Imager

Changes for version 0.44

modify the default include directories list to include the location FreeBSD keeps freetype1 headers after the freetype2 locations. This ensures that the -I options are generated with freetype2 locations first, so that those directories are searched before the freetype1 directories. This prevents problems compiling font.c on FreeBSD. Resolves: http://rt-cpan.fsck.com/Ticket/Display.html?id=1723
finish off a sentence in the "Producing an image from a CGI script" example in Imager::Files
method index didn't include errstr()
document that you don't want the FT2 freetype.h in the include path
- For latest versions check the Imager-devel pages: http://imager.perl.org/

Changes for version 0.43_03

change the "double-include" protection macro that imio.h uses.
updated download locations for the various libraries that Imager depends on. Added some advice for cygwin.
more information on gif library versions in README and Makefile.PL
creating an image where the size of the allocated image buffer would overflow an integer would cause too small a buffer to be allocated. This could potentially be a security hole. partly resolves https://rt.cpan.org/Ticket/Display.html?id=8213
set i_format to jpeg for jpeg files and test for it
set i_format to png when reading png files and test for it
i_yres was being set to the xres when reading a png file
added many bad BMP files to test various code paths in bmp.c, and made many minor fixes to bmp.c to make them work:
- it was possible for various types of read failures to SEGV, both as NULL pointer dereferences and buffer overflows
- some errors, like palettes being too large for the number of bits per pixel, were not being caught
- failing to read all of a packed data structure would not cause a read failure
- invalid compression types were not always caught
- error messages are more consistent (still not always great messages, but one step at a time)
added bmp_compression_name, bmp_used_colors, bmp_filesize, bmp_bit_count tags on reading a BMP file
added tools/imager to the distribution. This is still very experimental and untested. Patches welcome, if you write tests to go with them.
the BMP reader now validates the bfOffBits value from the BMP header and skips to that offset before reading image data. Previously this value was read but otherwise ignored.
added --palette action to tools/imager
i_img_pal_new() now releases the image object memory if creation fails.
set i_format to gif when reading gif files and test for it
set i_format to pnm when reading pnm files and test for it
set i_format to tga when reading tga files and test for it
handle image creation failures when reading tga files
set i_format to tiff when reading tiff images and test for it
handle image creation failures when reading tiff files
test t/t35ttfont.t no longer requires TTFONTTEST to be set (and hasn't for a long time,) removed that requirement from README
updated home page URLs throughout the .pods
added information on reporting bugs to the SUPPORT section of Imager.pm
regops.perl now sorts the dumped data structure to produce minimal diffs
quant.c now checks for integer overflow allocating its image data
i_readraw_wiol() now checks for image creation failure
i_readrgb_wiol() now checks for image creation failure
i_writergb_wiol() was an empty stub, it now pushes an error message and explicitly returns failure.
i_readrgb_wiol() now sets i_format to rgb.
set i_format to raw when reading tga files and test for it
document i_format tag
some tests were using $Config{ivsize} when they should have been using $Config{intsize} Resolves: https://rt.cpan.org/Ticket/Display.html?id=8645
tools/imager has been removed from the MANIFEST, it's way too late to test/debug for 0.44.
image.h had no prototype for i_ft2_bbox_r() and it was being called from Imager.xs
giflib 4.1.3 still doesn't have all the bugs fixed, update the notes in README and in the buggy_giflib.txt file that t105gif.t produces
make the inclusion of NO_META dependent on the version of EU::MM. I was going to include a META.yml but EU::MM's generation is too simplistic (and misses the leading document header too). For now I'll leave it out.
built 0.43_03 for testing

Changes for version 0.43_02

the changes to scale() had some problems with integer vs floating point calculations (only caught in tests under perl 5.8.5 <sigh>)
the FT2 glyph_names() method didn't do correct error handling when the string parameter wasn't supplied
i_ft2_glyph_name() accepted only an unsigned char as the character code to get the name for, which meant it didn't work for unicode characters \x{100} or above
the XS for i_ft2_glyph_name() had a similar problem
added NameTest.ttf to be used in checking unicode glyph names
added reliable_only optional parameter to the glyph_names() method so you can ignore theresult of FT_Has_PS_Glyph_Names()
handle errors given by i_ft2_glyph_name() a bit more correctly
the FT1 glyph_names() method didn't do correct error handling when the string parameter wasn't supplied
some memory allocated when creating a callback IO object (io_new_cb) wasn't being released (detected with valgrind)
the testtools.pl match[nx]() functions escapes the test string on test failure a bit better
the XS code for i_tt_glyph_name() used unsigned char to store a unicode character when it should have used unsigned long.
the XS code for i_t1_glyph_name() used unsigned char to store a unicode character when it should have used unsigned long.
resolves https://rt.cpan.org/Ticket/Display.html?id=7949
the type 1 glyph_names() method didn't do correct error handling when the string parameter wasn't supplied
renamed io.h to imio.h to prevent problems building under cygwin. resolve https://rt.cpan.org/Ticket/Display.html?id=7948
i_writegif_low() wasn't setting the returned global palette, which meant a rubbish palette was returned to the user (detected with valgrind)
built 0.43_02 for testing

