.. -*- mode: rst -*- .. This text is in reStucturedText format, so it may look a bit odd. .. See http://docutils.sourceforge.net/rst.html for details. =================== GraphicsMagick News =================== This file was last updated on November 7, 2015 Please note that this file records news for the associated development branch and that each development branch has its own NEWS file. See the ChangeLog file for full details. .. contents:: :local: 1.3.23 (November 7, 2015) ========================== .. _`GCC bug 53967` : http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53967 Special Issues: * Due to `GCC bug 53967`_, several key agorithms (e.g. convolution) may execute much faster (e.g. 2-3X) for x86-64 and/or when SSE is enabled for floating point math (`-mfpmath=sse`) if the GCC option `-frename-registers` is used. Default 32-bit builds do not experience the problem since they use '387 math. It is not clear in what version of GCC this problem started but it was not noticed by the developers until the GCC 4.6 timeframe. Other compilers do not suffer from this bug. Please lobby the GCC project to fix this embarrassing performance bug. Security Fixes: * ScaleImage(): While not strictly a security issue, requesting to scale an image while retaining the original number of rows will lead to a program crash or memory corruption due to double-free. Bug fixes: * ScaleImage(): Fix problem with new width/height match original (regression added by 1.3.22). * ScaleImage(): Fix double-free when new rows matches original rows (regression added by 1.3.22). * MinGW build fix related to eliminating a sleep() macro which conflicts with a MinGW-provided inline sleep() function. * PNG: Issue a warning instead of an error when attempting to read a PNG file containing a zero-length profile. This allows the file to be read. * identify: Fix problem in that `identify -format "%A"` (to test if transparency is supported in image) does not always produce the correct results. New Features: * None. Feature improvements: * None. Performance Improvements: * None. Windows Delegate Updates/Additions: * None. Build Changes: * Configure: Removed CFLAGS and LDFLAGS deduplication code which caused problems for user-provided CFLAGS and LDFLAGS which added and then removed compiler/linker options. Specifically, this fixes a problem with creating OS X universal builds. * Configure: Add tests for 'ps2write' and 'eps2write' which are available in recent Ghostscript. Behavior Changes: * None 1.3.22 (October 4, 2015) ========================== .. _`GCC bug 53967` : http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53967 Thanks: * Coverity: We thank Coverity for providing free service for free software projects, and thank Jodie Cunningham for getting the project set up in Coverity. Special Issues: * Due to `GCC bug 53967`_, several key agorithms (e.g. convolution) may execute much faster (e.g. 2-3X) for x86-64 and/or when SSE is enabled for floating point math (`-mfpmath=sse`) if the GCC option `-frename-registers` is used. Default 32-bit builds do not experience the problem since they use '387 math. It is not clear in what version of GCC this problem started but it was not noticed by the developers until the GCC 4.6 timeframe. Other compilers do not suffer from this bug. Please lobby the GCC project to fix this embarrassing performance bug. * Magick++: Any libraries or applications using Magick++ should be rebuilt in order to use this new release. Libraries and applications will be able to continue to use prior versions of Magick++ without being re-built, while benefiting from updated C libraries, provided that the system supports library versioning. Security Fixes: * General Coverity fixes. Some might have security consequences. * Ghostscript options concatenation is more secure against buffer overflow. * Windows: Built-in random number generator is now salted using CryptGenRandom(). This improves the robustness of the temporary file allocator. Bug fixes: * Coverity Fixes: Large amounts of fixes due to Coverity static analysis. See the ChangeLog and Mercurial for details. Coverity now reports zero issues. * General: Fix problems with reading filenames that include a colon. * General: Fixed performance problem with sub-image path extraction when there are many files in the directory. * General: Add missing options in utility help messages. * BMP: Reader was wrongly rejecting RLE-compressed files as being too small (regression added in 1.3.21 release). * BMP: Fix inverted alpha channel when writing BGRA8888 format. * DrawAffineImage(): Fix problem with negative x offset. * DrawAffineImage(): Fix problem that sometimes output rows are skipped when using OpenMP. * EXIF: Properly validate GPS_OFFSET. * -format: %Q now reports JPEG quality estimate if it is available. * -geometry: Fix handling of area geometries in the form "5000000@". * MagickGetImageGravity(): Prototype was missing in header files. * MIFF: Memory leak fixes. * MIFF: MIFF reader failed to read some MIFF headers properly. * MIFF: Detect buffer overrun attempt while reading zip compressed data. * PDF: Set image frame scene ids appropriately. * PNG: Memory leak fixes. * PS: Set image frame scene ids appropriately. * PTIF: Mark reduced frames as SubfileType 0x2 instead of 0x1. * SetImageProfile(): Avoid crash given NULL profile pointer. * TIFF: Fix reading Old JPEG and YCbCr sample images from libtiff pics-3.8.0.tar.gz image file collection. * TIFF: Disable matte channel for compression types which don't support it. * XPM: Memory leak fixes. * XWD: Memory leak fixes. New Features: * GRAYA: New subformat for gray coder which supports alpha channel. Format specifiers "R", "G", "B", "A", "C", "M", and "Y" may now be used to save and restore the associated channel using the same raw format as "GRAY". * Magick++: Image::repage() method added to support resetting 'page'. * PDF: Added '-define pdf:stop-on-error=true' optoin to cause PDF reading to quit immediately upon any error. * Subframe specification: Now specific PS and PDF pages may be selected, including re-ordering. Feature improvements: * PALM: Still a work in progress. Closer to working using netpbm's implementation as a reference. Performance Improvements: * None. Windows Delegate Updates/Additions: * dcraw: Update bundled dcraw to release 9.26.0. * lcms: Update bundled lcms2 to release 2.7. * png: Updated bundled libpng to release 1.6.17. * tiff: Update bundled libtiff to release 4.0.6. * ttf: Update bundled freetype to release 2.6. * webp: Updated bundled libwebp to release 0.4.3. * libxml: Update bundled libxml2 to release 2.9.2. Build Changes: * lcms ("Little CMS") v1 is no longer supported. * VisualMagick: Remember and re-use already given paths. Behavior Changes: Magick++: adaptiveThreshold() now accepts a 'double' value and the previous version of the method (using 'unsigned int') is deprecated. The STL function-object equivalent of the deprecated method is removed entirely. 1.3.21 (February 28, 2015) ========================== .. _`AddressSanitizer` : https://code.google.com/p/address-sanitizer/ .. _`Valgrind` : http://www.valgrind.org/ .. _`American fuzzy lop` : http://lcamtuf.coredump.cx/afl/ Thanks: * Gynvael Coldwind and Mateusz Jurczyk of the Google Security Team provided test files which allowed us to find and fix security problems in the software. * Hanno Böck provided test files which allowed us to find and fix security problems in the software. * Tobias Ospelt provided test files and advice which allowed us to find and fix security problems in the software. * Michal Zalewski provided test files which allowed us to find and fix security problems in the software. * Jodie Cunningham did lots of fuzzing to find issues and set up the project on Coverity for automatic analysis. * `American fuzzy lop`_ was used to produce and discover many of the files which caused problems for the software. * `AddressSanitizer`_ (ASan) was used to detect and isolate memory access issues. * `Valgrind`_ was used to detect and isolate memory access issues as well as memory leaks Special Issues: * Due to `GCC bug 53967`_, several key agorithms (e.g. convolution) may execute much faster (e.g. 2-3X) for x86-64 and/or when SSE is enabled for floating point math (`-mfpmath=sse`) if the GCC option `-frename-registers` is used. Default 32-bit builds do not experience the problem since they use '387 math. It is not clear in what version of GCC this problem started but it was not noticed by the developers until the GCC 4.6 timeframe. Other compilers do not suffer from this bug. Please lobby the GCC project to fix this embarrassing performance bug. * Magick++: Any libraries or applications using Magick++ should be rebuilt in order to use this new release. Libraries and applications will be able to continue to use prior versions of Magick++ without being re-built, while benefiting from updated C libraries, provided that the system supports library versioning. Security Fixes: * Annotate: Some requestable text-substitution attributes caused a crash. * All formats: Image dimensions are checked to assure that they are within limits before proceeding to read the image. * BMP: Fix hang (endless loop) for certain files. * DCM: Fix crash as well as small heap over-write. * DPX: Fix crash due to DPX file reporting more elements than it has. * MNG: Validate MHDR chunk length to avoid huge memory allocation and DOS. * PCX: Fix for CVE-2014-8355. Validate file header in order to avoid buffer overun later. * PDB: Detect arithmetic overflows when calculating buffer sizes. Fix crash in writer when image width is not even multiple of 16. Fix buffer overrun with 2 and 4-bit PDB image files. * PNM: Validate PGM, PPM, and PAM header MaxValue parameter to avoid crash on poorly-formed input. * PNG: Impose a 10-million limit on dimensions when reading a PNG file to avoid denial of service. * PSD: Avoid problems caused by huge PSD colormap size. * PSD: Fix small stack over-write if more than 99 layers are written to PSD format. * PSD: Returns immediately if pixel limit was exceeded. * RLE: URT RLE reader is now more robust with errant files. * SUN: Header validation is now made fully robust, and arithmetic overflows in buffer-size calculations are detected to avoid heap overwrite. * TIFF: Fix crashes for photometrics which may deliver one or three samples per pixel (was assuming always three). * VIFF: Fixes to prevent buffer overflow. Validate colormap indexes. * Windows delegates: Fix unexpected argument splitting when invoking an external delegate program via delegates.mgk. * WPG: Fix use of NULL pointers. Fix buffer overflows. * XPM: Detect truncated row and quit with error rather than over-running a buffer. * XWD: Improve header validation. Added to UnstableCoderClass since the reader for this format should not be entrusted with untrustworthy input. Bug fixes: * CIN: Fix problem with text attribute values which are not NULL terminated. Validate sizes claimed by Cineon header. * Coverity: Fixes for many issues detected by Coverity scan (see ChangeLog). * DPX: Fix problem with text attribute values which are not NULL terminated. * DPX: Fix severe corruption of little-endian 32-bit packed output. Corruption was severe enough that it would have been noticed immediately. * Delegates: Fix possible memory leaks when invoking external application. * FITS: Properly validate values provided by file header. * GIF: Fix use of uninitialized data. * JBIG: Fix memory leaks. * JNG: Fix double-free error in error path. * JPEG: Verify the number of output components before attempting to decode the image. * Magick++: Image resolutionUnits() was not always returning correct value. * Magick++: Locking has not been working properly since the code was written in 1998. Apparently the issue has not been significant enough to cause run-time issues. * ICO: Windows icon reader is now much more robust. * MIFF: Reader now quits with an error if zip or bzip2 stream is corrupted. * MAT: Fix memory leaks. * PALM: Reader now reads various input formats (up to version 2) correctly whereas it was crashing or otherwise malfunctioning before. More work remains, particularly in the writer. * PCX: Eliminate memory leaks in error paths. * PDB: In PDB writer, void possible under-allocation due to arthimetic overflow when allocating packets. * PICT: Fix PICT reader crash with corrupted file. * PNG: Fix double-free error in error path. * PNG: Fixed handling of transparency when writing indexed PNG. * PNG: Avoid reading beyond the end of a tEXt keyword. * PSD: Fix error when reading PSDs files which have no layers. * RLA: Fix possible crash due to file header. * Signal Handling: Signal handling is now more robust and handles SIGSEGV and other critical signals. The sole purpose of the default signal handling is to remove any temporary files and quit. An informative message is printed for signals other than SIGINT. * SUN: Sun raster reader was not completely robust. Now it is. * SWF: Fix pixel cache access errors in 'ping' mode. * Text annotation: An empty text string is no longer treated as an error. * Text annotation: Fix regression added in 1.3.19 