The declaration of `edge2' can be reached with `edge1 == NULL' and
`axis->edges == 0' which results in undefined behaviour.
* src/autofit/afloader.c (af_loader_load_glyph): Initialise `edge2'
after checking `axis->num_edges > 1'. `edge1 != NULL' can be assumed.
* include/freetype/config/ftheader.h (FT_COLOR_H): New macro.
* include/freetype/internal/ftobjs.h (FT_Colr_Internal): Change
type of `load_flags' to `FT_Int32'.
* include/freetype/internal/sfnt.h (TT_Load_Colr_Layer_Func): Change
type of `idx' to `FT_UInt'.
(TT_Blend_Colr_Func): Change type of `color_index' to `FT_UInt'.
* src/base/ftobjs.c (FT_Render_Glyph_Internal): Change type of
`load_flags' to `FT_Int32'.
* src/sfnt/ttcolr.c (find_base_glyph_record,
tt_face_load_colr_layers): Change type of `glyph_id' to `FT_UInt'.
(tt_face_find_color, tt_face_colr_blend_layer): Change type of
`color_index' to `FT_UInt'.
Fix signedness and type issues.
* src/sfnt/ttcolr.h: Updated.
This platform uses backslashes for paths, which docmaker didn't
understand correctly.
* src/tools/docmaker/tohtml.py (HtmlFormatter::blockEnter): Use
`os.path.normpath' to normalize the path for the platform being
used.
This generalizes magic outline shifts that make Harmony LCD
rendering work in terms of precise two-dimensional RGB subpixel
positions. These coordinates are now set in time of the `smooth'
module initialization and later used to shift a glyph outline for
rendering. FT_RENDER_MODE_LCD and FT_RENDER_MODE_LCD_V use the same
coordinates. The letter, however, rotates them before using.
The LCD bitmap padding is also calculated using these coordinates.
* include/freetype/internal/ftobjs.h (FT_LibraryRec): New array field
`lcd_geometry'.
* src/base/ftlcdfil.c (ft_lcd_padding): Reworked.
* src/base/ftobjs.c (ft_glyphslot_preset_bitmap): Updated accordingly.
* src/smooth/ftsmooth.c [!FT_CONFIG_OPTION_SUBPIXEL_RENDERING]
(ft_smooth_init): Initialize `lcd_geometry'.
(ft_smooth_render_generic): Formalize outline shifts.
* include/freetype/internal/ftobjs.h (FT_Glyph_LayerRec): Move this
structure to...
* include/freetype/freetype.h (FT_Glyph_LayerRec): ... this
header file.
(FT_Glyph_Layer): New typedef.
Update code to use it where appropriate.
* src/base/ftobjs.c (FT_Get_GlyphLayers): New function.
It is rather fundamental to set monochrome bitmap based on rounded
CBox because the b/w rasterizer turns on pixels when their centers are
inside the glyph outline. The dropout control is unpredictable and can
distort narrow glyphs if the bitmap is too wide.
Reported by Chris Liddell.
* src/base/ftobjs.c (ft_glyphslot_preset_bitmap): If BBox boundaries
are too close, adjust them before rounding.
* src/truetype/ttgload.c (TT_Load_Glyph): Free old `layers' array
before reassigning allocated memory.
Only allocate `color_layers' if we don't have one already.
* src/sfnt/ttcolr.c (tt_face_load_colr_layers): Return FT_Err_Ok if
current glyph is not a `COLR' base glyph.
* src/truetype/ttgload.c (TT_Load_Glyph): Don't allocate
`color_layers' if there are no color layers.
This commit enables OpenType's COLR/CPAL table handling; a typical
application are color emojis that can be scaled to any size.
If the color palette does not exist or is invalid, the rendering
step rasterizes the outline instead. The current implementation
assumes that the foreground is black.
Enable this by defining option TT_CONFIG_OPTION_COLOR_LAYERS.
There are still some issues with metrics; additionally, an API to
fetch color layers is missing.
* devel/ftoption.h, include/freetype/config/ftoption.h
(TT_CONFIG_OPTION_COLOR_LAYERS): New macro.
* include/freetype/internal/ftobjs.h (FT_Glyph_LayerRec,
FT_Colr_InternalRec): New structures.
(FT_Slot_InternalRec): Add `color_layers' field.
