error returns in gif

 - handle zero length extensions, previously this would cause a null
   pointer dereference
   Thanks to Krzysztof Wojtaś for the test data and fix for this.

 - an integer division meant that preview scaling to below 1 pixel
   wide or high (which isn't too useful anyway) was calculating using
   NaNs on most platforms, and causing an exception on others.
   Thanks to David Cantrell for producing a backtrace of the crash on
   his Alpha-NetBSD CPAN test box which made it possible to track this

Imager 0.69 - 08 Sep 2009

Bug fixes:

 - broken test fix - was attempting to call a function skip_all, when
   that should be a parameter to plan().

 - briefly document apidocs.perl, the tool used to build
   Imager::APIRef and make some minor enhancements

 - various minor documentation enhancements and fixes.

Imager 0.68 - 07 Sep 2009

 - Imager->new(file => $filename) and other similar incantations will
   load the given file.

Bug fixes:

 - avoid using CHECK as a label in Imager::Test

 - re-work most image file test files that require a library into
   separate library present/not present files to remove stupidly long

 - don't treat rubthrough() outside the bounds of the target image as
   an error.

Imager 0.67_01 - 02 Sep 2009

Bug fixes:

 - correct documentation of default of raw image interleave read

 - add raw_ prefix to raw read parameters, though the original names
   still work.

 - fail the read if an invalid raw_interleave parameter is supplied

 - warn if no interleave or raw_interleave parameter is supplied,
   since the documented default was wrong, and incompatible with the
   write format

 - for reading raw images, if raw_storechannels > raw_datachannels,
   set the extra channels in the image to 0

 - when probing for executables like freetype-config, search for .bat
   and .cmd on MSWin32, as well as .exe.

 - re-work the external libraries section of README:
   - list Debian and Redhat package names for each library
   - reformatting
   - update URLs

 - use the new EU::MM META_MERGE facility instead of generating
   META.yml from scratch

 - use Devel::CheckLib (bundled, modified) to check which release of
   libtiff is installed and reject 3.9.0

Imager 0.67 - 12 Dec 2008

Bug fixes:

 - fix a packaging error

Imager 0.66 - 12 Dec 2008

 - 24-bit color .ICO/.CUR files can now be read.

Bug fixes:

 - an optimization skipping 0 src alpha values could cause the
   rubthrough() to read past the end of a buffer.

 - corrected a reference leak where writing GIFs would leak memory.
   This could also happen calling to_paletted().
   Also documented the underlying long existing feature where the
   colors parameter is filled with the generated color table and added
   tests for it.

 - write out the image size in bytes field of a BMP correctly.

 - add limited tests for Imager::ExtUtils

 - make Imager::ExtUtils->includes use an absolute path, since
   a relative path could cause failures using Inline::C.

 - re-arrange the POD for Imager::Font::BBox:
   - mark total_width(), pos_width(), end_offset() obsolete, since
     they're mostly for backwards compatibility
   - group width methods and height methods

Imager 0.65 - 20 May 2008

Bug fixes:

 - In some cases when an error occurs reading those parts of a JPEG
   file after the image the scan-line buffer could be freed a second
   time.  In cases where the the error occured while reading the image
   data it's possible that the buffer could have leaked.
   Thanks to Gabriel Vasseur for reporting this and help in tracking
   it down.

 - the gif_screen_height tag was overriding the screen width and being
   ignored for the screen height when present.

Imager 0.64 - 23 April 2008

This is a bug fix release.  This includes a fix for a possible
security issue.

Bug fixes:

 - Possible security issue: The floating point sample path for image
   based fills had a buffer overflow.  This would overwrite the end of
   a malloc()ed buffer with double precision floats.

 - check that the result of fileno($fh) is defined rather than simply
   true when read() or write() is supplied with an fh parameter.

 - i_scale_axis() wasn't checking the result of i_img_new_ch()
   resulting in a SIGSEGV when attempting to scale an image to a size
   too large to fit in memory.  This is a NULL pointer access issue,
   not a buffer overflow.
   Added a check for the failure.
   scale_calculate() (and hence scale()) will now fail if any of the
   scale size parameters are a reference.

 - Regression: filling a greyscale image with a hatch used the wrong
   color channels from the supplied fg/bg colors.

 - fixed a related problem for image fills.

Imager 0.63 - 7 April 2008

This release primarily contains changes to improve ease of use -
rather than you having to convert images to the appropriate number of
channels, Imager handles it internally.  How to handle drawing colors
and the default combine mode is a thornier problem left for some other

 - the font libraries are now only initialized when needed.