which caused spurious drawing errors to be produced while rendering with text when all of the text is off the left-hand side of the image. * TIFF: Fix unreliable reading JBIG compressed files by forcing use of strip reader rather than sometimes using scanline reader (which libtiff's JBIG codec does not support). * TIFF: Fix reading or writing planar min-is-white or min-is-black images with an associated alpha channel. * WebP: WebP writer now writes truely lossless output when requested. * identify / GetImageStatistics(): Failed to compute statistics for the Black channel of CMYK image files. * VICAR: Fix problem with continuing to "read" data when there is no more data left to read. * WMF: Fix memory leaks. * WPG: Fix potential DOS due to long reads during an error condition. * XPM: Avoid strncpy() of overlapping memory. Fixed memory leaks in error paths. Fixed bad memory access caused by empty file. New Features: * compose: Supports composite operator names similar to the major \*Magick brand, without losing any compatibility with previous naming. * ICO: Windows ICO reader now supports reading PNG-encoded files. * Magick++ Geometry: New methods limitPixels() and fillArea() to support '@' and '^' geometry qualifiers. This enhancement breaks the ABI due to previous use of inline methods and no place to put the new flags. * Magick++ Image::extent(): New method to place image on sized canvas of constant color using gravity. * Magick++ Image::formatExpression(): New method format a string based on a format similar to command-line -format. * Magick++ Image::resize(): New method to resize image specifying geometry, filter, and blur. * Magick++ STL extentImage: New function object to invoke image extent method. * Magick++ Image::quiet(). New method which blocks (ignores) warning exceptions when passed a 'true' argument. * Resource limits: Added support for image Width and Height limits. Default image Width and Height limits are based on the range of a 32-bit signed integer, even for 64-bit builds which may have sufficient numeric range to image an entire galaxy. Limits may be increased as desired. * TIFF: Use define tiff:ignore-tags to ignore tags in 'corrupted' files with unknown and invalid tags. Use to read TIFF files which otherwise can not be read due to errors. * TIFF: Use '-define tiff:report-warnings=true' to enable that warnings reported by libtiff are thrown as warning exceptions so that they may be caught or will be reported at the gm command-line. * Windows Exceptions: A handler is registered (due to calling InitializeMagick()) to capture Windows Exceptions in a similar manner to the existing POSIX signal handler. If an application is using the library and wants to provide its own Windows exception handling, then it should make any changes after invoking InitializeMagick(). Feature improvements: * None. Performance Improvements: * None. Windows Delegate Updates/Additions: * PNG: Update bundled libpng to 1.6.16. Resolves known security issues. * FreeType: Update bundled Freetype to 2.5.4. Resolves known security issues. * WebP: Update bundled WebP to 0.4.2 release. * WebP is auto-linked in Visual Studio. Build Changes: * WebP is not included in the build when building with Visual Studio 6 (1998 vintage compiler!) since it requires more modern C. Behavior Changes: * AVI: Support for this format is removed since the implementation was worthless. * TIFF: Now uses YCbCr encoding when JPEG compression is requested for an RGB image. 1.3.20 (August 16, 2014) ========================= Special Issues: * Due to `GCC bug 53967`_, several key agorithms (e.g. convolution) may execute much faster (e.g. 2-3X) for x86-64 and/or when SSE is enabled for floating point math (`-mfpmath=sse`) if the GCC option `-frename-registers` is used. Default 32-bit builds do not experience the problem since they use '387 math. It is not clear in what version of GCC this problem started but it was not noticed by the developers until the GCC 4.6 timeframe. Other compilers do not suffer from this bug. Please lobby the GCC project to fix this embarrassing performance bug. Security Fixes: * No security issues were reported or fixed. Bug fixes: * Compilation: No longer undefine __attribute__ since this may be used by system or compiler headers and cause problems. * BMP: Alpha channel from BMP3 format was inverted. * PNG: Fix round-trip repeatability issue (due to rounding algorithm) with modern versions of libpng. Prefer the less accurate method which does not alter the image. * PNG: Fix some memory leaks in error-handling paths. * PNM: Scaling of alpha in sub-ranged pixels is fixed. * Wand API: Removed development debug fprintf which causes each drawing primitive to be printed to stderr. * PS, PS2, PS3, PDF: Only use resolution from image or -density if units were properly specified. Without units, resolution is worthless. * PS, PS2, PS3, PDF: Use resolution from image if it appears to be valid. * WebP: Fix inverted return status which caused failure to be reported instead of success. * Rotation clipping/shearing errors for short wide images at some angles are fixed. * -geometry: Deal with resize geometry missing width or height (e.g. '640x' or 'x480') by substituting the missing value with one which preserves the image aspect ratio. This has been documented to be supported since almost the dawn of GraphicsMagick but was not actually supported until now. * -geometry: Support '>' and '<' qualifiers with '@' qualifier to specify if image should be resized if larger or lesser than given area specification. New Features: * Wand API: MagickSetImageGravity() - New function to set image gravity. * Wand API: MagickGetImageGravity() - New function to get image gravity. * Wand API: MagickSetImageMatte() - New function to set the image matte channel enable flag. * Wand API: MagickGetImageMatte() - New function to read the image matte channel enable flag. * Wand API: MagickSetImageGeometry() - New function to set the image geometry string. * Wand API: MagickGetImageGeometry() - New function to get the image geometry string. * Wand API: MagickOperatorImageChannel() - New function to apply an operator to an image channel. * Magick++ API: New Image::thumbnail() method for fast image resizing, particularly to make thumbnails. * Core C API: Added SetLogMethod() to allow an application/library to specify a function to be called for logging. * Clang/LLVM: Provide support for clang/llvm attribute and builtin specifiers similar to that provided for GCC. * OpenMP: OpenMP native locking and thread specific data is supported via a configuration option (is not the default). This offers a "pure" OpenMP compilation mode. No real value for this compilation mode has been observed yet but it seems worthy to support. * Coders: Added BrokenCoderClass to mark coders which often malfunction or are not very useful in their current condition. * Composition: Added HardLight composition operator, which is now used by PSD and XCF formats, and available via command line, Magick++ API, PerlMagick API, and Wand API. * Composition: Added ScreenCompositePixels composition operator. * Composition: Added missing Photoshop separable compositing operations, Overlay, Exclusion, ColorBurn, ColorDodge, SoftLight, LinearBurn, LinearDodge, LinearLight, VividLight, PinLight, HardMix. * +set: Command line utilities now support +set to remove an existing image attribute. * -format: Support additional format specifiers 'g', 'A', 'C', 'D', 'G', 'H', 'M', 'O', 'P', 'Q', 'T', 'U', 'W', 'X', and '@', similar to the major brand. * -operator: New quantum operators ThresholdBlackNegateQuantumOp and ThresholdWhiteNegateQuantumOp. These correspond to -operator "Threshold-Black-Negate" and "Threshold-White-Negate". * TIFF: Now support setting the TIFF "Software" tag for users who do not want to admit to using GraphicsMagick. * WebP: All of the WebP encoder encoder options are now supported by -define arguments. Feature improvements: * Pixel interpolation quality is greatly improved, with minimal impact on performance. Pixel interpolation now also works well given an alpha channel. * WebP: WebP support is now prepared to compile with most WebP library versions and supports all features except for those pertaining to "RIFF" container support. Performance Improvements: * Non-integral image rotation performance has been improved by about 40%, with lower memory usage as well. * GradientImage: Update image is_grayscale and is_monochrome flags based on gradient color properties. Windows Delegate Updates/Additions: * PNG: Libpng 1.6.12 - June 12, 2014. * JPEG: libjpeg 9a of January 19, 2014. * FreeType: FreeType 2.5.3 of March 6, 2014. * WebP: webp 0.4.0 of January 20, 2013. * zlib: zlib 1.2.8 of April 28, 2013. Build Changes: * --without-threads no longer disables use of OpenMP. Use the already existing option --disable-openmp to disable OpenMP. * Makefiles: Include paths are now exceedingly pedantic to make sure that only the required directories are included. * VisualMagick configure: Improve configure program so that it is possible to select QuantumDepth, OpenMP, and 64-bit build via configure dialog boxes as well as options on the command line. Also automatically detects and deals with similarly named files in subdirectories so that WebP support can now build successfully. Behavior Changes: * MultiplyCompositePixels: Multiply composition now uses SVG interpretation of how alpha should be handled. No longer does a simple multiply of alpha channel. * Composition: The Difference, Darken, Lighten, and HardLight composition operators were modified to support alpha in their computations. * PNG: Using -optimize no longer triggers palette and depth optimizations since their implementations have been problematic. 1.3.19 (December 31, 2013) ========================== Special Issues: * Due to `GCC bug 53967`_, several key agorithms (e.g. convolution) may execute much faster (e.g. 2-3X) for x86-64 and/or when SSE is enabled for floating point math (`-mfpmath=sse`) if the GCC option `-frename-registers` is used. Default 32-bit builds do not experience the problem since they use '387 math. It is not clear in what version of GCC this problem started but it was not noticed by the developers until the GCC 4.6 timeframe. Other compilers do not suffer from this bug. Security Fixes: * EPT: Fix crash observed when Ghostscript fails to produce useful output. This was particularly noticeable when Ghostscript was not installed. This crash could be used to cause denial of service. * PNG: With libpng 1.6.X, avoid a crash while copying a PNG with a "known incorrect ICC profile". This crash could be used to cause denial of service. Bug fixes: * Build: Fix cross-compilation for MinGW64 on Linux build machine. * Build: configure FreeType test no longer insists that can be included. * CMS profile: Only delete the CMS transform if it is non-null. Fixed assertion observed when lcms returned a null profile and GraphicsMagick attempted to deallocate it. * Drawing: Improve error handling logic so that drawing returns quickly on pixel access errors rather than plowing on ahead. This avoids problems with SVGs which take seemingly forever to render. * Drawing via C/C++ APIs: `BevelJoin` no longer causes a MVG parsing error. * EPT: Fix crash observed when Ghostscript fails to produce useful output. This was particularly noticeable when Ghostscript was not installed. * OpenMP: Revert use of omp_set_dynamic() since it caused performance issues when using GCC's GOMP implementation and the number of threads to use is specified. * EXIF profile: Support the `SubjectArea` EXIF tag. * MIFF writer: PseudoClass format was written incorrectly for depth greater than 8. * MIFF writer: RLE compressed format used inverted alpha from the other subformats contrary to the MIFF specification. * MIFF reader: Fixes to be able to read MIFF written by ImageMagick 6.X, including DirectClass grayscale images (except for RLE compressed). * Mosaic: Fixed unsigned underflow problem with -mosaic when page offset is negative and exceeds image width or height, resulting in assertions, out of memory errors, or pixel cache limit errors. * PDF: Consistently initialize Image page width and height to image width and height. While general to all of GraphicsMagick, this change is to assure that the PDF writer computes page dimensioning consistently. PDF page dimensioning was wrong if the image had been resized with -geometry "100%". * PAM: Fix MAXVAL scaling when reading PAM images. PAM was only working correctly for images with 256 or 64k levels. * PNM: PGM "P2" format writer wrote bad output for 8-bit depth. * PNG: With libpng 1.6.X, avoid a crash while copying a PNG with a "known incorrect ICC profile". * PNG: Q8 GM build now correctly reads 16-bit PNG files. * TIFF writer: Try to avoid writing more than 32k strips per image by increasing rows-per-strip since some programs fail to read images with more than 32k strips per image. * TIM reader: PSX TIM reports 8-bit depth (rather than 16). * TTF font rendering: Improve FreeType rendering error logic so that rendering returns immediately on pixel access errors rather than