* include/freetype/internal/sfnt.h (TT_Load_Colr_Layer_Func,
TT_Blend_Colr_Func): New function types.
(SFNT_Interface): Add `load_colr', `free_colr', `load_colr_layer',
and `colr_blend' fields.
* include/freetype/internal/tttypes.h (TT_FaceRec): Add
`colr_and_cpal' field.
* include/freetype/internal/tttags. (TTAG_COLR, TTAG_CPAL): New
macros.
* src/sfnt/ttcolr.c, src/sfnt/ttcolr.h: New files.
* src/base/ftobjs.c (ft_glyphslot_done, FT_Render_Glyph_Internal):
Handle glyph color layers.
* src/sfnt/Jamfile (_sources), src/sfnt/rules.mk (SFNT_DRV_SRC): Add
`ttcolr.c'.
* src/sfnt/sfdriver.c: Include `ttcolr.h'.
(PUT_COLOR_LAYERS): New macro.
Update call to `FT_DEFINE_SFNT_INTERFACE'.
* src/sfnt/sfnt.c: Include `ttcolr.c'.
* src/sfnt/sfobjs.c (sfnt_load_face): Load `COLR' and `CPAL' tables.
(sfnt_done_face): Updated.
* src/truetype/ttgload.c (TT_Load_Glyph): Handle color layers.
AC_COMPILE_IFELSE only tries to compile a `*.c' to a `*.o'. The
Solaris Studio 12.1 through 12.5 compilers see the
`-fvisibility=hidden' flag, but ignore it with a warning of:
cc: Warning: Option -fvisibility=hidden passed to ld,
if ld is invoked, ignored otherwise
AC_LINK_IFELSE does the compile and then tries to link the result,
at which point the Solaris linker will issue an error:
ld: fatal: option '-fvisibility=hidden' is incompatible with
building a dynamic executable
If we don't use AC_LINK_IFELSE to catch the error, then configure
will fail further tests which attempt to link, such as those testing
dependencies like `libbz2'.
Also, don't try adding `-fvisibility' if we have already added
`-xldscope', just use one of them, since Sun Studio 12 and earlier
compilers only issue a warning, and don't try passing through to the
linker to generate an error, so AC_LINK_IFELSE doesn't catch them.
Tested on Solaris 11.4 beta with compiler versions:
Sun Studio 8 (Sun C 5.5)
Sun Studio 10 (Sun C 5.7)
Sun Studio 11 (Sun C 5.8)
Sun Studio 12 (Sun C 5.9)
Sun Studio 12.1 (Sun C 5.10)
Oracle Solaris Studio 12.2 (Sun C 5.11)
Oracle Solaris Studio 12.3 (Sun C 5.12)
Oracle Solaris Studio 12.4 (Sun C 5.13)
Oracle Developer Studio 12.5 (Sun C 5.14)
Oracle Developer Studio 12.6 (Sun C 5.15)
gcc 5.5.0
gcc 7.3.0
and verified the libfreetype.so.6 generated by each of those
compilers exported the same set of symbols.
* builds/unix/configure.raw: Implement it.
Problem reported by Shailesh Mistry <shailesh.mistry@hotmail.co.uk>.
* src/autofit/afshaper.c (af_shaper_buf_create,
af_shaper_buf_destroy) [!FT_CONFIG_OPTION_USE_HARFBUZZ]: Don't
allocate and free a four-byte buffer. Instead, make those functions
no-ops; the calling functions will provide a pointer to a buffer
instead.
* src/autofit/afcjk.c (af_cjk_metrics_init_widths,
af_cjk_metrics_init_blues, af_cjk_metrics_check_digits),
src/autofit/aflatin.c (af_latin_metrics_init_widths,
af_latin_metrics_init_blues, af_latin_metrics_check_digits)
[!FT_CONFIG_OPTION_USE_HARFBUZZ]: Use pointer to local variable for
`shaper_buf'.
* CMakeLists.txt: Test for CMake build directory being diffent from
source directory. Provide other parts of the build system access the
full include directory.
This is due to calling the configure script via `make' (within the
top-level `configure' wrapper script). The same can happen for all
other secondary make targets that are used to only modify the
primary one, e.g., `make setup devel'.