Changes for version 0.43_01

only call FT_Get_Postscript_Name() on FT 2.0.6 and later
put the IM_LIBPATH and IM_INCPATH values first in the search path so the builder gets their local versions if desired rather than the system versions they might be trying to avoid
document the exp() and log() transform2() functions
document the constants normally set by transform2().
refer the user to appropriate documents in the example in Imager.pm
change the list of documents in Imager.pm to move the document names out of the =item lines so we can make them into links
the scale() method now produces a warning when called in void context.
font.c now only uses the defined T1Lib error codes
update ppport.h and remove the duplicate definitions from Imager.xs. Had to mangle ppport.h to prevent duplicate global function definitions.
newer versions of tifflib require that all of the function pointers passed to TIFFClientOpen be non-NULL, the mmap() and munmap() pointers were always NULL and the sizeproc was sometimes NULL.
there would be a link or load time error on m_init_log() if Imager was built with IM_NOLOG, fixed by renamed init_log() to m_init_log() in log.c (thanks to Takumi Yamani)
moved some variable definitions to the right place
the Win32 font driver bounding box function now returns the right number of values (both thanks to Takumi Yamani)
the Win32 font driver now uses DEFAULT_CHARSET rather than ANSI_CHARSET are the lfCharSet value for the LOGFONT, as suggested by Takumi Yamani.
fontfiles/ExistenceTest.{pfb,ttf} weren't marked as binary in the CVS repository (caused test failures if you built from CVS on Win32)
Makefile.PL should now handle INCLUDE or LIB with spaces in them correctly on Win32.
the pnm reader read maxval for ppm/pgm files and then ignored it, it's now validated (0 < maxval < 65536) and used to scale samples. Note that binary ppm/pgm files (P6/P5) with maxval > 255 result in an error, since I didn't want to add new features just yet, just get the code that's there working correctly. Thanks to Elthek on rhizo for reporting this and help in tracking it down. Resolves https://rt.cpan.org/Ticket/Display.html?id=7465
added a bunch of tests for reading pnm files.
previously, if you supplied to_paletted and empty color map and set make_colors to 'none', quant_translate() would segfault. This was because it was making the reasonable assumption that you'd have colors to map to. quant_translate() now checks there is at least one color and return NULL if there isn't.
i_img_to_pal() now does error checking of the value returned by quant_translate().
Imager::to_paletted() now checks for success/failure of i_img_to_pal() correctly and does appropriate error handling.
i_writegif_low() did no error checking on the result of quant_translate(), it now does
we now test that trying to write a GIF image with no palette allowable by the quant options is a failure.
Imager::write() was doing nothing with the result of the call to i_writegif_gen(), in particular it wasn't returning () on failure.
added tests for paletted image handling and the methods specific to those images
the XS for i_setcolors() was missing the OUTPUT clause for RETVAL, and hence wasn't returning failure on failure.
supplying a sufficiently small scaling value could make the scale() method return an image with zero height or width. Several of the above together resolve https://rt.cpan.org/Ticket/Display.html?id=7467
the void context warning for scale() now includes the callers filename/line (instead of the default of Imager.pm line 15xx)
Imager->new will now return undef if the dimensions or number of channels specified for an image are out of range. An error message can be retrieved with Imager->errstr.
added the C<builtin> color specifier and the Imager::Color::Table class which defines those colors. Resolves https://rt.cpan.org/Ticket/Display.html?id=2593
added the equals() method to Imager::Color. Resolves https://rt.cpan.org/Ticket/Display.html?id=2238
prevent a test warning introduced by the above
the rotate() and matrix_transform() methods now accept a 'back' parameter specifying a background color. This is only used where there is no source data (think of the corners around an image rotated 45 degrees) and it not combined with transparent pixels from the source image. Resolves https://rt.cpan.org/Ticket/Display.html?id=6140
removed a warning generated by the new matrix_transform() test
added a method index to Imager.pm
corrected "flood fill" to "flood_fill" in Imager/Draw.pod
removed compose() method from Imager/Transformations.pod since it isn't implemented yet
the image resulting from a crop is now the same type as the source image (paletted vs direct, bits/sample) Resolves https://rt.cpan.org/Ticket/Display.html?id=7578
the parameters to crop() weren't handled correctly in most cases other than supplying left,top,right,bottom.
clarified the documentation for crop() providing more detail and more examples
the edges of the cropped area are now cropped against the edges of the source image
cropping to zero width/height is treated as an error (no image is returned and $src->errstr has a message) Resolves https://rt.cpan.org/Ticket/Display.html?id=7581
built 0.43_01 for testing