Grigori Goronzy [Mon, 9 Aug 2010 00:55:30 +0000 (02:55 +0200)]
Fix drawing leakage: delay glyph allocation
Delay allocation of the "faux" glyph until a drawing is parsed.
This helps with fixing a (pretty bad) memory leak and also reduces frame
overhead a little.
Grigori Goronzy [Sun, 1 Aug 2010 21:52:30 +0000 (23:52 +0200)]
Fall back to first charmap as last resort
If we're already after the font reselection stage and a glyph cannot be
found in a font, try again with the first charmap as a last resort.
This is needed for old "symbol" fonts, for example.
Grigori Goronzy [Sun, 1 Aug 2010 01:30:59 +0000 (03:30 +0200)]
Limit (bitmap) glyph cache size
Now that bitmap glyphs (for vector clip masks) are stored in the glyph
cache, make sure it cannot grow indefinitely easily. Similar to the
bitmap cache, track approximate size of the cached elements and reset
if the cache exceeds a certain limit.
Also, reduce default bitmap cache size to 30 MB as we have essentially
two bitmap caches now. That's still plenty in all use cases where
caching matters.
Grigori Goronzy [Thu, 29 Jul 2010 00:38:29 +0000 (02:38 +0200)]
Fix drawing cbox calculation
Use extremes of all points fed to the parser, instead of only points
that are added to the outline.
Fixes a rendering problem reported on IRC and should fix Issue #7.
Grigori Goronzy [Thu, 22 Jul 2010 17:02:12 +0000 (19:02 +0200)]
Better PAR correction
PAR correction so far merely applied a horizontal scaling factor to
glyphs right after loading. This screws up transformed text,
unfortunately.
It's a much better approach to typeset text without correction and scale
everything at the very end. In MPlayer, an EOSD rendering now looks
exactly the same as a video filter rendering (except for more sharpness
in the EOSD case).
Grigori Goronzy [Sat, 26 Jun 2010 23:25:56 +0000 (01:25 +0200)]
Reduce number of glyph copies
Do not copy glyphs when it isn't really needed.
Profiling with valgrind's callgrind showed that FT_Glyph_Copy takes a
big fraction of the rendering time (~40%) for "simple styled softsubs"
cases. It turned out that glyphs are duplicated much more often than
is necessary. Reducing the number of copies cuts down the rendering
time for the "simple styled softsubs" case by about 70%!
Grigori Goronzy [Sat, 26 Jun 2010 03:15:18 +0000 (05:15 +0200)]
Add fixups for various font defects
Reverse "outside" contours with wrong winding correction, require
that a contour is "inside" for considering its removal; Move this
hack into ass_font.c, where it belongs.
Grigori Goronzy [Sat, 26 Jun 2010 00:17:03 +0000 (02:17 +0200)]
Improve fullname matching
Only match outline fonts and take slant and weight into consideration.
This matches all font faces which have at least the slant/weight that is
requested. This fixes issue 23; however, corner cases can be constructed
in which wrong fonts will be matched, if multiple slant/weight variants
of the same font exist, but that is very unlikely in practice.
Grigori Goronzy [Sun, 30 May 2010 23:38:11 +0000 (01:38 +0200)]
Add API function to flush all events
Add a function to flush all events in a track and increment
LIBASS_VERSION to indicate the addition.
Initial patch by Aurelien Jacobs (aurel AT gnuage DOT org).
Grigori Goronzy [Sat, 17 Apr 2010 23:28:25 +0000 (01:28 +0200)]
Use first Windows charmap as fallback
Extend the charmap selection heuristics to select the first Windows
charmap if no Unicode charmap is found.
Fixes one particular font, FGP楷書体NT-M, and possibly others.
Grigori Goronzy [Sat, 17 Apr 2010 20:59:47 +0000 (22:59 +0200)]
Support multiple faces per attachment correctly
Store the real font face id instead of using 0 all the time and use the
real font face id for font lookup as well. This makes font attachments
with multiple faces work correctly.
Grigori Goronzy [Thu, 18 Feb 2010 04:27:16 +0000 (05:27 +0100)]
Basic @font support
Do not skip '@' at the start of a font name in styles; detect '@' at
font name start and set a new attribute in ASS_Font accordingly.