* builds/dos/detect.mk (emx, turboc, watcom, borlandc, borlandc16),
builds/os2/detect (visualage, watcom, borlandc, devel),
builds/unix/detect.mk (devel, lcc, unix), builds/windows/detect.mk
(visualc, watcom, visualage, lcc, mingw32, bcc32, devel-bcc,
devel-gcc): Use no-op recipe.
Reported by Kiyoshi Kanazawa:
https://lists.gnu.org/archive/html/freetype-devel/2018-05/msg00008.html
Thanks to the suggestions by Alexei and Alan Coopersmith.
* builds/unix/configure.raw: Check if "-xldscope=hidden" is
accepted, and if so, it is added to CFLAGS. This is the option
making Sun / Oracle C compilers hide the symbols from global
scope.
* include/freetype/config/ftconfig.h: Use "__global" prefix
for FT_EXPORT() macro, if SunPro C is newer than Sun ONE
Studio 8 (2003).
* builds/unix/ftconfig.in: Ditto.
* builds/vms/ftconfig.h: Ditto.
*/* [FT_CONFIG_OPTION_PIC]: Remove all code guarded by this
preprocessor symbol.
*/*: Replace `XXX_GET' macros (which could be either a function in
PIC mode or an array in non-PIC mode) with `xxx' arrays.
* include/freetype/internal/ftpic.h, src/autofit/afpic.c,
src/autofit/afpic.h, src/base/basepic.c, src/base/basepic.h,
src/base/ftpic.c, src/cff/cffpic.c, src/cff/cffpic.h,
src/pshinter/pshpic.c, src/pshinter/pshpic.h, src/psnames/pspic.c,
src/psnames/pspic.h, src/raster/rastpic.c, src/raster/rastpic.h,
src/sfnt/sfntpic.c, src/sfnt/sfntpic.h, src/smooth/ftspic.c,
src/smooth/ftspic.h, src/truetype/ttpic.c, src/truetype/ttpic.h:
Removed.
Bug introduced after release 2.8.
* src/base/ftbitmap.c (ft_bitmap_assure_buffer): We use
`FT_QALLOC_MULT', which doesn't zero out the buffer. Adjust the
bitmap copying code to take care of this fact.
This will be part of the forthcoming Unicode 11.0.
* src/autofit/afblue.dat: Add blue zone data for Mtavruli.
* src/autofit/afblue.c, src/autofit/afblue.h: Regenerated.
* src/autofit/afscript.h: Add Mtavruli standard character.
But no support new scripts (volunteers welcomed).
* src/autofit/afranges.c (af_arab_nonbase_uniranges,
af_beng_nonbase_uniranges, af_cakm_nonbase_uniranges,
af_deva_nonbase_uniranges, af_geor_uniranges,
af_gujr_nonbase_uniranges, af_mlym_nonbase_uniranges,
af_nkoo_nonbase_uniranges, af_telu_nonbase_uniranges,
af_hani_uniranges): Add new data.
* CMakeLists.txt, builds/cmake/FindHarfBuzz.cmake: Extensive
modernization measures.
* .gitignore: Add build/, as that's the example directory used in
CMakeLists.txt.
This brings up the minimum required CMake version to 2.8.12.
The installation paths follow the GNU defaults now, e.g. installing on a
64 bit host will place binaries into the lib64/ folder on e.g. Fedora.
Symbols are hidden by default (e.g. `-fvisibility=hidden' on GCC).
CMake will no longer look for a C++ compiler.
Library and .so version now match the Autotools build.
Comments in the build file and informational messages now use platform
agnostic example commands.
ftoption.h and ftconfig.h are written directly without a redundant
`-new' copy.
External dependencies are expressed as option()s and will turn up as
such in cmake-gui.
Internal: Properties such as dependencies and include directories are
now privately set on the freetype library instead of globally.
The CPack definitions have been cleaned up, the `make dist' has been
removed. Source packages generated with CPack don't contain Autotools
files and aren't used by the maintainters anyway.
On Windows, src/base/ftver.rc is compiled to decorate the library with
version and copyright information.
A pkg-config file is now generated and installed.
Reported as
https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=7371
* src/cff/cffload.c (cff_load_private_dict): Sanitize
`priv->blue_shift' and `priv->blue_fuzz' to avoid overflows later
on.
* src/type1/t1load.c (T1_Open_Face): Ditto.
This fixes the `make refdoc' using Cygwin: $(CAT) is `type' on this
platform, and this program only understands backslashes in paths.