 - moved the imtoc.perl code into Imager::Preprocess

 - paste() and rubthrough() now adapt the source image data to the
   destination, so you can now safely paste/rubthrough from greyscale
   images to color images or back, or from alpha channel images to
   noalpha channels or back.

 - rubthrough() now falls back to pasting when the source doesn't have
   an alpha channel.  This effectively treats the source as having a
   max alpha channel, the right thing to do.

 - re-worked most of the area filling code to use a common set of
   functions when filling.
   Corrected normal combine mode.
   Rewrote most of the combine modes to match the way the SVG draft
   defines them with respect to a translucent source and destination.
   Added tests for translucent source and destination.
   Added tests to check 8-bit/sample and double/sample combines work

 - writing a 2 or 4 channel image to a JPEG file will now write that
   image as if composited against a background, black by default,
   overridable with the i_background tag/parameter.

 - writing a 2 or 4 channel image to a PGM/PPM file will now write
   that image as if composited against a background, black by default,
   overridable with the i_background tag/parameter.

 - writing a 2 or 4 channel image to a BMP file will now write that
   image as if composited against a background, black by default,
   overridable with the i_background tag/parameter.

Bug fixes:

 - Imager::Matrix2d->translate() now only requires one of the x or y

 - mixing qtype scaling now sets all channels of a pixel to zero if
   the pixel has zero coverage (zero alpha).  This should produce more
   compressible output files.

 - removed the pointless #! line from lib/Imager/Font/
   Noticed when I saw:
   I'm not changing the #! lines of the sample code, since it's sample
   code, not intended for installation.

 - some TGA images weren't being detected correctly as TGA images

 - handling of the left-over bit for 16-bit/pixel TGA images has been
   changed to match the behaviour of the GIMP.  Previously the bit
   being set was treated as an opaque pixel, but one user reported a
   problem with loading such an image.  I haven't been able to find any
   tools beyond the GIMP that handle alpha-channel 16-bit TGAs, so
   I'll match it's behaviour.  See issue 114913 in the GIMP's

Imager 0.62 - 10 December 2007

 - Makefile.PL now expands ~/path supplied to --incpath or --libpath
   to /path under your home directory.

 - the old dynaload code used Mach API functions to load dynamic
   libraries on Mac OS X.  These APIs have been deprecated in OS X
   10.5 and were causing some build problems.
   So henceforth Imager uses the dlopen() family of functions, and you 
   will need version 10.3 or later of OS X.

 - added the det() function to the transform2() engine.
   added the sample
   Courtesy Richard Fairhurst.

Bug fixes:

 - samples/ sourced the base value for gif_top from
   Thanks to Eleneldil G. Arilou for pointing this out.

 - t/t82inline.t no longer loads B at runtime, to work around a bug
   in some 5.005_0[45] installations.

 - work around Module::Depends::Intrusive bug #21229

 - the hardinvert filter no-longer inverts the alpha channel.

 - the hardinvert filter now supports large samples

Imager 0.61_02 - 28 November 2007

 - major TIFF support re-work

 - added a C level image interface for accessing samples from 1-32
   bits, exposed this at the perl level in getsamples()

 - the conv filter now works at floating point precision for high bit

 - added is_bilevel method to test whether an image should be written as
   a bilevel image if the image format supports it.

 - added -log-stderr as an Imager import list option

 - added some important types to Imager::APIRef

 - added test_image_double() to Imager::Test

Bug fixes:

 - Imager::Fountain couldn't read GIMP gradient files with 10 or more

 - the scale() method with qtype mixing now handles images with an
   alpha channel correctly.

 - fixed a broken link from the "animated GIF" entry in the concept index.
   Thanks to Slaven Rezic.

 - on some perl's the infix expression parser test would fail due to
   actions in the grammar returning false.  Made sure all actions return
   a true value.
   Thanks to Richard Fairhurst for spending a lot of time in tracking
   down this problem.

Imager 0.61 - 5 November 2007

 - added samples/, which adjusts the screen size/position tags
   when scaling an animated gif

Bug fixes:

 - correct handling of sz in matrix_transform() - this should allow
   perspective type transformations to work now.

 - prevent a cast to integer warning on x64 builds in datatypes.c
   also fixed some other type warnings

 - some sub-directory tests depended on files produced by the parent
   directory tests

 - Imager::Font::Wrap doesn't correctly set savepos
   thanks to Nikita Dedik and Eleneldil G. Arilou for reporting this.