plowing on ahead. * TTF font rendering: Support rendering UTF-8 up to 21-bit code points. Was only supporting 16-bit code points. * Wand API: DrawSetStrokeDashArray() / DrawGetStrokeDashArray(), fix failure to work properly due to this code path never being tested. * Windows Ghostscript: 64-bit GraphicsMagick no longer requires both 32-bit and 64-bit builds of Ghostscript to be installed in order to read Postscript and PDF formats. * XPM reader: Reported depth now depends on the colormap rather than always claiming to be 16-bit. New Features: * JPEG: Add support for writing 'XMP' profile. * PNM: As a simple non-standard extension to the standard PNM and PAM formats, support writing and reading 32-bit sample depth. Writing such files is only supported by the Q32 build although they may be read by any build. * WebP: Now supports reading and writing Google's WebP format. This feature is not currently supported by the Windows Visual Studio build. Feature improvements: * Pixel composition based on BlendCompositePixel() is enhanced to completely eliminate under-color from the blending if the under-pixel is fully transparent. Also blends based on the average opacity of both pixels rather than only the over-pixel. This change did not result in any change in the GM test suite results but it is possible that there could be some negative impact from it. Please report any issues noticed which are due to this change. * X11 `display`: For DirectClass image, use ThumbnailImage() rather than SampleImage() when creating the panner icon to improve the quality of the image. Performance Improvements: * PNG: `ping` a PNG faster by avoiding reading the image data. Windows Delegate Updates: * Updated IJG JPEG library to release 9. * Updated PNG library to release 1.6.8. * Updated lcms2 library to release 2.5. * Updated libxml2 library to release 2.9.1. * Updated FreeType library to release 2.5.2. Behavior Changes: * MIFF: Now writes PseudoClass images correctly when depth is greater than 8. This impacts the reader, which will not be able to read previously written incorrect format correctly. Images like this should be very rare. The solution is to use an older GraphicsMagick version to convert such images to a valid storage format (with a depth of 8) so that they may be read with this version. * MIFF: Now writes RLE-compressed RGBA images with correct alpha. This impacts the reader, which will not be able to read previously written incorrect format correctly. Images like this should be very rare. A solution is to use an older GraphicsMagick version to use a compression algorithm other than RLE so that they are read correctly with this version. Another solution is to process problematic images with '-operator Opacity Negate 0' to invert the alpha channel. * TIFF: Returns DirectClass images by default for MINISWHITE and MINISBLACK TIFF formats (rather then colormapped). * Windows: Also search c:\gs\fonts for Ghostscript font files. This search path is normally hard-coded into Ghostscript binaries and is a convenient place to put fonts so they may be shared by multiple Ghostscript versions. * XPM: Now limits color resolution to 16-bits, even with Q32 build. 1.3.18 (March 10, 2013) ========================== Special Issues: * Due to `GCC bug 53967`_, several key agorithms (e.g. convolution) may execute much faster (e.g. 2-3X) for x86-64 and/or when SSE is enabled for floating point math (`-mfpmath=sse`) if the GCC option `-frename-registers` is used. Default 32-bit builds do not experience the problem since they use '387 math. It is not clear in what version of GCC this problem started but it was not noticed by the developers until the GCC 4.6 timeframe. Other compilers do not suffer from this bug. Security Fixes: * None. Bug fixes: * Fixed bug with format substitutions if input string ends with a single '%'. * BMP: Fixed an old bug with decoding chromaticity primaries. * PNG: Fixed reading of interlaced images. Fix reading of sub-8-bit palette and grayscale images. Some PNG sub-formats were written incorrectly. Fix crash in PNG8 writer if image colors happened to be non-zero but image was not actually colormapped. * PNG: Configure script now also searches for libpng versions 16 and 17. * TIFF: Fix a crash which was noticed when writing RGBA separated (planar) format. * `--enable-symbol-prefix` was not prefixing all of the C symbols. Some core C library functions were not prefixed. This option applies to the Wand library API as well now. * C API: When input is from a user-provided file descriptor, the file position is restored after reading the file header bytes. Previously the file position was rewound to the beginning of the file. This allows reading embedded image data from the current offset in a file, and allows continuing to use the stream after GraphicsMagick has returned the image. * C API: It is now possible to invoke CloseBlob() multiple times. * display: Display was supposed to respond to +/-usePixmap, but was not. It was responding to +/-use_pixmap. Now it responds to both. * Windows/VisualMagick: Fix building GraphicsMagick with Intel ICC compiler driven by Visual Studio Professional 2012. * Windows: Avoid a crash and produce a useful diagnostic if Ghostscript is needed but not yet installed. New Features: * GM utility: New 'batch' command was contributed by Kenneth Xu which supports executing any number of other GM utility sub-commands in a single invocation in a sort of "batch" script. Input may be piped from standard input, from a specified file, or from a 'GM >' command prompt. This utilities front-end allows any other program/script to drive 'gm' using a co-process model and speeds up execution by eliminating utility start-up/shut-down time. * WIN64 (64-bit Windows): Windows 64-bit is now officially supported. * convert/mogrify: Now support -auto-orient to automatically rotate the image upright for viewing based on its current orientation setting. Also support -orient to support setting the current image orientation. Please note that the orientation property of EXIF profiles is not yet updated so the EXIF profile will be wrong after using -auto-orient. * C API: AutoOrientImage(), new function to automatically orient the image so that it is upright for normal viewing. * Wand API: MagickGetImagePage()/MagickSetImagePage(), new functions to support getting and setting the image page size and offsets. * PNG: Added PNG48 and PNG64 support. Added PNG00 support (png encoder that inherits its color-type and bit-depth from the input, if the input was a PNG datastream). Feature improvements: * GraphicsMagick TAP tests may now be run stand-alone using Perl's 'prove' TAP test driver. Performance Improvements: * Detection of glob specifications in file names is more efficient. Windows Delegate Updates: * None. Behavior Changes: * ltdl: Libltdl is no longer bundled. Libltdl must be previously installed on the system in order to build the modules configuration. * AppendImages() now converts subsequent images to the colorspace of the first image, and no longer converts the first image to RGB. Instead, it is assumed the user knows what she/he is doing. * SetImageColorRegion() no longer automatically converts the image to RGB. The user is responsible for assuring that the provided color is in the same colorspace as the image. 1.3.17 (October 13, 2012) ========================== Security Fixes: * PNG: Fix for CVE-2012-3438. The Magick_png_malloc function in coders/png.c in GraphicsMagick 6.7.8-6 does not use the proper variable type for the allocation size, which might allow remote attackers to cause a denial of service (crash) via a crafted PNG file that triggers incorrect memory allocation. * Automake (derived): Fix for CVE-2012-3386: The "make distcheck" rule in GNU Automake before 1.11.6 and 1.12.x before 1.12.2 grants world-writable permissions to the extraction directory, which introduces a race condition that allows local users to execute arbitrary code via unspecified vectors. Bug fixes: * PNG: Reading sub-8-bit palette images is fixed (images looked stretched). * SVG: Fixed bug which allowed MVG and SVG files with long vector paths to crash the software. * SVG: Ignore XML headers rather than rendering them as text. * MVG/SVG/WMF/-draw: It is now possible to draw a plain ',' character. * WMF: Fixed a bug which caused wrong centered-text placement. * import: Return status was inverted. * configure: Don't force that liblzma is used just because libtiff is used. New Features: * The configure script now supports a --enable-quantum-library-names option to enable that shared library name includes quantum depth to allow shared libraries with different quantum depths to co-exist in same directory (only one can be used for development). * JNX: Support is added for reading the Garmin proprietary Image Format. * BMP: Support an alpha channel in uncompressed 32-bit BMP. Feature improvements: * `-lat`: The adaptive threshold algorithm is replaced with a new algorithm which scales linearly (rather than quadratically) with area size. * Tests: Test suite is re-written to use TAP-based tests. * GIF: Reader tries to be better at detecting and reporting failures. Performance Improvements: * -lat: Adaptive threshold is much faster with large area sizes. Windows Delegate Updates: * Dcraw 9.16 is now included in the build (with JPEG and JPEG2000 support). * Libxml2 is updated to the 2.9.0 release. * Libtiff is updated to the 4.0.3 release. * Lcms2 is updated to the 2.4 release. * Libpng is updated to the 1.5.13 release. Behavior Changes: * Loading modules is only supported for the modules build. Previously any build using shared libraries could load modules. * Bundled libltdl is now configured as 'installable' rather than 'convenience'. * -enhance: Only filter based on color channels (ignore opacity). * BrowseDelegate: Web browser (for viewing help information) now defaults to 'xdg-open', but if it is not found, then configure will search for firefox, google-chrome, mozilla (in that order). 1.3.16 (June 24, 2012) ========================== Security Fixes: * Don't translate 'comment' and 'label' attributes if the request is made while a file is being read. Only translate such attributes if they come from the command line or API user. Bug fixes: * SWT: SWT reader suffered from a number of implementation errors which caused it not to work any more. Works again. * XBM: Fix memory leak observed when reading file in 'ping' mode. * Support -trim on images which use a consistent (single color) transparent background. In this case, trim is done based on opacity rather than foreground color. * Include in order to assure that 'size_t' and 'ssize_t' are declared. This is necessary since MagickExtentImage() uses these types as part of its definition. * `+repage` was not working because parser was insisting that it should include an argument. * -units was scaling existing resolution the wrong way around (i.e. multiplying rather than dividing). * PerlMagick: Fix compilation with Perl 5.16. * PingBlob(): PingBlob was not working for all cases. Is now based on BlobToImage() for assured reliability. New Features: None Feature improvements: * MAT: Animated movies inside 4D matrices are loaded now. * PDF: File base name is used as the document title. * PNG: Fix issues observed specifically with libpng 1.5.10. Performance Improvements: * Pixel iterators should be more efficient now if the image uses a file-backed cache. * Motion blur algorithm does scale well as cores are added so include OpenMP support for it by default. Windows Delegate Updates: * JPEG: Updated to IJG 8d release. * PNG: Updated to 1.5.11 release * TIFF: Updated to 4.0.2 release. * Zlib: Updated to 1.2.7 release. * libxml2: Updated to 2.8.0 release. Behavior Changes: None 1.3.15 (April 28, 2012) ========================== Security Fixes: * Libpng in Windows build is updated to 1.5.10 release. Provides a fix for CVE-2011-3048. Bug fixes: * PNG - fixed problem with bit depth when the encoder decides to write RGBA instead of indexed PNG. * Fixed some temporary file leaks which were caused by the temporary file name being automatically extended to include a scene number, and therefore fail to be deleted. New Features: * Added '+noise random' and '-operator noise-random' to 'convert' and 'mogrify'. This modulates the existing image data with uniformly random noise. * Added -strip option in composite, convert, mogrify, and montage to remove all profiles and text attributes from the image. * Added -repage option to composite, convert, mogrify, and montage subcommands to reset or adjust the current image page offsets based on a provided geometry specification. * New C function StripImage() to remove all profiles and text attributes from the image. * New C function ResetImagePage() to adjust the current image page canvas and position based on a relative page specification. * C functions GenerateDifferentialNoise(), AddNoiseImageChannel(), QuantumOperatorRegionImage(), AddNoiseImage() updated to support RandomNoise enumeration. * New C++ Image method strip(), and unary function stripImage() to remove all profiles and text attributes from the image. * XCF format now respects image subimage and subrange members so that returned image layers may be selected. * The INFO coder (e.g. output file "info:-") now respects the -format option so that its output may be adjusted identically to how -format works for 'identify'. * TclMagick now supports Random noise. Feature improvements: * C function ThumbnailImage() now allows the user to override the filter used, but still defaults to using the box filter. Performance Improvements: * None Behavior Changes: * No longer add a printf-style scene formatting specification to filenames which do not have one and no longer automatically operate in 'adjoin' mode in such cases. If multiple numbered files are intended to be output, then add +adjoin to the command line and use an output filename specification similar to "image-%d.jpg". Output files are now completely specified and predictable but this may break some existing usages which anticipate the automatic file numbering. 