Rotate affected glyphs after loading and calculate a suitable advance.
Grigori Goronzy [Sat, 10 Apr 2010 02:02:36 +0000 (04:02 +0200)]
Ignore global advance width unconditionally
Some (arguably) broken CJK fonts claim to have a fixed advance width,
but they actually use two different widths for half-width and full-width
characters. Even worse, some use proportional width.
This can cause problems under certain circumstances. A big project
(Cairo) uses a FreeType load flag to ignore the global advance width
for such fonts, so I assume it can be used safely. Let's do it just
like them.
Grigori Goronzy [Wed, 10 Feb 2010 22:51:11 +0000 (23:51 +0100)]
fontconfig: consider fullname for matching fonts
Additionally match fonts via the fullname (or name for humans) and
prefer these matches. Previously, libass matched against the family name
only and this name can be completely different from the fullname (which
is used by VSFilter!).
Grigori Goronzy [Thu, 21 Jan 2010 05:31:45 +0000 (06:31 +0100)]
Skip border generation if glyph is collapsed
If a character is completely collapsed in x or y direction (\fscx0,
\fscy0), the rasterizer will not generate any visible pixels. This also
means vsfilter will never draw any border in this case, since it works
on the pixel data. FreeType's stroker OTOH happily draws a border around
a character, no matter how collapsed it is. Emulate vsfilter's behavior
by skipping border generation if the glyph is collapsed.
Grigori Goronzy [Mon, 18 Jan 2010 23:42:22 +0000 (00:42 +0100)]
Up size limit for script files
Change size limit from 10MB to 50MB and clean up code a bit. A limit
still might be handy if one selects a very big file for subtitles by
accident. 50MB should be enough for even the biggest karaoke files.
Grigori Goronzy [Wed, 6 Jan 2010 22:51:01 +0000 (23:51 +0100)]
Improve rotation cache accuracy
In some cases 16.16 precision is not good enough. Instead use 10.22 and
use modulo 360.0 on the angles to make overflows impossible and improve
cache hit ratio sometimes.
Grigori Goronzy [Mon, 4 Jan 2010 18:54:19 +0000 (19:54 +0100)]
Reset clip_mode on new events
Make sure that the clip mode is reset to regular on new events.
Otherwise events after an event that used inverse clip will also use
inverse clip. Ouch! Usually the symptom is that subtitles become
invisible as the whole screen is clipped.
Grigori Goronzy [Sun, 3 Jan 2010 13:54:02 +0000 (14:54 +0100)]
Replace strtod with locale-independent strtod
strtod respects the locale and in some locales, the decimal separator is
not a point, leading to parsing errors in tags like \pos(23.4,5), which
are perfectly valid.
As there isn't a really portable way to use a particular locale just for
one call to strtod, reimplement it. The implementation was taken from
the 1.8 branch of Ruby.
Grigori Goronzy [Sun, 11 Oct 2009 07:07:26 +0000 (09:07 +0200)]
Fix libass w/o fontconfig
Make sure that libass does not crash even if no default font is
provided. Additionally, fix crashes related to drawings in this and
possibly other unusual cases.
Document the default_path and default_family parameters of
ass_set_fonts.
Grigori Goronzy [Thu, 10 Sep 2009 02:47:21 +0000 (04:47 +0200)]
Render NBSP (\h) as normal space
Many fonts don't have a non-breaking space (NBSP) character and this
can cause problems when these glyphs are substituted. Render them
like a regular space.
Grigori Goronzy [Tue, 8 Sep 2009 20:21:43 +0000 (22:21 +0200)]
Disable kerning by default and make it configurable
Kerning causes problems with many (broken) fonts. VSFilter doesn't use
kerning so these problems won't be apparant, while libass uses kerning
by default.
For compatibility reasons, disable kerning by default. In addition,
make it configurable through style overrides and the Script Info
header.
I.e. put "Kerning: yes" into the Script Info header to enable kerning.
With MPlayer, using "-ass-force-style Kerning=yes" will always
enable kerning.
This also includes a minor ABI change for tracking the kerning boolean.
As this only affects code that creates or modifies ASS_Tracks manually,
and I'm not aware of anyone doing that, the SONAME will not be bumped.