- Enable Bytecode. Weird_W's fonts finally look humanly readable.
Thanks to Waxdragon for testing.
<http://www.freetype.org>
LATEST CHANGES BETWEEN 2.1.10 and 2.1.9
I. IMPORTANT BUG FIXES
- The size comparison for BDF and PCF files could fail sometimes.
- Some CFF files were still not loaded correctly. Patch from
Derek Noonburg.
- The stroker still had some serious bugs.
- Boris Letocha fixed a bug in the TrueType interpreter: The
NPUSHW instruction wasn't skipped correctly in IF clauses. Some
fonts like `Helvetica 75 Bold' failed.
- Another serious bug in handling TrueType hints caused many
distortions. It has been introduced in version 2.1.8, and it is
highly recommended to upgrade.
- FreeType didn't properly parse empty Type 1 glyphs.
- An unbound dynamic buffer growth was fixed in the PFR loader.
- Several bugs have been fixed in the cache sub-system.
- FreeType behaved incorrectly when resizing two distinct but very
close character pixel sizes through `FT_Set_Char_Size' (Savannah
bug #12263).
- The auto-hinter didn't work properly for fonts without a Unicode
charmap -- it even refused to load the glyphs.
II. IMPORTANT CHANGES
- Many fixes have been applied to drastically reduce the amount of
heap memory used by FreeType, especially when using
memory-mapped font files (which is the default on Unix systems
which support them).
- The auto-hinter has been replaced with a new module, called the
`auto-fitter'. It consumes less memory than its predecessor,
and it is prepared to support non-latin scripts better in next
releases.
- George Williams contributed code to read kerning data from PFM
files.
- FreeType now uses the TT_NAME_ID_PREFERRED_FAMILY and
TT_NAME_ID_PREFERRED_SUBFAMILY strings (if available) for
setting family and style in SFNT fonts (patch from Kornfeld
Eliyahu Peter).
- A new API `FT_Sfnt_Table_Info' (in FT_TRUETYPE_TABLES_H) has
been added to retrieve name and size information of SFNT tables.
- A new API `FT_OpenType_Validate' (in FT_OPENTYPE_VALIDATE_H) has
been added to validate OpenType tables (BASE, GDEF, GPOS, GSUB,
JSTF). After validation it is no longer necessary to check
for errors in those tables while accessing them.
Note that this module might be moved to another library in the
future to avoid a tight dependency between FreeType and the
OpenType specification.
- A new API in FT_BITMAP_H (`FT_Bitmap_New', `FT_Bitmap_Convert',
`FT_Bitmap_Copy', `FT_Bitmap_Embolden', `FT_Bitmap_Done') has
been added. Its use is to convert an FT_Bitmap structure in
1bpp, 2bpp, 4bpp, or 8bpp format into another 8bpp FT_Bitmap,
probably using a different pitch, and to further manipulate it.
- A new API `FT_Outline_Embolden' (in FT_OUTLINE_H) gives finer
control how outlines are embolded.
- `FT_GlyphSlot_Embolden' (in FT_SYNTHESIS_H) now handles bitmaps
also (code contributed by Chia I Wu). Note that this function
is still experimental and may be replaced with a better API.
- The method how BDF and PCF bitmap fonts are accessed has been
refined. Formerly, FT_Set_Pixel_Sizes and FT_Set_Char_Size
were synonyms in FreeType's BDF and PCF interface. This has
changed now. FT_Set_Pixel_Sizes should be used to select the
actual font dimensions (the `strike', which is the sum of the
`FONT_ASCENT' and `FONT_DESCENT' properties), while
FT_Set_Char_Size selects the `nominal' size (the `PIXELSIZE'
property). In both functions, the width parameter is ignored.
III. MISCELLANEOUS
- The BDF driver no longer converts all returned bitmaps with a
depth of 2bpp or 4bpp to a depth of 8bpp. The documentation has
not mentioned this explicitly, but implementors might have
relied on this after looking into the source files.
- A new option `--ftversion' has been added to freetype-config to
return the FreeType version.
- The memory debugger has been updated to dump allocation
statistics on all allocation sources in the library. This is
useful to spot greedy allocations when loading and processing
fonts.
- We removed a huge array of constant pointers to constant strings
in the `psnames' module. The problem was that compilations in
PIC mode (i.e., when generating a Unix shared object/dll) put
the array into the non-shared writable section of the library
since absolute pointers are not relocatable by nature.
This reduces the memory consumption by approximately 16KByte per
process linked to FreeType. We now also store the array in a
compressed form (as a trie) which saves about 20KByte of code as
well.
- Kirill Smelkov provided patches to make src/raster/ftraster.c
compile stand-alone again.
svn path=/trunk/; revision=16675
- Moved RtlpCreateUnicodeString to ntoskrnl for now, it however is depreciated but still used in various places
- Added RtlpAllocateMemory and RtlpFreeMemory for rtl memory allocations (from paged pool in ntoskrnl and from the process heap in ntdll) that replace the ExAllocatePool* and ExFreePool implementations in ntdll
svn path=/trunk/; revision=16673
- Implemented FindFirstFileExW() and have removed InternalFindFirstFile() in /lib/kernel32/file/find.c, now FindFirstFileA (), FindFirstFileExA () and FindFirstFileW called FindFirstFileExW ()
NOTE: Filip has asked revision "16661" (the next one) to be done by him... sorry if this sounds silly, but please respect his wish :)
svn path=/trunk/; revision=16660
- Created include/drivers and moved driver headers there.
- Brought back core.h as diskdump.h, and re-enabled diskdump...it seems even though I was told it doesn't work, it's still very much used.
svn path=/trunk/; revision=16657
- Remove core.h and disable diskdump driver (never worked, was told it's an abandonned, obsolete driver and I tend to agree.)
- Rename ROS ntifs.h to ccros.h and moved to root include directory (where all the "temporary" includes are for now).
- Moved ROS class2.h to drivers/storage/include, per my own and other's recommendation, to make it clear that it's a ros-intenral file only useful for low-level storage drivers (no 3rd party driver needs this, which is why MS doesn't export it).
svn path=/trunk/; revision=16656
* Support SxS when available (add new fiber member for this as well).
* Support FLS when available.
* Added new Fiber Members for new features in XP/2003 (GuaranteedStackBytes and FlsData)
* Support FPU State Restore/Save in switch code.
* Use new common stack/context initialization routines instead of ROSRTL.
* Change Fiber structure to use CONTEXT directly in order to speed up and generalize
context creation.
* Fix BaseFiberStartup to send right parameters.
Note: not yet fully tested with fibertest, I will do this later today and fix any bugs.
svn path=/trunk/; revision=16650