Reported by Nikhil Ramakrishnan <ramakrishnan.nikhil@gmail.com>.
Do not limit DLL builds to MinGW, since we already have
`__declspec(dllexport)' directives in `ftconfig.h'.
Also suppress more warnings for POSIX functions.
It is now widely recommended that ELF shared libraries hide symbols
except those with explicit __attribute__((visibility("default"))).
This is supported by all major compilers and should rather be an
option in libtool.
* builds/unix/configure.raw: Add -fvisibility=hidden to CFLAGS.
* builds/unix/ftconfig.in, builds/vms/ftconfig.h,
include/freetype/config/ftconfig.h (FT_EXPORT): Use visibility
attribute.
* src/truetype/ttgxvar.c (ft_var_to_normalized): Don't emit number
of coordinates.
(TT_Get_MM_Var): Trace instance indices names.
(TT_Set_Var_Design): Updated.
* src/tools/apinames.c (PROGRAM_VERSION): Set to 0.3.
(OutputFormat): Add `OUTPUT_GNU_VERMAP'.
(names_dump): Handle it.
(usage): Updated.
(main): Handle new command line flag `-wL'.
* builds/unix/configure.raw: Check for resource compiler.
* builds/unix/unix-cc.in: Conditionally set up resource compiler.
* builds/freetype.mk: Add conditional rule for `ftver.rc'.
* src/base/ftver.rc: Copyright notice and year update.
=======================
Tag sources with `VER-2-9'.
* docs/VERSION.TXT: Add entry for version 2.9.
* README, Jamfile (RefDoc), builds/windows/vc2005/freetype.vcproj,
builds/windows/vc2005/index.html,
builds/windows/vc2008/freetype.vcproj,
builds/windows/vc2008/index.html,
builds/windows/vc2010/freetype.vcxproj,
builds/windows/vc2010/index.html,
builds/windows/visualc/freetype.dsp,
builds/windows/visualc/freetype.vcproj,
builds/windows/visualc/index.html,
builds/windows/visualce/freetype.dsp,
builds/windows/visualce/freetype.vcproj,
builds/windows/visualce/index.html,
builds/windows/ftver.rc,
builds/wince/vc2005-ce/freetype.vcproj,
builds/wince/vc2005-ce/index.html,
builds/wince/vc2008-ce/freetype.vcproj,
builds/wince/vc2008-ce/index.html: s/2.8.1/2.9/, s/281/29/.
* include/freetype/freetype.h (FREETYPE_MINOR): Set to 9.
(FREETYPE_PATCH): Set to 0.
* builds/unix/configure.raw (version_info): Set to 22:0:16.
* CMakeLists.txt (VERSION_PATCH): Set to 1.
* builds/unix/configure.raw: Call AC_USE_SYSTEM_EXTENSIONS.
(LIB_CLOCK_GETTIME): Define; this will hold `-lrt' if necessary.
* builds/unix/unix-cc.in (LIB_CLOCK_GETTIME): New variable.
According to the CFF specification, charstrings can have up to 96 stem
hints. Due to hint replacement routines in Type 1 charstrings, some
glyphs are rejected by the Adobe engine, which implements the above
limit. This fix turns off hinting for such glyphs.
* src/psaux/pshints.c (cf2_hintmap_build): Reset the error from calling
`cf2_hintmask_setAll' on a problematic Type 1 charstring and turn off
hinting.
This is necessary in case the application's memory routines differ
from FreeType. A typical example is a Python application on Windows
that calls FreeType compiled as a DLL via the `ctypes' interface.
* include/freetype/ftmm.h, src/base/ftmm.c (FT_Done_MM_Var): Declare
and define.
* docs/CHANGES: Updated.
Behdad reported that setting blend coordinates, then getting design
coordinates did incorrectly return the default instance's
coordinates.
* src/truetype/ttgxvar.c (tt_set_mm_blend): Fix it.
Problem reported by Hin-Tak.
* builds/windows/vc2010/freetype.vcxproj: Add files `ftbdf.c' and
`ftcid.c'.
Sort entries.
* builds/windows/vc2010/freetype.vcxproj.filter: Ditto.
Fix members of `FT_MODULE' group.
* include/freetype/internal/ftpsprop.h: Use `FT_BASE_CALLBACK'.
(ps_property_get): Harmonize declaration with corresponding
function typedef.
