src/base/ftlcdfilt.c, src/smooth/ftsmooth.c: API change for
the LCD filter, the FT_LcdFilter value is a enum describing
which filter to apply, new values FT_LCD_FILTER_LIGHT and
FT_LCD_FILTER_LEGACY (the latter implements the LibXft original
algorithm which produces incredible color fringes for everything
except very-well hinted text)
* src/autofit/aflatin.c: various tiny improvements that drastically
improve the handling of serif fonts and of LCD/LCD_V hinting modes.
src/base/ftobjs.c: fixed a bug in the automatic unpatented
hinting support which prevented normal bytecode hinting to
work properly
* src/autofit/aftypes.h: undefining AF_DEBUG to get rid of
traces
for an upcoming 2.2.2 release
* include/freetype/ftlcdfil.h, src/freetype/ftlcdfil.c:
added a new API to support color filtering of subpixel glyph
bitmaps. In default build, the function FT_Library_SetLcdFilter
returns FT_Err_Unimplemented_Feature; you need to #define
FT_CONFIG_OPTION_SUBPIXEL_RENDERING in ftoption.h to compile
the real implementation
* src/smooth/ftsmooth.c: adding support for sub-pixel color
filtering; simplifying a few function calls
* include/freetype/config/ftheader.h: adding FT_LCD_FILTER_H
macro that points to <freetype/ftlcdfil.h>
I. IMPORTANT BUG FIXES
- Various integer overflows have been fixed.
- PFB fonts with MacOS resource fork weren't handled correctly on
non-MacOS platforms.
- The PCF font loarder has been seriously hardened against malformed
font files.
II. IMPORTANT CHANGES
- the unpatented hinter is now part of the default build of the
library, and we added code to automatically support "tricky"
fonts that need it.
what this means is that FreeType should "just work" with certain
Asian fonts, like MingLiu, which cannot properly load without a
bytecode interpreter, but fortunately do not use any of the
patented bytecode opcodes.
Note that the API didn't change, so you can still force
unpatented hinting with a special parameter to FT_Open_Face
as well.
if you're an embedded systems developer, you might want to
*disable* the feature to save code space by undefining
TT_CONFIG_OPTION_UNPATENTED_HINTING in ftoption.h.
- LCD-optimized rendering is now disabled in all default builds
of the library, mainly due to patent reasons. For more information
see:
http://lists.gnu.org/archive/html/freetype/2006-09/msg00064.html
a new configuration macro FT_CONFIG_OPTION_SUBPIXEL_RENDERING has
been introduced in ftoption.h; manually define it in this file
if you want to re-enable the feature.
the change only affects the implementation, not the FreeType API.
This means that clients don't need to be modified, because the
library still generates LCD decimated bitmaps, but with the added
constraint that R=G=B on each triplet.
- Some computation bugs in the TrueType bytecode interpreter were found,
which allow us to get rid of very subtle and rare differences we had
with the Windows renderer.
III. MISCELLANEOUS
- TrueType glyph loading is now about 25% faster.
- the anti-aliased rasterizer has been optimized and is now 15% to 25%
percent faster than the previous one, depending on content
- the Type 1 loader has been improved; as an example, it now skips
over top-level dictionaries properly
======================================================================
Fix miscellaneous compiler warnings.
* freetype2/include/freetype/internal/ftobjs.h: close
comment with `*/' to avoid `/* in comment' compiler warning.
* freetype2/src/base/ftdbgmem.c (ft_mem_table_get_source): Turn
cast `(FT_UInt32)(void*)' into `(FT_UInt32)(FT_PtrDist)(void*)'
since on 64-bit platforms void* is larger than FT_UInt32.
* freetype2/src/base/ftobjs.c (t_validator_error): cast
away volatileness of argument to ft_longjmp. Spotted by
Werner `Putzfrau' Lemberg.
* freetype2/src/bdf/bdflib.c (bdf_load_font): initialize
local variable `lineno'.