 - test 171 in t/t01introvert.t was failing on perls configured to
   use long double.

 - the code for the transform2() uminus operator was missing a break.
   Added tests for better code coverage of the ops.

 - the SGI RLE compression code could overflow its compression buffer

 - the 32-bit output function used by the SGI code only handled values
   under 0x10000.  This was most noticable when writing large RLE images.

 - validate chan_count for chans == NULL for each of the i_gsamp()

 - attempt to work around the test failure at

 - improve the error messages produced when attempting to read or write
   an unknown image file format.

 - improve the transform2() documentation

 - correctly generate the author key in META.yml

 - correctly blend a rotated (or matrix_transformed()) image when
   performing interpolation in the presence of an alpha channel.
   Also corrected the centring of the rotated image on the output

Imager 0.60 - 30 August 2007

 - Finished/rewrote Arnar's old SGI RGB file format support, so Imager
   now has full SGI RGB image format, including RLE and 16-bit/sample

 - logging functions are now available in the API

 - applied Gabriel Vasseur's patch
   added documentation, further tests, and support for greyscale images
   Obviously problems are my fault :)

 - the mask for ICO/CUR images is now applied as an alpha channel to
   the returned image.  For the old behaviour, supply ico_masked => 0
   to read() or read_multi().  This should be less confusing when
   using Imager as a general image processor.

Bug fixes:

 - in some cases it's possible for giflib/libungif to return color 
   indexes outside the range of colors defined by the image's palette.
   We now expand the palette to match the indexes used.
   Thanks to Gabriel Vasseur for reporting this.

 - fixed various memory leaks that could occur when failing to read png,
   jpeg, bmp or tga files.

 - to avoid confusion, channels not present in the image are returned as
   zero by getscanline().  This has no effect on the C level i_glin()
   and i_glinf() API functions which continue to not set the unused

 - the convert() method now returns an image of the same sample size as
   the source image.

 - remove repeated text in Imager::Files

 - be even more explicit that scale() and friends don't modify the source
   image, but return a new image.

 - improve the error message from errstr() when you try to load a font
   for which the driver hasn't been built in Imager.

 - transparency is now enabled by default when writing GIF images

 - Imager would not load on Windows 98

Imager 0.59 - 14 June 2007

Bug fixes:

 - fixes a regression introduced by the fixes for RT 11972

 - cropping outside the image would return an Imager object with
   no low-level image object, instead of returning false.
   Fixed by: Philip Gwyn (Leolo)

Imager 0.58 - 16 May 2007

No significant changes from 0.57_01.

Imager 0.57_01 - 11 May 2007

 - added to_rgb16 to produce a 16-bit/sample version of an image

 - improve freetype 1.x text output efficiency

Bug fixes:

 - search another place for rgb.txt, and check all the places 
   Imager::Color checks when deciding whether to skip testing it

 - use a convolution kernel size based on the stddev rather than a
   fixed size when performing a gaussian blur

 - document the difference() method's mindist parameter, and debug it.

 - put the Imager release number in the Inline::C generated code to
   regenerate Inline code when a new release of Imager is installed.

 - fix rendering on alpha channel images for the FreeType 1.x driver.

 - fix rendering on alpha channel images for the T1lib driver.

 - reworked library probing, we can now set more than one probe
   function for a library.  Disabled the default (non-freetype-config)
   library probe and added an extra probe function that searches for
   both ft2build.h and whatever it includes, and adds -I as needed.
   Hopefully this will fix build problems like

Imager 0.57 - 30 Apr 2007

This is a maintenence release fixing a security issue in Imager.

 - CRITICAL: a specially crafted compressed BMP file can cause a buffer
   overflow in malloced memory.  There will be further discussion of
   this issue in the ticket below.
   CVE-2007-2459  CVE-2007-2413
   The descriptions at varied in quality, please see the
   ticket at for a more accurate description of the issue.

Imager 0.56 - 1 Apr 2007

 - added support for reading 16-bit/sample PGM/PPM images

 - added support for writing 16-bit/sample PGM/PPM images

 - improved performance of reading PBM/PGM/PPM images

 - added support for writing PBM images if the image is paletted and
   contains only black and white

 - added a new make_colors value - "mono"

 - switched from the svn log Changes to a manual Changes to reduce

 - new sample code - samples/

Bug fixes:

 - CRITICAL: the "Imager" typemap entry (not used by Imager itself)
   was returning an image object with an extra reference, this
   resulted in a memory leak.