* include/freety[e/internal/fttrace.h: Add `trace_psprops'.
* src/base/ftpsprop.c: Include necessary header files.
(FT_COMPONENT): Define.
(ps_property_set): Tag with `FT_BASE_CALLBACK_DEF'.
(ps_property_get): Tag with `FT_BASE_CALLBACK_DEF'.
Harmonize declaration with corresponding function typedef.
We exit early if the current design or blend coordinates are
identical to the new ones.
* src/truetype/ttgxvar.c (tt_set_mm_blend, TT_Set_Var_Design):
Implement it, returning internal error code -1 if there will be no
variation change.
* src/type1/t1load.c (t1_set_mm_blend): Ditto.
* src/base/ftmm.c (FT_Set_Var_Design_Coordinates,
FT_Set_MM_Blend_Coordinates, FT_Set_Var_Blend_Coordinates): Updated.
The subsetted demo font of the report that exhibits the bug has a
very unusual type 2 cmap for Unicode(!): It contains only two
sub-headers, one for one-byte characters (covering the range 0x20 to
0xFA), and a second one for higher byte 0x01 (just for character
code U+0131).
Before this commit, the iterator wasn't able to correctly handle a
sub-header for higher byte 0x01.
* src/sfnt/ttcmap.c (tt_cmap2_char_next): Fix character increment
for outer loop.
* src/truetype/ttgload.c (TT_Process_Simple_Glyph): Always adjust
`pp1' to `pp4', except if we have an HVAR and/or VVAR table.
* src/truetype/ttgxvar.c (TT_Vary_Apply_Glyph_Deltas): Handle
alternative code branch identically w.r.t. presence of an HVAR
and/or VVAR table.
* builds/detect.mk (dos_setup): Don't use literal `>' character at
all. Mixing the different escaping rules from make, dos, and
windows is too fragile.
Stuff like
{
<bla>
}
confused the parser, which incorrectly treated `<bla>' as a markup
tag.
* src/tools/docmaker/content.py (ContentProcessor::process_content):
Apply `re_markup_tags' only outside of code sections.
This reduces redundancy and increases synergy; it also reduces the
number of header files.
* include/freetype/config/ftheader.h (FT_DRIVER_H): New macro.
(FT_AUTOHINTER_H, FT_CFF_DRIVER_H, FT_TRUETYPE_DRIVER_H,
FT_PCF_DRIVER_H, FT_TYPE1_DRIVER_H): Make them aliases to
FT_DRIVER_H.
* include/freetype/ftautoh.h, include/freetype/ftcffdrv.h,
include/freetype/ftpcfdrv.h, include/freetype/ftt1drv.h,
include/freetype/ftttdrv.h: Replaced with...
* include/freetype/ftdriver.h: ...this new file.
(FT_CFF_HINTING_ADOBE, FT_T1_HINTING_ADOBE): Renamed to...
(FT_HINTING_ADOBE): ... this new macro.
(FT_CFF_HINTING_FREETYPE, FT_T1_HINTING_FREETYPE): Renamed to...
(FT_HINTING_FREETYPE): ... this new macro.
* src/*/*: Updated accordingly.
Also reported as
https://bugs.chromium.org/p/chromium/issues/detail?id=791317
* src/base/ftbitmap.c (ft_bitmap_assure_buffer): If increasing the
bitmap size needs a larger bitmap buffer, assure that the new memory
areas are initialized also.
* include/freetype/internal/fttrace.h: Remove unused tracing macros.
s/pshalgo2/pshalgo/.
Add `trace_cffdecode'.
* src/pshinter/pshalgo.c (FT_COMPONENT): Updated.
* src/cff/cffload.c: Include FT_INTERNAL_POSTSCRIPT_AUX_H.
* src/cff/cffobjs.c: Include FT_SERVICE_METRICS_VARIATIONS_H and
FT_SERVICE_CFF_TABLE_LOAD_H.
* src/cid/cidriver.c: Include FT_INTERNAL_POSTSCRIPT_AUX_H.
* src/psaux/cffdecode.c: Include FT_FREETYPE_H and
FT_INTERNAL_DEBUG_H.
(FT_COMPONENT): Define.
* src/psaux/cffdecode.h: Include FT_INTERNAL_POSTSCRIPT_AUX_H.
* src/psaux/psauxmod.h: Include FT_INTERNAL_POSTSCRIPT_AUX_H.