1.3.14 (February 25, 2012) ========================== Security Fixes: * Windows bundled libpng updated to the 1.5.9 release, which fixes the dire CVE-2011-3026 buffer overrun bug. Bug fixes: * EMF format : Fixed wrong module mapping which caused EMF reading to not work under Windows. * TGA format: Assume that 32-bit TGA files have an alpha channel, even if they are not marked as such. * XCF format: Fix reading XCF which is comprised of different sized layers. * JPEG & CineonLog: Convert RGB-compatible colorspaces (e.g. CineonLog) to RGB by default since that was the case prior to release 1.3.13. * RAW formats: Small memory leak in dcraw module was fixed. * Resize: ResizeImage() was ignoring its resize filter argument and was using the filter setting from the Image structure instead. * The mirror virtual pixel method was broken. New Features: * Open64 Compiler Suite: Version 5.0 is fully supported. * Wand API: Added MagickExtentImage(). * MEF RAW: Mamiya Photo RAW "MEF" format is now supported. Feature improvements: * DPX format: Original file endianness is preserved by default. * PNG library: Updated libpng to 1.5.9 release. * TIFF library: Updated libtiff to 4.0.1 release. * Zlib library: Updated to zlib 1.2.6 release. Performance Improvements: * Despeckle algorithm (-despeckle) is many times faster. Behavior Changes: * DPX format: Original file endianness is preserved by default. 1.3.13 (December 24, 2011) ========================== Security Fixes: None Bug fixes: * In I/O blob, don't rewind already open file handle passed to OpenBlob() since we don't know the intended state of this file handle, and because it prevents appending to an existing file. * In AppendImageProfile(), don't leak profile buffer while appending a chunk to an existing profile. * Fix deadlock in ClonePixelCache() which was caused by using the same semaphore pointer in the source and destination images. * Removed bogus SyncBlob() code which sometimes caused a crash and was not useful. * Fixed crash or hang which occured when the user entered CONTROL-C while threaded code was being executed. * Fix core dump in AcquireOneCacheViewPixelInlined() when the image is in CMYK space. * In MontageImages (montage), fix crash observed with "-geometry x+0+0". * The TIFF reader was crashing for images which use the TIFFTAG_OPIIMAGEID tag. * AppendImages() (-append) was failing when only one image was provided. * The `animate`, `display`, and `identify` commands now report any error only once, and then proceed to the next file name rather than quitting. * Don't change the locale settings in InitializeMagick() since this may cause problems for international users. API users are still responsible for assuring that locale settings don't break floating point parsing and output (i.e. floating point decimal needs to be '.' rather than ','). * RPM build is fixed (PerlMagick build was broken). * RPM build installs documentation to expected places on Red Hat type systems. * Fixes for usage with OpenSolaris. * DESTDIR is supported by PerlMagick build. * The matte channel was not being properly enabled or respected for TXT images. * InitializeMagick() and DestroyMagick() are now fully thread safe. * When a shear angle was zero, the shear request was being ignored entirely. * In DispatchImage(), the `K` channel was always output as black for "CMYK" specification unless the image matte flag was True. * MATLAB fixes. * PNG fixes. * PCL fixes for printing bi-level image on Konica-Minolta printers. * EPT error handling fixes. * JPEG reader was sometimes truncating large IPTC profiles. * JPEG writer now handles errors properly rather than allowing libjpeg to exit the program (or hanging if driven by Magick++). * JPEG reader now treats an unhandled EXP marker as a warning rather than a hard error. * File open errors are now reliably reported. * Improved rendering precision when using the drawing APIs. * For the Magick++ Image backgroundColor(), borderColor(), and matteColor() methods, preserve the opacity part of the user-specified color. New Features: * Add support for drawing text using a bitmap font. * benchmark command supports a -stepthreads option to execute the specified command with an increasing number of threads to measure how an algorithm benefits from threading. This mode includes a column to show the speedup compared with one thread, and the Karp-Flatt metric * Added support for invoking "gs-cmyk" and "gs-cmyka" entries in delegates.mgk when ColorSeparationType or ColorSeparationMatteType is requested. These cause Ghostscript to always output CMYK PAM format (even if the input file was not in CMYK format). * EXIF profiles are preserved when writing JPEG files. * The -mosaic command now respects the composition option specified by -compose as well as the image background color specified by -background. * The TXT coder now supports multiple image frames. * For image normalization (-normalize), add support for histogram-threshold setting to specify the percentage of the histogram to discard when computing image normalization parameters (default is 0.1%). For example `-set histogram-threshold 0.01 -normalize`. * Added an `INFO` coder which produces textual image description output similar to `identify` but may be used with convert like "gm convert myfile info:-". * Support application of the PDF crop box via '-define pdf:use-cropbox=true'. * For PCL printer output, define pcl:fit-to-page in order for the printer to scale the image to fit the page. * Added order dither 5x5, 6x6, and 7x7 circular dither patterns to create a halftone effect. * PNM subformats are now reported as the specific subformat rather than just "PNM". * NetPBM's PAM format is now supported. * MacPaint image format reader is added. * Added TIFF LZMA compressor support. * Added TIFF support for a tiff:group-three-options define to allow power-users to set the value of the GROUP3OPTIONS tag. * New core C API function SetImageColorRegion() to set the constant pixel color for a specified region of the image. * New Wand C API function MagickWriteImagesFile() to append images to a provided file handle. * New Wand C API function MagickSetImageSavedType() to allow specifying the storage type used when saving the file (rather than changing the current image characteristics). * In Wand C API, the functions NewPixelWand(), NewDrawingWand(), and NewMagickWand() invoke InitializeMagick() automatically in case user forgets to do so. * New Wand C API function MagickSetFormat() to allow setting the file or blob format before it has been read. * New Wand C API function MagickSetDepth() to set the depth used when reading from an image format which requires that the depth be specified in advance. Feature improvements: * Now compiles properly with libpng 1.4.X and 1.5.X. * Lcms 2.X is supported. Performance Improvements: * TGA read performance improved. * PNM read/write performance improved. * Convolution (-convolve, -sharpen, -guassian, etc.) is faster. * Adaptive threshold image (-lat) is faster. * Image trimming (-trim) is faster. Behavior Changes: * For DPX format and packed 10 bits, datums are now represented in the same (reversed) order for all RGB and YCbCr formats. Previously YCbCr 4:4:4 formats were not swapping the word datums because the only real-world files encountered did not swap the word datums. * The -colors, -map, and -monochrome options now take effect immediately rather than at the end of all other processing. * Removed non-standard multi-frame extension for SGI format. * Windows install footprint is more consistent between DLL and static builds. * LZMA compressed tarball is in 'xz' format rather than deprecated 'lzma' format. 1.3.12 (March 8, 2010) ========================== Security Fixes: * Updated libpng Windows sources to 1.2.43 in order to resolve CVE-2010-0205 as it pertains to the GraphicsMagick Windows build. Bug fixes: * Filter mode (write to stdout) was completely broken. * Should now compile with libpng 1.4. * Windows PerlMagick build identified itself as the wrong version. New Features: * None Feature improvements: * None Performance Improvements: * None Behavior Changes: * DCX output format is only written on request. Previously the PCX coder would automatically switch to DCX format if multiple frames would be written. 1.3.11 (February 21, 2010) ========================== Security Fixes: * Fixed array underflow on systems using signed char which could result in a program crash due to extended characters in filenames or in certain file formats. Bug fixes: * Fixed array underflow on systems using signed char which could result in a program crash due to extended characters in filenames or in certain file formats. New Features: * Added a -thumbnail command to 'convert' and 'mogrify'. This is a faster way to scale down the image when speed is a primary concern. * Added a -extent command to 'convert' and 'mogrify' which composites the image on top of a backing canvas image of solid color. * Added support for -compose to the 'convert' and 'mogrify', which were documented to support it (but did not). Feature improvements: * None Performance Improvements: * Requests for 'Over' and 'Atop' composition are converted to a request for the (faster) 'Copy' composition when both images are opaque. Behavior Changes: * None 1.3.10 (February 10, 2010) ========================== Security Fixes: * None Bug fixes: * +adjoin was not working correctly for the case when only one image frame is present. With +adjoin and writing one frame to "foo%d.jpg" it was outputting "foo%d.jpg" rather than "foo0.jpg". * When drawing paths, memory allocation for the points was much larger than it needed to be (patch by Vladimir Lukianov). New Features: * None Feature improvements: * None Performance Improvements: * None Behavior Changes: * To reiterate the change which first appeared in 1.3.9, there is no longer an implicit +adjoin if the output file name happens to contain a %d sequence, or there are multiple frames and the output file format only supports storing one frame. Specify +adjoin if scene number substition is desired in the output file names. 1.3.9 (February 4, 2010) ======================== Security Fixes: * None Bug fixes: * Fix "double free" error when using gm import -frame. * XPM does not support RGBA color syntax, so return RGB instead. * The display '-update' option was only working in conjunction with the '-delay' option with a delay setting of 2 or greater. * For formats which support multiple frames, output with +adjoin to filenames containing a scene specification (e.g. foo%02d.tiff) was resulting in wrong output file names. * -convolve was crashing rather than reporting an error. * Fixed crash if the number of OpenMP threads was reduced from the original value via '-limit threads' or omp_set_num_threads(). * -blur was not blurring the opacity channel for solid-color images. * When installing HTML documentation, many files were included which are not part of the formatted documentation. * Several deleted global string constants are restored with deprecated status in order to assure that symbols are not removed from the ABI. New Features: * None Feature improvements: * None Performance Improvements: * None Behavior Changes: * There is no longer an implicit 'adjoin' if an output filename contains an apparent scene specification (e.g. foo%02d.tiff) and multiple files are not needed to save the image.. It is necessary to use +adjoin. For example ``gm convert foo.pdf +adjoin %02d.tiff``. * -flatten now applies the image background color under the first image in the list if it is not already opaque. 