* freetype2/src/gxvalid/gxvmod.c (classic_kern_validate):
mark local variable `error' volatile.
include/freetype/internal/tttypes.h, src/base/ftgloadr.c,
src/base/ftobjs.c, src/truetype/ttgload.c, src/truetype/ttinterp.c,
src/truetype/ttobjs.c: improvements to native TrueType hinting,
this is a first try, controlled by the FIX_BYTECODE macro in
src/truetype/ttinterp.c
add T1_TOKEN_TYPE_KEY. (struct T1_FieldRec_) add `dict'. Add macros
T1_FIELD_DICT_FONTDICT and T1_FIELD_DICT_PRIVATE. Change T1_NEW_XXX and
T1_FIELD_XXX macros to take the dictionary where the PS keywords is
expected as an additional argument.
* freetype2/src/cid/cidload.c: (T1_FieldRec): Adjust invocations
of T1_FIELD_XXX.
* freetype2/src/cid/cidtoken.h: Adjust invocations of T1_FIELD_XXX.
* freetype2/src/psaux/psobjs.c: Add macro FT_COMPONENT for tracing.
(ps_parser_to_token): Report a PostScript key as T1_TOKEN_TYPE_KEY
instead as T1_TOKEN_TYPE_ANY. (ps_parser_load_field): Make sure a token
that should be a string or name is a string or name indeed. Avoid
memory leak if a keyword has been already encountered and its value
is overwritten.
* freetype2/src/type1/t1load.c: (T1_FieldRec): Adjust invocations of
T1_FIELD_XXX. (parse_dict): Ignore keywords that occur in the wrong
dictionary (e.g., in Private instead of FontDict).
* freetype2/src/type1/t1tokens.h: Adjust invocations of T1_FIELD_XXX.
src/psaux/t1cmap.c, src/psnames/psmodule.c: Fix for the memory
leak described in bug #16759.
We change 'ps_unicodes_init' so that it also takes a 'free_glyph_name'
callback to release the glyph names returned by 'get_glyph_name'
src/autofit/afcjk.c, src/autofit/afhints.c, src/autofit/aflatin.c,
src/autofit/aftypes.h, src/base/ftcalc.c, src/base/ftoutln.c,
src/gzip/ftgzip.c, src/psaux/psconv.c, src/truetype/ttgload.c,
src/type1/t1gload.c:
this is a major patch used to drastically improve the performance
of loading glyphs. This both speeds up loading the glypn vector
themselves and the auto-fitter.
note that we've started using inline assembler with GCC to
implement FT_MulFix, given that this function is so damn
important for the engine's performance.
the resulting speed-up is about 25%.
src/base/ftmac.c, src/base/ftrfork.c, src/lzw/ftzopen.c,
src/raster/ftrend1.c, src/sfnt/ttpost.c, src/truetype/ttgxvar.c,
src/type42/t42parse.c, src/winfonts/winfnt.c: hardening the code
against out-of-bounds conditions when allocating arrays. This is
for the cases where FT_NEW_ARRAY and FT_RENEW_ARRAY are not used
already. Introducing the new FT_ALLOC_MULT and FT_REALLOC_MULT
macros.
include/freetype/internal/ftmemory.h, src/base/ftdbgmem.c,
src/base/ftutil.c: udpating the memory management functions and
macros to safely deal with array size buffer overflows, this
corresponds to attemps to allocate arrays that are too large. For
an example, consider the following code:
count = read_uint32_from_file();
array = malloc( sizeof(Item) * count );
for ( nn = 0; nn < count; nn++ )
array[nn] = read_item_from_file();
if 'count' is larger than FT_UINT_MAX/sizeof(Item), the multiplication
will overflow and the array allocated will be smaller than the data
read from the file. In this case, the heap will be trashed, and this
can be used as a denial-of-service, or make the engine crash later.
the FT_ARRAY_NEW and FT_ARRAY_RENEW macro now check that the new
count is no more than FT_INT_MAX/item_size, otherwise, a new error,
named 'FT_Err_Array_Too_Large' will be returned.
note that the memory debugger now works again when FT_DEBUG_MEMORY
is defined, and FT_STRICT_ALIASING has disappeared, the corresponding
code being now the default.
Werner Lemberg
David Turner
Jens Claudius