 - fix rendering on alpha channel images for the FreeType 2.x driver

 - reading bmp files now consitently handles short reads.  You can now
   supply a parameter to treat a short read as successful and set

 - previously, reading ASCII PBM files required spaces between samples,
   even though the format doesn't require that

 - improved documentation of the unsharpmask filter (I hope)

 - force flushing of the output from i_tt_dump_names() and test output
   in t/t35ttfont.t to prevent output from being mixed up.

 - rewrite a conditional expression as an if() to hopefully work around
   a bug in the pre-4.0 GCC Apple shipped with OS X 10.4.

 - avoid Data::Dumper in regops.perl to support older releases of perl

Imager 0.55 - 16 Dec 2006

This is primarily a bug fix release.

Note: Test::More is now a pre-requisite for Imager and is no longer bundled.

There is one new feature:

 - the Win32 font driver now supports UTF8 (RT 22166)

Several bugs were fixed:

 - the string() method would not output the string "0"

 - fills.c was failing to compile on Solaris 10 (compiler unknown)

 - the gif_disposal and gif_user_input tags weren't being read from
   the file correctly

 - gif.c was failing to build under MSVC

 - in some cases strings passed to the string() method were treated as
   terminated by NUL (chr 0)

 - on "MSWin32" perl builds we now link to -lzlib instead of -lz since
   that's the default build name for zlib on Win32.

 - search $Config{incpath} for headers too, which we should have been
   doing all along.

Win32 font driver fixes:

 - the global descent value from bounding box was the wrong sign

 - if the first or last glyph overflowed the left or right side of the
   advance width they would be clipped

Imager 0.54 - 14 Sep 2006

This is primarily a feature release:

 - a new qtype value 'mixing' has been added to the scale()
   method. This is faster than 'normal', slower than 'preview'. This
   is based on the method used by pnmscale, and seems to produce less
   blurry results than normal.

 - the rubthrough() method can now render onto images with an alpha

 - the read_multi() method now falls back to calling doing a single
   image read via the read() method and write_multi() will now fall
   back to calling write() if a single image is supplied. This means
   you can simply call the read_multi() or write_multi() functions
   without having to check if the type is formatted by that method.

 - the GIF loop extension can now be written. If you don't have
   libungif/giflib 4.1.4 (or some distribution's bugfixed equivalent) you
   should upgrade.

 - getscanline() and setscanline() can now read/write palette index
   based data from/to the image for paletted images, by setting type to

 - we no longer hassle you to disable GIF support

 - minor documentation fixes

Imager 0.53 - 26 Jul 2006

This is a bugfix release.

Some test code was left in a code path not covered by the test
suite. A test was added to cover this code path and the test code was

Imager 0.52 - 25 Jul 2006

This is primarily a feature release, but contains a fair few bug
fixes, new features:

 - ability to read and write MS Windows ICO and CUR files

 - you can now add file format plugins to support new file formats

 - add POD coverage tests

 - setcolors() and addcolors() now accept color names and so on
   instead of requiring Imager::Color objects.

 - flood_fill() can now fill to a specified border color instead of
   just to the area the same color as the seed.

Bug fixes:

 - bounding_box for the T1 driver wasn't converting UTF8 to ascii when
   calculating the advance width.

 - bounding_box for the T1 driver wasn't including leading and
   trailing spaces in the bounding box as the other drivers did, it also
   produced strange results for empty strings or strings containing only

 - when reading CMYK jpeg images they were being transferred to the
   image object as is, producing a four channel image. It only looked ok
   due to an old still unfixed Photoshop bug. We now convert from the
   inverted CMYK that photoshop (and Corel for example) produce into RGB.

 - reading a CYMK TIFF would result in a 4 channel image, reading any
   image with more than 4 channels (eg. RGB with 2 alpha channels) would
   result in an error.

 - added /usr/local/include to the default include search path, since
   we were already searching /usr/local/lib for libraries.

And various minor fixes and documentation updates.

Imager 0.51 - 23 Apr 2006

 - fix a validation bug when processing JPEG EXIF data that can cause
   a crash

 - fix mis-processing of the src_maxx and src_maxy parameters of the
   paste() method

 - fix a problem in Imager's "smart" handling of the color parameter
   to various methods.

Imager 0.50 - 29 Mar 2006

 - CRITICAL: fixes a segmentation fault from attempting to write a 2
   or 4 channel image to jpeg or a 2 channel image to tga where the
   output is an in-memeory buffer.

 - fixes an incorrect pointer parameter in the PNG code

 - skip Inline::C tests when building in a directory with spaces