1.3.8 (January 21, 2010) ======================== Security Fixes: * Fix for CVE-2009-1882 "Integer overflow in the XMakeImage function". * Fix lockup due to hanging in loop while parsing malformed sub-image specification (SourceForge issue 2886560). * Libltdl: Updated libtool to 2.2.6b in order to fix security issue. Resolves CVE-2009-3736 as it pertains to GraphicsMagick. Bug fixes: * -convolve, -recolor: Validate that user-provided matrix is square when parsing -convolve and -recolor commands in order to avoid a core dump. * CALS: Reading images taller than the image width resulted in a failure. * ConstituteImage(), DispatchImage(): 'A' and 'T' should indicate transparency and 'O' should indicate opacity. Behavior was inconsistent. In some cases 'O' meant transparency while in other cases it meant opacity. Also, in a few cases, matte was not getting enabled in the image as it should. * DCRAW: Module name was not registered so modules based builds were not supporting formats provided via 'dcraw'. * GetOptimalKernelWidth1D(), GetOptimalKernelWidth2D(): In the Q32 build, convolution kernel size was estimated incorrectly for large sigmas on 32-bit systems due to arithmetic overflow. This could cause wrong results for -convolve, -blur, -sharpen, and other algorithms which use these functions. * Image Size: Fixed the ability to pass the image size via the filename specification like "myfile.jpg[640x480]" rather than needing to use -size. * IPTC: Blob data needed to be padded to an even size. Size is now correctly reported. * IPTC: Returned IPTC string values were one character too short. * Large Files: Large pixel cache files were not working under GNU Linux. * JP2: Fixed some value scaling problems. * JP2: Fix possible crash at exit when Jasper is used by a modules build. * MPC: is_monochrome and is_grayscale flags were not managed properly for the MPC coder. * PCL: Page was not always being ejected. * PNG: The png8 encoder would fail when trying to write a 1-color image. * PSD: PSD parser was confused by 0x0 pixel layers, resulting in image data corruption of all following layers. * -rotate, -shear: Some internally-reported errors were potentially being lost. * Subrange/stdin: Commands now support reading an image from stdin in conjunction with a subrange specification (e.g. "-[1]"). * Magick++ STL ShadeImage: Implementation was completely botched. New Features: * CALS Type 1 files may now be written (Work contributed by John Sergeant). CALS support is dependent on the TIFF library. * GROUP4RAW encoder supports reading/writing RAW Group4 data. * JP2: JPEG 2000 may now be written in arbitrary bit depths ranging from 2 to 16 rather than just 8 or 16. * JPEG: IJG JPEG library version 7 is now supported. * JPEG: Added jpeg:block-smoothing and jpeg:fancy-upsampling defines to control these JPEG library options. * JPEG: Detect and apply colorspaces appropriately for ITU FAX JPEG. * Resource Limits: There is now a "threads" resource limit which allows specifying the number of OpenMP threads which may be used, similar to the OMP_NUM_THREADS environment variable. * TIFF: Allow CIELAB TIFF to be read. * MagickGetImageAttribute()/MagickSetImageAttribute(): New Wand methods to support getting and setting an image attribute. Contributed by Mikko Koppanen. * ClonePixelWand(): New Wand method to deep-copy an existing pixel wand. * ClonePixelWands(): New Wand method to deep-copy an array of existing pixel wands. * MagickCdlImage(): New Wand method to apply the ASC CDL to an image. * MagickGetImageBoundingBox(): New Wand method to return the crop bounding box required to remove any solid-color border from the image. * MagickGetImageFuzz(), MagickSetImageFuzz(): New Wand methods to get and set the color comparison fuzz factor. * MagickHaldClutImage(): New Wand method to apply a Hald CLUT to an image. * MagickSetResolution(): New Wand method to set the wand resolution. * MagickSetResolutionUnits(): New Wand method to set the wand resolution units. * Magick++: Allow Magick++ library to built as a DLL under MinGW and Cygwin. This requires a modern GCC in order for C++ exceptions to work. Feature improvements: * Cygwin: Cygwin 1.7 is now supported. * JPEG compression settings are preserved (if possible) when inserting JPEG blobs into formats which use JPEG. * PDF: If the original file used JPEG compression, then use JPEG compression with original settings (if possible). * TIFF: Update Windows build to use libtiff 3.9.2. * X11 Display: Apply a checkerboard pattern underneath transparent images which use more than simple binary transparency. Performance Improvements: * Gamma: Performance is improved for Q8 and Q16 builds. Also preserve full precision in Q32 build. * String data is dealt with a bit more efficiently (fewer allocations, less memory, and less CPU). Behavior Changes: * InitializeMagick() MUST be invoked prior to using any Magick API function. Failure to do so will likely lead to an immediate application crash. This is due to initialization and runtime changes intended to improve thread safety and efficiency. Previously it was only strongly recommended to invoke InitializeMagick(). * ConstituteImage(), DispatchImage(): 'A' and 'T' should indicate transparency and 'O' should indicate opacity. Behavior was inconsistent. In some cases 'O' meant transparency while in other cases it meant opacity. Also, in a few cases, matte was not getting enabled in the image as it should. * colors.mgk: Is now empty to default and is optional. Previous content is now compiled into the library in an efficient way, but existing values may be modified, or new values added by adding entries to color.mgk. * DisableSlowOpenMP is now the default. Use --enable-openmp-slow to enable OpenMP for algorithms which sometimes run slower rather than faster. * magic.mgk: This configuration file is no longer used since this data is now compiled into the library in an efficient way. * modules.mgk: Is now empty to default and is optional. Previous content is now compiled into the library in an efficient way, but existing values may be modified, or new values added by adding entries to modules.mgk. * Third party executables not included in the Visual Studio build are no longer bundled in the GraphicsMagick installer. This means that hp2xx.exe, mpeg2dec.exe, and mpeg2enc.exe are no longer distributed. 1.3.7 (September 17, 2009) ========================== Security Fixes: * PCX: Detect improper rows, columns, or depth. Fixes CVE-2008-1097 "Memory corruption in ImageMagick's PCX coder". * DrawDashPolygon: Avoid a crash which sometimes occured with tiny polygons. Bug fixes: * JPEG: Profile chunks need to be concatenated in order to build the whole profile. This was not working so embedded profiles larger than 32K or maybe 64K were being corrupted. This bug was introduced in GraphicsMagick 1.2. * Meta: Fix memory leaks. * Meta: Work better with with IPTC record 2 blocks and deal better with IPTC embedded in an 8BIM profile. Fixes by John Sergeant. * MPC: Fix crash when reading MPC and the input image is modified. * PNG: Ensure that the opacity channel is properly initialized. * -profile: Lowercase arguments were sometimes not working as expected. * Topol: Topol reader actually works now and is included in test suite. * TIFF: Read and write JPEG-compressed grayscale TIFF correctly. * VisualMagick configure now works properly when output paths are specified. * WMF: Eliminate memory leaks. New Features: * MagickWand: New method MagickSetCompressionQuality() to allow setting the compression quality. * MagickWand: New method CloneDrawingWand() to deep-copy a drawing wand. * MagickWand: New method DrawGetException() to retrieve information regarding the last drawing wand exception (if any). * MagickWand: New method DrawClearException() to clear a drawing wand exception. * Magick++: New Image method cdl() to apply the ASC CDL. * Magick++: New Image method colorMatrix() to apply a color matrix to the image channels. * Magick++: New Image method haldClut() to apply a color lookup table (Hald CLUT) to the image. * MSL/Conjure: Added a new 'profile' command which applies, adds, or removes one or more IPTC, ICC or generic profiles from a file. Work contributed by John Sergeant. * Added a 'time' subcommand to provide Unix-style 'time' output when a 'time' capability is missing, or the reporting format is inconsistent. For example 'gm time convert ...'. Feature improvements: * ColorMatrixImage(): Add opaque opacity channel when needed. * PDF & PS: Use '-type palette' prior to input file name to cause Ghostscript to return a dithered colormapped image. * PNG: Now compiles with libpng-1.4.0beta74 and later. * TIFF: Libtiff in Windows build is upgraded to 3.9.1. This allows GraphicsMagick to read and write 16 and 24 bit float TIFF files. * Windows code to find Ghostscript is rewritten from scratch. Performance Improvements: * Drawing of points, lines, and polygons (and complex shapes based on these) is now accelerated using OpenMP with excellent speed-up. * ICC color transforms now see linear speedup from OpenMP. * Rotate: For rotations of 90 or 270 degrees, tile sizes are selected more appropriately. Behavior Changes: * No longer clear the exception structure at the start of ReadImage() and other similar functions since this sometimes masks errors. The API user is expected to make sure that the exception structure is clean prior to invoking a function. * SVG: Writer is now disabled since it usually does not work properly. 1.3.6 (July 25, 2009) ===================== Security Fixes: * None. Bug fixes: * Composition was failing when the change image overlaps off the left side of the canvas. * EPT, PDF, PS: PDF bounding box is sometimes incorrect or not globally applicable so don't specify bounding box when reading PDF files. * OpenMP: Fix (benign) multi-thread cross-contentions (detected by valgrind's Helgrind). * TIFF: Fix problem with reading one bit per sample RGB images. * TIFF: Writer was using rows-per-strip of 8 when writing JPEG-compressed TIFF. This does not work for vertical subsampling, and some TIFF readers insist on 16. The rows-per-strip is now required to be a multiple of 16. * TIFF: In some cases, the TIFF reader and writer were accessing planar TIFF in row-order rather than plane-order, which resulted in sever buffering problems in libtiff, and failure when compression was used. * -write now works usefully as documented. * Temporary file name generator was not random enough, resulting in some file name collisions for GraphicsMagick processes started at the same time. * PerlMagick: Fixed Ping on a BLOB. * GetImageDepth was leaking memory. * Convert/mogrify -mask option was leaking memory. * Mogrify -output-directory option was leaking memory. * DPX: Fixed memory leak encountered when subsampling to 4:2:2. * DPX: Values read received insuficient scaling, which round-tripped correctly, but rounded-down excessively if any image processing was applied. New Features: * Added HRS reader for slow scan TV (contributed by Fojtik Jaroslav). * Pthreads (POSIX threads) API may now be used under the WIN32 API. * New access confirmation facility (MagickConfirmAccess) to allow the API user to monitor and/or block access to files and URLs. This allows the API user to implement a security policy based on actual accesses. * New color matrix function (ColorMatrixImage) to apply a color matrix similar to Adobe Flash Flash.filters.colorMatrixFilter(), and Windows GDI+ ColorMatrix class, (order up to 5x5) to the image pixels. This is accessible via the -recolor command option. * Added an IDENTITY coder to return a Hald identity CLUT image of specified order (e.g. "identity:8"). * Added a Hald CLUT capability as described at http://www.quelsolaar.com/technology/clut.html. This allows a color transformation to be easily created and replicated on any number of images. The algorithm is accessed by the -hald-clut option of 'convert' and 'mogrify'. Original algorithm by Eskil Steenberg and adapted for GraphicsMagick by Clément Follet, with additional work by Bob Friesenhahn. * Added support for the ASC CDL transform. Available as -asc-cdl via the 'convert' and 'mogrify' subcommands. Original implementation by Clément Follet but considerably re-worked by Bob Friesenhahn. Implementation passes the +/- 1 count accuracy requirement required by the ASC CDL SOP tests. * Added support for reading CALS Type 1 format (contributed by John Sergeant). CALS is a standard raster format used by the US Department of Defense for storing blueprint images. * Added a random number generation system based on George Marsaglia's multiply-with-carry generator. Somewhat slower than rand() but produces better random numbers with a period >2^60. This is a much better random number generator than the C library rand() and the algorithm is integrated in a way which maximizes multi-thread performance. * The 'compare' command now supports a -maximum-error option to specify the maximum image error so that it may be used to support boolean logic in automated test scripts. * For OpenMP-builds, the '-list resource' output now indicates the number of threads which will be used. Feature improvements: * Image resize now avoids adding "halos" around objects when resizing an image which contains transparency (patch contributed by Pavel Merdin). * DICOM: The DICOM reader is completely re-written and is much more functional now. A few features (e.g. RLE compression) are still missing. This work is contributed by John Sergeant. * EXIF: Unprintable characters in EXIF attribute strings are now returned using three-digit octal notation. Unknown tags are identified via their four-character hex value. * PCL: PCL writer is rewritten to fix many bugs, add support for compression, add support