Declare `cff_builder_funcs' and `ps_builder_funcs'.
* src/psaux/psft.c: Include `psobjs.h' and `cffdecode.h'.
* src/psaux/psobjs.c : Include `psauxmod.h'.
In version 2.6.1, this has been disallowed to better reject
malformed fonts; however, this restriction was too strong. This
time, we only take the first 256 elements into account, since
encoding arrays are always accessed with a 8bit integer, according
to the PostScript Language Reference.
* src/type1/t1load.c (parse_encoding): Implement it.
* builds/unix/configure.raw (ftoption_set, ftoption_unset): New
auxiliary functions to construct...
(FTOPTION_H_SED): ... this new variable.
Apply it as a sed argument while copying `ftoption.h' to the
`builds/unix' directory (using `AC_CONFIG_FILES').
Simplify code of test that checks cpp's computation of bit length
(the test previously created an empty `ftoption.h' file and deleted
it immediately afterwards); without this change, it can happen on my
GNU/Linux box that `configure's execution of `config.status' doesn't
create `ftoption.h' (no idea why this happens).
* builds/unix/install.mk (install): Install
`builds/unix/ftoption.h'.
* builds/unix/unix-def.in (DISTCLEAN): Updated.
* builds/unix/.gitignore: Updated.
Some static function declarations cause unused function warnings if
certain config options are turned off via `ftoption.h'.
* src/base/ftbase.h, src/base/ftrfork.c, src/sfnt/ttbdf.h,
src/truetype/ttgxvar.h: Add #ifdef guards around these sections.
Glyph advance widths were being written to the new `PS_Decoder' but not
saved to the underlying format specific decoder. This caused pure CFF
fonts to have bad advance width.
* include/freetype/internal/psaux.h (PS_Decoder): Change `glyph_width'
field to pointer.
Remove unused fields.
* src/psaux/psobjs.c (ps_decoder_init): Change `glyph_width' from copy
to reference.
Remove unused.
* src/psaux/psft.c (cf2_setGlyphWidth): Update code.
Since Ewald Hew factored the Adobe hinting engine out of the CFF
driver code, we can now use it on Type 1 (and CID) font formats, as
both have the same hinting philosophy.
This change activates the Adobe hinter when in LIGHT mode, and
therefore always unless explicitly asking for the auto-hinter. This
makes LIGHT behavior consistent with CFF fonts. As of this commit,
the hinting engine table looks as follows.
LIGHT NORMAL
-------------------------
TrueType Auto v40
CFF Adobe Adobe
Type 1 Adobe Adobe
The interpreter in Type 1 mode rewinds the charstring after collecting
all hints for building the initial hintmap (commit d52dd7f). However,
some charstrings use `endchar' in a final subroutine call, rewinding to
the start of that subroutine, and only a small section of the actual
glyph is drawn.
* src/psaux/psintrp.c (cf2_interpT2CharString) <cf2_cmdENDCHAR>:
Ensure we are on the top level charstring before rewinding.
See the report by Yang Yinsen.
https://lists.gnu.org/archive/html/freetype-devel/2017-11/msg00000.html
* src/truetype/ttobjs.c (trick_names): Add `DFGothic-EB',
`DFGyoSho-Lt', `DFHSGothic-W5', `DFHSMincho-W3' and `DFHSMincho-W7'.
(tt_check_trickyness_sfnt_ids): Add checksums for DFGothic-EB,
DFGyoSho-Lt, DFHSGothic-W5, DFHSMincho-W3 and DFHSMincho-W7. Also
add checksums for DLCLiShu and DLCHayBold which their family names
were already listed but their checksums were previously unknown.
We used to split large glyphs into horizontal bands and continue
bisecting them still horizontally if that was not enough. This is
guaranteed to fail when a single scanline cannot fit into the
rendering memory pool. Now we bisect the bands vertically so that
the smallest unit is a column of the band height, which is guranteed
to fit into memory.
* src/smooth/ftgrays.c (gray_convert_glyph): Implement it.
At large sizes almost but not exactly horizontal segments can quickly
drain the rendering pool. This patch at least avoids filling the pool
with trivial cells. Beyond this, we can only increase the pool size.
Reported, analyzed, and tested by Colin Fahey.
* src/smooth/ftgrays.c (gray_set_cell): Do not record trivial cells.
Werner Lemberg