for 8 bit PseudoClass images, and dramatically improve usability (work contributed by John Sergeant). * TIFF: Allow the user to force the returned image to be TrueColor type for min-is-white and min-is-black TIFF files. * TIFF: User can now specify the predictor using syntax like '-define tiff:predictor=2'. * TIFF: User can now specify the rows-per-strip value when using JPEG compression. * TXT: The TXT reader is now capable of reading image files written by the TXT writer, as well continuing to render ASCII text into an image (work contributed by Fojtik Jaroslav). * Utilities @file.txt syntax for including a list of files to use as an argument now really works as expected. This may be used to inject any other text into the command line as well. As a result, the 'mogrify' utility may be invoked on thousands of files at once while obtaining the list of files to process from a text file. * The 'mogrify' utility now caches argument images so that they are loaded only once when mogrify is used to process multiple image files. Performance Improvements: * -median and -noise now see reliable linear speedup as threads are added. Behavior Changes: * PerlMagick is configured but no longer built by default. * Use '-interlace Line' to produce an interlaced GIF, PNG, or progressive JPEG. 1.3.5 (January 26, 2009) ========================= Security Fixes: * BMP and DIB formats were throwing an assertion for negative height values. This caused the process to crash. Bug fixes: * Don't install Magick++ headers if C++ is disabled. * Linux RPM SPEC file needs to always install the loadable module .la files or else the modules won't load. * Windows runtime DLLs were for the wrong compiler version, resulting in failure to execute if the correct runtime DLLs are not available. New Features: * None Feature improvements: * FITS: Parsing is more robust. Performance Improvements: * None 1.3.4 (January 13, 2009) ========================= Security Fixes: * None. Bug fixes: * Now runs under Windows Vista (as a 32-bit application). * Fix for colorspace transform math overflow in Q32 build. New Features: * Windows build supports OpenMP and requires Windows 2000 or later (source code still supports Windows '98). * Support large files under Windows. * Support reading/writing 16 and 24 bit float TIFF files. * Support reading/writing 64 bit integer TIFF files. * Added "Log", "Max", "Min", and "Pow" options to -operator. Feature improvements: * Debug logging now properly prints 64-bit offset values. Performance Improvements: * Improve resource estimation for Microsoft Windows systems. 1.3.3 (December 9, 2008) ======================== Security Fixes: * None. Bug fixes: * 'identify' was throwing an assertion when used on colormapped files (this bug was introduced by 1.3.2). * With the -segment option, eliminate trashing the image colors when used on huge images. * 'identify -format "%c"' now reports the entire comment regardless of size. * Argument to -convolve is no longer arbitrarily truncated so huge convolution kernels may now be specified from the command line. Performance Improvements: * Image segmentation (-segment) is now accelerated using OpenMP and uses several other tactics to improve execution performance. * 'identify "*"' now successfully works in a 32-bit application when used in a directory containing a million files. * 'identify' now executes quickly when used on TIFF files. 1.3.2 (November 29, 2008) ========================= Security Fixes: * None. Bug fixes: * -roll was failing for colormapped images. * VID: Memory leak fix. * PREVIEW: Solarize parameter was wrong. * Delegates previously using 'spawn' needed an ampersand so that starting the child process does not hang the GUI. Feature improvements: * +profile now supports an exclusion syntax. For example ``+profile '!icm,*'`` removes all of the profiles except for the ICM profile. The new syntax also allows multiple profiles to be listed at once. Performance Improvements: * AdaptiveThreshold, Blur, Convolve, and MotionBlur no longer process the opacity channel unless the image has one. 1.3.1 (November 17, 2008) ========================= Security Fixes: * None. Bug fixes: * RPM build, Fixes to successfully build binary RPMs for Red Hat Linux 4. * MSL/conjure, Fix bug with attributes becoming appended to themselves. Fix memory leaks. Feature improvements: * New --disable-openmp-slow configure option for disabling use of OpenMP for algorithms which may run slower on operating systems with crummy thread libraries. * JPEG, Allow user to specify DCT encoding method via jpeg:dct-method define. Also allow control over whether huffman encoding is used via jpeg:optimize-coding define. Performance Improvements: * OpenMP (parallel processing) improvements for these functions: - Rotate by 90 and 270 degrees (-rotate) 1.3 (November 9, 2008) ====================== Security fixes: * AVI reader: Re-worked to be more robust against crash or DOS. * AVS reader: Re-worked to be more robust against crash or DOS. * DCM reader: Re-worked to be more robust against crash or DOS. * EPT reader: Re-worked to be more robust against crash or DOS. * FITS reader: Re-worked to be more robust against crash or DOS. * MTV reader: Re-worked to be more robust against crash or DOS. * PALM reader: Re-worked to be more robust against crash or DOS. * RLA reader: Re-worked to be more robust against crash or DOS. * TGA reader: Re-worked to be more robust against crash or DOS. * Avoid possible crash in GetImageCharacteristics() when substituting text in comment read from file. * Cineon reader: Fixed crash with broken file from Sami Liedes. * Palm reader: Fixed crash with broken files from Sami Liedes. * PICT reader: Fixed crash with broken files from Sami Liedes. * DPX reader: Validate file data better to avoid improper operation with intentionally (or accidentally) defective files. * XCF reader: Fixed crash with broken files from Sami Liedes. Bug fixes: * Libbz2 is now detected for MinGW. * Install documentation under /usr/local/share/doc/GraphicsMagick by default, according to GNU conventions. * In PerlMagick, Dissolve composition was not working right. * FITS: Ensure that written format conforms to specification. * TIFF: - Don't accidentially convert CMYK images to RGB. - Eliminated a memory leak in the codec support detection code. * JPEG: Removed over-write of image->client_data. * PDF: Try to properly deal with reading rotated PDFs. * PNG: Fixed crash when writing PNG images with transparency and either optimize is requested, or the image is colormapped. * Configure: Fixed the --enable-magick-compat configure option, which had stopped working. * Configure: Fixed --without-magick-plus-plus so that it works again. This stopped working in the 1.2 release cycle. * Configure: Fixed MagickLibVersion text string generation so that it is now correct when a component of the release number exceeds '9'. Now components can safely count up to '99' before there is a problem. Performance Improvements: * OpenMP (parallel processing) improvements for these functions: - Affine transform (-affine -transform) - Average images (-average) - Add noise (+noise) - Black threshold (-black-threshold) - Blur (-blur) - Border (-border) - Channel import, export, and depth-setting (-channel, -depth) - Clip path - Coalesce (-coalesce) - Colorize (-colorize) - Colorspace transformation (-colorspace) - Compare images ('compare' command) - Composition ('composite' command) - Convolution (-convolve, -edge, -emboss, -gaussian, -sharpen) - Contrast adjust (-contrast) - Crop (-crop) - CycleColormap (-cycle) - Depth setting (-depth, -operator depth) - Despeckle (-despeckle) - Enhance (-enhance) - Equalize (-equalize) - Flatten (-flatten) - Flip (-flip) - Flop (-flop) - Frame (-frame) - Gamma adjust (-gamma, -operator gamma) - Gradient - Implode (-implode) - Levels adjust image (-level) - Local adaptive threshold (-lat) - Median filter (-median) - Minify image (-minify) - Modulate image (-modulate) - Morph image (-morph) - Mosiac (-mosaic) - Motion blur (-motion-blur) - Negate image (-negate) - Noise filter (-noise) - Normalize image (-normalize) - Oil Paint (-paint) - Opaque (-opaque) - Ordered dither (-ordered-dither) - Operators (-operator) - Profile adjust (ICC) (-profile) - Random threshold (-random-threshold) - Resize image (-resize) - Raise image (-raise) - Roll image (-roll) - Rotate image (-rotate) - Shade image (-shade) - Shear image (-shear) - Shave (-shave) - Solarize image (-solarize) - Spread image (-spread) - Statistics computation (identify -verbose) - Swirl (-swirl) - Threshold channel (-threshold, -operator threshold) - Threshold image (-threshold) - Transparent (-transparent) - Trim image (-trim) - UnsharpMaskImage (-unsharp) - Wave (-wave) - White threshold (-white-threshold) * Improved coder management performance. * XCF (GIMP) reader is much faster. New Features: * Use MAGICK_CODER_STABILITY environment variable to enable a subset of the coders based on their stability classification. * Use MAGICK_IO_FSYNC environment variable to cause written file to be synchronized to disk to avoid possible data loss on power fail. * Added 'compare' command to statistically or visually compare two image files. * Added new channel operators (-operator): - Assign - Gamma - Depth - Negate - Noise-Gaussian - Noise-Impulse - Noise-Laplacian - Noise-Multiplicative - Noise-Poisson - Noise-Uniform - Threshold - ThresholdBlack - ThresholdWhite * New composition operators (-compose): - CopyBlack - CopyCyan - CopyMagenta - CopyYellow - Divide * Added -motion-blur to motion blur the image. * Mogrify and convert now support -black-threshold and -white-threshold. * MAT: Now supports reading compressed files. * FITS: Now supports 8, 16, 32 bit integer, float, and double images and writes correct FITS format. * DCRAW: Coder proxy module allows reading digital camera files as if they were natively supported. * New C API functions: - AddNoiseImageChannel(), add noise to an image channel. - BlurImageChannel(), blur an image channel. - GaussianBlurImageChannel(), gaussian blur an image channel. - ImportImageChannelsMasked(), import selected image channels. - SharpenImageChannel(), sharpen an image channel. - UnsharpMaskImageChannel(), unsharpmask an image channel. - New cache view interfaces to correct shortcomings of original ones. New interfaces are AcquireCacheViewPixels(), AcquireOneCacheViewPixel(), AcquireCacheViewIndexes(), GetCacheViewPixels(), SetCacheViewPixels(), and SyncCacheViewPixels(). The deprecated functions are AcquireCacheView(), GetCacheView(), SetCacheView(), and SyncCacheView(). - GetCacheViewRegion() reports region bounded by a cache view. - GetCacheViewArea() reports area bounded by a cache view. - ExportViewPixelArea() exports a cache view as formatted pixels. - ImportViewPixelArea imports formatted pixels into a cache view. * Removed C API functions: - ReadStream() - WriteStream() * Magick++ C++ API improvements - Color class no longer considers transparent black to be an invalid color. - New Image methods addNoiseChannel(), blurChannel(), gaussianBlurChannel(), motionBlur(), randomThresholdChannel(), randomThresholdChannel(), sharpenChannel(), unsharpmaskChannel(). Feature improvements: * -ordered-dither and -random-threshold may now be used to individually dither any named channel. * Mogrify and convert now support -minify to halve the image size. * Mogrify and convert now support -magnify to double the image size. 1.2 (April 29, 2008) ==================== Security fixes: * Fixes for CERT security alert TA04-217A described at "http://www.us-cert.gov/cas/techalerts/TA04-217A.html". * AVI, BMP, & DIB security fixes. * PSD security fixes. * P7 format security fix. * Fix EXIF IFD stack overflow vulnerability. * SGI security fix for RLE encoding (CVE-2006-4144) * XCF security fix (CVE-2006-3743) * PALM heap overflow fix (CVE-2006-5456) * DCM security fix (CVE-2006-5456) * Fix for shell command injection in delegate code via file names) (CVE-2005-4601). Delegate execution is much more secure now. * Don't use filenames as printf specifications (CVE-2006-0082). * Fix integer overflow in DCM coder (CVE-2007-1797). * XWD integer overflow fix (CVE-2007-1797). * Implementation has replaced usage of strcpy, strcat, and strncat with the more security conscious strlcat and strlcpy. * DCM, DIB, XCF, XBM, and XWD security fix for integer overflow vulnerability (IDefense 09.19.07). * Do not access X11 or invoke convenience or stealth delegate programs based on the file extension. In particular, these file extensions are rejected for consideration as a format specifier: 'autotrace', 'browse', 'dcraw', 'edit', 'gs-color', 'gs-color+alpha', 'gs-gray', 'gs-mono', 'launch', 'mpeg-encode', 'print', 'scan', 'show', 'win', 'xc', and 'x'. Bug fixes: * The configure script now searches for a web browser in the order mozilla, firefox, and finally netscape. * When the user specifies the -units option, the current image resolution values are now re-scaled to match the new units. * Properly determine Ghostscript font location for Ghostscript 8.0 and later. * GraphicsMagick now successfully builds and passes all tests under Digital Unix 5.1, using the vendor compiler. * Ghostscript sometimes displays an error message and fails, yet it returns a success error code to GraphicsMagick. Verify that Ghostscript has updated the output file before attempting to use it. * Fixed a configure script syntax error when testing for trio. * When requesting a list of formats, all of the modules in the module search path are considered. Previously only the modules in the same directory as the LOGO module were listed. * Ensure that an image clip mask is respected by the negate algorithm. * The BMP writer was sometimes writing incorrect BMP v4 files. * Support reading and writing large PCX files. * The Red Hat source RPM was failing to install the -config scripts with execute permissions. * Fixed a bug which could cause possible truncation while cloning the image cache. * Ensure that MIFF files indicate the compression which was actually used. * Properly handle errors from libtiff so that corrupted images are not output. * Fix for stripped-TIFF reader. Discard extra samples beyond alpha in scanline TIFFs. * Endian option now controls TIFF byte-order rather than bit-order. * TIFF writer can now write to pipes and other non-seekable output destinations. * JBIG writer was writing empty files for some libjbig releases. * Improved handling of corrupt GIF files. * Handle large SUN format images. * Properly compute image depth for 16-bit SGI image files. * For the gmdisplay program, ensure that only RGB data is sent to Windows. * Many memory leak fixes. * PDF writer is fixed so that Ghoscript 8.5 doesn't warn about the output. * PDF writer now writes proper output with CCITT compression. * Properly use fseeko() and ftello() if they are available. * Fixed a infinite loop bug in the XWD reader. * Fix minor memory leak in ProfileImage(). * Fixed -level command parsing when a percent symbol is supplied within the argument rather than at the end. * Fix pixel scaling problem caused by floating point rounding error (noticed under AIX). * Fixed a memory leak in the GIF coder in the error return path. * Fix for SourceForge bug id 1353744 "MagickGetQuantumDepth doesn't work". * Fix for SourceForge bug id 1315109 "segfault in InitializeMagick(NULL)". * Fix for SourceForge bug id 1391421 "problem doing resize on 273x1 JPEG". * Fix for SourceForge bug id 1510075 "Failed to write PDF with JPEG compression". * Fix for SourceForge bug id 1572357 "GetOnePixel definition appears incorrect". * Fix for SourceForge bug id 1576616 Fix includedir variable in pkg-config files". * Fix for SourceForge bug id 1173713 "segfault in ModifyCache". * Fix for SourceForge bug id 1431805 "clip art wpg files cause access violation in graphics magick". * Fix for SourceForge bug id 1743141 "Affine matrix option parsing". * Fix for SourceForge bug id 1625477 "Memory leak reading layered PSD Image". * Fix for SourceForge bug id 1878992 "literal square brackets in file name cause large delay and bug id 1783209 "converting runs slowly when subimage is specified". * Fix for SourceForge bug id 1883527 "compression of tiff-file has no effect". * Successfully read files in the form "file[123]". * Fix reading 12-bit grayscale JPEG. * Set image depth appropriately when importing image from X11 display. * Fix map resource tracking. * Fix reading recent variants of ImageMagick's MIFF format. * Output bilevel TIFF meeting the TIFF Class F specification. New Utilities: * A 'benchmark' subcommand is now available to benchmark the performance of any other arbitrary subcommand (e.g. 'convert'). Feature improvements: * LZW compression is now enabled by default. * Support industry-standard subsampling notation like "4:2:2". * If gm is executed under a traditional alternate name (e.g. convert), it will invoke the appropriate sub-command. This allows use of hard links, symbolic links, or just copying 'gm' to the desired sub-command name in order to achieve 100% ImageMagick 5.5.2 utility compatibility. * Provide the --enable-magick-compat option when configuring to install ImageMagick utilities compatibility links. * Identify -verbose output includes normalized (0.0-1.0) statistics. * Identify and convert now print "pixels per second" rates to help evaluate performance. * Added the identify +ping option to force reading the complete file. * The display program now supports the +progress option to disable any visual progress indication (and hourglass cursor) while loading images. * Support writing grayscale TGA files. * Provide explicit support for Rec 601 and Rec 709 grayscale spaces. * Include some support for a log RGB space based on the 2.048 density range as defined for the Cineon Digital Film System. * Added utilities command-line support for industry standard subsampling notation like 4:4:4 and 4:2:2. * Use MAGICK_IOBUF_SIZE to tune the size of the I/O buffer. * Use -type Bilevel, Grayscale, TrueColor, or TrueColorMatte to influence the type of image that Ghostscript returns. * Use '-define tiff:fill-order={msb2lsb|lsb2msb}' to control TIFF bit fill order. * The -version option now dumps a feature list as well as the build options. * The -endian option now supports the option 'native'. * A -monitor is added to enable progress monitoring for the command line utilities. * Use the -output-directory option to 'mogrify' to send output files to the specified directory. * Use the -create-directories option in conjunction with -output-directory and 'mogrify' to create any necessary subdirectories. * A Pixels resource limit is added. Use '-limit Pixels value' to limit the maximum number of pixels in an image to 'value'. * The already supported option '-type Optimize' is now honored by formats that need to choose a subformat based on the properties of the image. Grueling tests of many/all pixels are not performed unless '-type Optimize' is supplied. * Added a a -set option to the composite, convert, display, mogrify, import commands in order to allow setting an image attribute. * Display utility no longer defaults to reading from standard input if stdin is not a tty. * May now be configured to use the umem memory allocation library available in Solaris 9, Update 3 and later, or from the portable umem project. Coder additions/improvements: * Replaced existing DPX "support" with all-new DPX support conforming to the SMPTE 268M-2003 standard. * Cineon reader completely rewritten. * TIFF coder is completely re-written. Now supports reading and writing RGB, CMYK, and grayscale, scanline-oriented TIFF images with arbitrary (1 to 32 bits) depth. Includes support for tiled TIFF, floating point TIFF, LogLuv TIFF, BigTIFF, arbitrary depths, and associated alpha. * TIFF coder now supports retrieving and saving XMP profiles. * MATLAB support is much improved and supports writing as well. * WPG reader now supports CTM translations. * ART format now supports writing. * Support 32-bit raw RGB images. * Support 32-bit raw CMYK images. * Support 32-bit raw gray images. * JP2 coder reads images in YCbCr colorspace and retrieves an embedded ICC ICM color profile if present. API enhancements: * Added ExportImageChannel() and ImportImageChannel() APIs to support exporting and importing pixel regions with an arbitary range of (1 to 32) bits per quantum. * Added image leveling methods for Magick++. * Generalized GetImageAttribute() support for retrieving wildcarded attributes so that an identify -format specification like ``"%[dpx:*]"`` works as expected. * Incorporated changes changes necessary so that GraphicsMagick can work with the Ch C/C++ interpreter from SoftIntegration at http://www.softintegration.com/. * Added MagickAllocFunctions() to allow the API user to replace the underlying memory allocator functions. * Added MagickMalloc() and deprecated AcquireMemory(). * Added MagickCloneMemory() and deprecated CloneMemory(). * Added MagickMallocArray() to safely allocate N items of size S. * Added MagickRealloc() and deprecated ReacquireMemory(). * Added MagickFree() and deprecated LiberateMemory(). Performance improvments: * The DispatchImage() and ConstituteImage() functions incorporate special case code for BGR, BGRO, BGRP, RGB, RGBO, and I formats (8 bit only) in order to improve performance dramatically. * When writing very large JPEG images, don't enable Huffman compression since doing so requires libjpeg to buffer the entire image in memory. * When using the 'identify' -verbose option, -verbose must be specified twice in order to obtain the color count. This makes normal use of -verbose much faster. * Significantly improved read/write speed for bilevel and gray images. * TIFF I/O is considerably faster. * Postscript writer is 10-15X faster. * PNM formats writer is 10-100X faster. * Rotate by 90 or 270 degrees is 2-9X faster. Windows-specific improvements/changes: * For the MinGW and Cygwin builds, the Magick++ library is forced to build as a static library since otherwise C++ exceptions don't work. * MinGW cross-build is available from a Linux or FreeBSD host. * Determine location of Ghostscript fonts only once in order to improve performance. * Updated bzip2 to 1.0.4. * Updated Jasper library to version 1.900.1. * Updated jbigkit to 1.6 * Updated lcms to 1.17 * Updated libpng to 1.2.27. * Updated libtiff to 3.8.2 * Updated zlib to 1.2.3. * Libtiff supports LZW compression. * X11 is no longer part of the default build and will not be included in the distributed install packages (but can still be built). * Find latest Ghostscript which idenfies itself as "GPL Ghostscript". * Use GlobalMemoryStatusEx(), if available, to determine how much physical memory is available. Important for large-memory machines. * Fixed NTreaddir() so that it does not write beyond its buffer. * Fixed opendir() emulation function so it can't overwrite the stack. * FlashPIX library sources are no longer distributed in the Windows source package and building FlashPIX is disabled by default. FlashPIX may still be built by adding the library (separately distributed). * Fix bitmap handle leak in CropImageToHBITMAP() and ImageToHBITMAP(). 1.1 (Released April 4, 2004) ============================ Bug fixes: * Semaphore fix which is necessary for proper multi-threaded operation. * Configure script fix to ensure that -lfpx is not supplied to the C compiler during subsequent tests since this fails on some systems. * Fix for East and West gravity computations. * System error reports (errno) associated with an exception are now correctly obtained from the context existing when the exception was thrown rather than the context of the reporting function. * JNG encoder fix. Files were being written with incorrect alpha_sample_value in the header. These can be repaired by reading them into GM 1.1 and rewriting them. * XPM fix to module registration. * PSD fix for index calculation when QuantumDepth>8. * Validate that geometry specifications only include allowed characters. * SGI fix to save compression type while writing. * EXIF attributes were not being retrieved when requested. * Fix for bug when reading an image via a user-provided file descriptor. * The reported image magick string is now always that of the original input file (it was sometimes being reported as the format produced by an intermediate delegate program). * Fixes to color profiling of CMYK images. * Memory leak fixed in DrawClipPath(). * Arc drawing is fixed. * Command-line parsing bug under Linux due to Linux's sscanfs inability to parse strings like "0x1" as "%fx%f" is fixed. * Scaling of 5 and 6-bit colors was slightly incorrect in BMP, AVI, DIB, and TIM datastreams. * GM utility now reports an error rather than silently returning if an unsupported sub-command is provided. * TIFF coder was writing 16-bit per sample RGB images incorrectly on little-endian CPUs. Performance improvements: * Texture tiling is now 7X faster. * Color profile processing speed improvements for colormapped images. Utilities enhancements: * For Unix, 'gm version' now includes a dump of the configure and build parameters. * Logging of thrown exceptions is now supported. Use '-debug exception'. This is useful to learn when and where errors are reported. * The -define option is added in order to support supplying additional options to coders without needing to add additional command line options or structure members. * The output of 'gm identify -verbose' now provides a nice dump of EXIF data. * The -sampling-factor option now accepts as many HxV pairs as there are JPEG components. Omitted ones default to 1x1. * The convert and montage commands now support an -operator command to perform arithmetic and bitwise operations on specified image channels. Coder additions/improvements * The META coder supports wide characters for the IPTC and 8BIM formats. * The XTRN coder now supports wide characters. * An "IMAGE" coder is provided which provides access to a large number of images (derived from XFig) suitable for use as patterns, or as test images. * The "PATTERN" coder now returns an image pattern tiled to size (equivalent results to TILE:IMAGE:pattern). This is for ImageMagick compatibility. * The CINEON coder now supports reading and writing images in CINEON "CIN" linear gray and RGB formats. The read support is still very weak, but it works for common images. * The JPEG coder now estimates the original JPEG quality and sampling factors and will reuse these options when writing JPEG if the image is of the same type and the option "-define JPEG:preserve-settings" is supplied. * The JPEG-2000 coder now supports all Jasper library arguments using command line syntax similar to '-define jp2:rate=0.5'. * Reading and writting compressed SVG (SVGZ) is now supported. * The TXT coder now observes depth when writing. * The TIFF coder now outputs colormapped images with 1, 2, 4, and 8 bits per sample in order to provide much smaller file sizes for images with very few colors. * Many TIFF coder enhancements. Now reads colormapped and grayscale images at arbitrary (even odd) bits-per-sample sizes. Now properly supports an opacity channel (at any bits-per-sample value) for grayscale images. Bilevel grayscale images are treated similar to any other grayscale image unless CCITT FAX3/FAX4 compression is requested. Now allows the user to specify an arbitrary bits-per-sample value for grayscale images (even odd values) using "-define tiff:bits-per-sample=value". Now automatically stores the image as TrueColor RGB pixels if the image compression is set to JPEG. TIFF files are now written in using the TIFF library's default endian order rather than always big endian. * The WPG coder now renders embedded WMFs. * The PS3 coder is completely re-written to work much better and support more features (see ChangeLog). Code structure enhancements: * Error handling has been improved and validated through testing. Some errors were being lost, unnecessarily ignored, or reported as something else entirely. * The number of error text messages to be maintained has been reduced by consolidating similar messages. * The memory allocator functions have been replaced with similar macros in order to eliminate warnings with GCC 3.3, avoid accidentally casting away const, and allow memory debuggers to report memory allocations and frees against the correct functions. The previously-used functions remain in the library for the purpose of compatibility. * no longer depends on magick_config.h defines. * The text string localization code has been replaced with a simpler version written by Bill Radcliffe. * Added key,value "map" APIs (somewhat similar to C++'s ) for internal use. API enhancements: * Incorporated John Cristy's Wand API's in a new GraphicsMagickWand library. * API definition is no longer dependent on types which vary in size (e.g. size_t) depending on large file compilation options. This means that applications may now be compiled without any special large file options and still work properly with the library. * Thrown exceptions (ExceptionInfo structure) now include source file, source line, function name, and current system error number. * The GetMagickInfoArray() function is added to replace use of GetMagickInfo() for code which needs to access the coder list. This is necessary since invoking GetMagickInfo() may re-order the coder list, causing problems for code which traverses the list. Using GetMagickInfo() to access individual list elements is safe. * Added the CopyException function to support copying exception info from one structure to another. * Added the ReplaceImageInList function to replace an image in an image list. * Added the DrawPeekGraphicContext function to access the current DrawInfo structure in the drawing context stack. Use of this function is not recommended since it voilates proper programming practices. It is added to support the Wand API's. * GetImageDepth() now returns an integral value between 1 and QuantumDepth and is no longer limited to the values 8, 16, and 32. * SetImageDepth() supports setting the image modulus depth to any integral value between 1 and QuantumDepth. This effects the effective numeric precision, not the storage depth, since the quantum storage type is still a Quantum. * GetImageChannelDepth() supports retrieving the modulus depth for a specified channel. * SetImageChannelDepth() supports setting the modulus depth for a specified channel. * ProfileImage is updated to handle alpha channels and grayscale images. * Added GetImageProfile() to retrieve a CMS profile from an image. * Added SetImageProfile() to attach a CMS profile to an image without adjusting the image pixels. * Added DeleteImageProfile() to remove a CMS profile from an image. * ConstituteImage() and DispatchImage() now support 'T' (transparency), 'O' (opacity), and 'P' (pad) options. * CompositeImage() now supports CopyCyanCompositeOp, CopyMagentaCompositeOp, CopyYellowCompositeOp, and CopyBlackCompositeOp, composition operators. * GetColorHistogram() obtains a color histogram for the image. * QuantumOperatorImage() and QuantumOperatorRegionImage() support arithmetic and bitwise operations on specified image channels. * The semaphore.h header is no longer installed or included in the API headers since these functions are private interfaces. * Configure using --enable-symbol-prefix or define PREFIX_MAGICK_SYMBOLS to use the C preprocessor to prefix all library symbols with "Gm". This prevents library symbol conflicts with other libraries. PerlMagick fixes: * Adjusted a number of function option names so that they match the documentation. * Memory leak fixed. * Reading files (e.g. GIF) via a file descriptor is fixed. Build improvements: * The TRIO library may be used to provide a replacement for vsnprintf if the C library doesn't provide it. This improves security on old systems. * Configure only configures for C & C++ languages. * Configure now does a better job of figuring out how to build a thread-safe library across multiple operating systems. * Configure incorporates a new mmap() test which tests the functionality which is needed so that mmap() is not unnecessarily rejected on a number of systems. This improves performance for large files on those systems. * Configure/build fixes for IBM's AIX operating system. Windows-specific improvements: * The static install package now uses the "uninstalled" configuration so that it does not depend on the Windows registry in order to run. This allows files from the static install package to be copied to another computer without running an installer. * The executable search path is extended at run-time to include the directory where the CORE DLLs reside in order to ensure that they are found. * Adding the -t option to VisualMagick configure enables building all of the coders into one library in order to save build time and simplify linkage. * The XTRN coder now supports wide characters. * LCMS library updated to version 1.10. * GMDisplay displays a checkerboard pattern behind transparent images. * Support is provided for issuing log messages to the Windows standard logging system. * Project files are now provided for use with Borland C++ Builder 6.0. * Updated LCMS version to 1.12. * Updated FreeType version to 2.1.5. * Updated JBIG-KIT to version 1.5. * Updated libpng to version 1.2.5. * Updated libwmf to version 0.2.8.2. * Updates zlib to version 1.2.1. * ActivePerl 5.8.1 Build 807 now supported. * GraphicsMagick now compiles using Visual Studio .NET 2003. --------------------------------------------------------------------------- 1.0 (Released in May, 2003) =========================== GraphicsMagick support services: * Master web site at "http://www.GraphicsMagick.org/". * Mailing lists, bug tracking, and forums available via "http://sourceforge.net/projects/graphicsmagick/". * CVSWeb via "http://cvs.graphicsmagick.org/cgi-bin/cvsweb.cgi/". * CVS via cvs.graphicsmagick.org (find instructions at "http://www.graphicsmagick.org/www/cvs.html"). * FTP via "ftp://ftp.graphicsmagick.org/pub/GraphicsMagick". Project maintenance improvements: * ChangeLog conforms to the GNU standard and all CVS commits include useful log messages. * CVS commit messages posted to graphicsmagick-commit mail list. * CVS commit messages contain CVSWeb URL references. Footprint changes from ImageMagick: * Library -lMagick renamed to -lGraphicsMagick. * Library -lMagick++ renamed to -lGraphicsMagick++. * Utilities consolidated into a single 'gm' utility (e.g. use 'gm convert'). * Script Magick-config renamed to GraphicsMagick-config. * Script Magick++-config renamed to GraphicsMagick++-config. * Headers installed under ${PREFIX}/include/GraphicsMagick. * PerlMagick namespace renamed from "Image::Magick" to "Graphics::Magick". * Pkgconfig files GraphicsMagick.pc and GraphicsMagick++.pc are installed in $libdir/pkgconfig to assist pkg-config users. * Coder modules installed to lib/GraphicsMagick-1.0/modules-Q8/coders. * Filter modules installed to lib/GraphicsMagick-1.0/modules-Q8/filters. Many performance enhancements: * Default QuantumDepth is 8 since this is adequate for most purposes and more efficient than 16. * The Magick++ demo (compiled with QuantumDepth=8 and -O2) runs about 1.8X faster under SPARC/Solaris than the same demo with ImageMagick 5.5.4. * Colorspace transformations are much faster. * Grayscale/monochrome image handling is much faster. * PseudoClass image handling is faster. * Text annotations using FreeType are much faster. * Image file I/O is much faster. * RLE-compressed MIFF reading much faster. Code structure enhancements: * All utility support functions moved to magick/command.c in order to significantly reduce link dependencies, allowing statically-linked programs to be smaller. * Use of MogrifyImage() eliminated except for by utilities. * Re-builds due to changes to reduced by splitting the header into multiple headers. * ISO C '99 typedefs (gm_int16_t, gm_uint16_t, gm_int32_t, gm_uint32_t, gm_int64_t, gm_uint64_t) are available for use. Feature enhancements: * Module loader always enabled for shared builds to allow extension. * Loading of arbitrary filter modules (via -process option) supported under Unix as well as Windows. * SVG coder allows specifying size and initial background color. * JPEG-2000 coder (JP2) updated to work with Jasper 1.7. * HWB and HSL image colorspace translation support. * JNG/MNG/PNG format support tracks ImageMagick version. * BMP encoder can write 16-color indexed BMPs now as well as 2-color BMPs (not restricted to monochrome). * TIFF decoder now includes optimized support for tiled and stripped TIFF. * The -random-threshold option (RandomThresholdImage()) is added to threshold an image to bilevel using random thresholding. Major bug fixes: * 100% successful test completion at all quantum depths (8/16/32). * A temporary file management subsystem is added to ensure that all temporary files are removed before program exit. Temporary files are created and used in a secure fashion to avoid the possibility that a "trojan" temporary file (e.g. a symbolic link, or a file containing unsafe content) is created before a delegate has the chance to write to it. The environment variable MAGICK_TMPDIR allows the user to specify where temporary files are created without altering where other programs create their temporary files. Temporary filenames are created in 8+3 format to hopefully be more acceptable to ralcgm. * When dithering is disabled, don't dither when converting to a PseudoClass, grayscale, or monochrome image. Disabling dithering may cause these translations to be much faster. * PICON format works with BLOBs. * No longer removes input file when pinging a FlashPIX file. * Arc drawing and texture fill fixes from ImageMagick. * Sample, scale and affine fixes from ImageMagick. * MIFF colormaps are now scaled properly while reading. * CMYK translation works for QuantumDepth=32. * ConstituteImage now works properly for grayscale images. * Built-in tilde expansion and filename globbing now works properly. * InitializeMagick now registers signal handlers to ensure that resources are released before program exit. This helps avoid temporary file leaks due to the user using "CONTROL-C". * The installed header now only contains the few definitions required by the API headers. This should significantly reduce or eliminate conflicts with other package headers. Windows platform enhancements: * Configure updated for Visual C++ 7.0. * OLE object (ImageMagickObject) re-written to work with Visual C++ 7.0 * New function, CropImageToHBITMAP(), to return a region of the image as a Windows HBITMAP. * Use vsnprintf to format strings under Windows (safer). --------------------------------------------------------------------------- On November 19, 2002, GraphicsMagick was created as a fork of ImageMagick, several days before the ImageMagick 5.5.2 release. The objectives of GraphicsMagick are to: * Use an open development model. * Encourage new developers to join the project. * Avoid unnecessary source code "churn". * Establish and preserve a stable API. * Use efficient coding practices which result in fast code. * Improve memory efficiency. * Use a release process which assures a working product. * Maintain an accurate ChangeLog. --------------------------------------------------------------------------- .. |copy| unicode:: U+000A9 .. COPYRIGHT SIGN Copyright |copy| GraphicsMagick Group 2002 - 2015