Stefano Pigozzi [Thu, 5 Dec 2013 18:07:02 +0000 (19:07 +0100)]
coretext: also lazy load fonts based on Family and PostScript names
Previously, the lazy load of fonts was only using display name. Also use the
other names available through the CoreText API (FamilyName and PostScriptName).
Stefano Pigozzi [Sat, 2 Nov 2013 21:01:37 +0000 (22:01 +0100)]
fontselect: coretext: allow to match fontname using the provider
Not all APIs cache everything the same way that fontconfig does. This allows
to first perform a match based on the font name and then score the matched
fonts using the common code using and in memory database approach.
The benefit is the application doesn't have to load all of the fonts and
query for weight, slant, width, path and fullnames.
I left both code paths inside ass_coretext.c. This allows to test matching
problems and have a term of comparison with the slower implementation.
To activate it one just has to flip the CT_FONTS_EAGER_LOAD define to 1.
Here are some benchmarks with a pretty typical OS X font library of ~600 fonts
and using Libass's test program to load a script with 'Helvetica Neue':
CT_FONTS_EAGER_LOAD=0
0.04s user 0.02s system 79% cpu 0.081 total
CT_FONTS_EAGER_LOAD=1
0.12s user 0.06s system 44% cpu 0.420 total
Stefano Pigozzi [Sat, 2 Nov 2013 11:07:07 +0000 (12:07 +0100)]
fontselect: coretext: allow selection based on PostScript name
Up until now fontselect used the face index to identify which font to load
from a font collection. While this pretty convenient when using something
freetype based like fontconfig, it seems to be somewhat freetype specific.
CoreText uses the PostScript name as the unique identifier of a font. This
commit allows to use that instead of the index to decide which face to open
with FT_New_Face. To use the PostScript name the provider must return a -1
index and the PostScript name.
Stefano Pigozzi [Fri, 1 Nov 2013 14:44:25 +0000 (15:44 +0100)]
fontselect: implement a coretext font provider
Fontconfig is known to be very slow on OS X and Windows, this has to do with
the extremely prohibitive cache times (which are getting even longer with
latest versions of Fontconfig).
This commits starts to address the problem by using CoreText on OS X to load
the font data. The commit uses the simplest possible approach to load all of
the data in memory and then use it to match. This causes a somewhat slow
startup time (around ~400ms on my i7) but it is already better than waiting
*minutes* for Fontconfig to cache the fonts data.
A later commit will improve the speed of the match by using a hybrid approach
that lazy loads in the libass database only the necessary fonts.
Grigori Goronzy [Sun, 4 Sep 2011 13:54:38 +0000 (15:54 +0200)]
Support multiple font family names
Some fonts use localized family names, especially CJK fonts, which
often have English and Japanese or Chinese names. Handle these cases
just like full names.
Grigori Goronzy [Sat, 20 Aug 2011 16:19:25 +0000 (18:19 +0200)]
Export font provider interface
Add wrapper to the ASS_Renderer to create a font provider from
its internal font selector and shuffle some code around to export
everything that's needed for font providers to the public. Document
font provider functions.
Grigori Goronzy [Fri, 19 Aug 2011 02:36:05 +0000 (04:36 +0200)]
Add convenience defines for slant values
There is no standard scale for slant. This is almost a boolean
attribute. However, a font can have a real italic variant, and/or a
simple oblique variant. fontconfig's notation supports both of these,
so it makes sense to reuse that notation for the sake of flexibility;
we might need to differentiate between them.
Grigori Goronzy [Thu, 18 Aug 2011 23:16:24 +0000 (01:16 +0200)]
Use TrueType font weight scale
fontconfig uses an unusual scale from 0-215 for the font weight. It
looks like it is somewhat derived from the typographic scale some font
families use, but is still rather nonstandard. Nowadays the TrueType
scale from 100-900 seems to be standard. CSS uses it, for example.
However, most importantly, VSFilter also uses the TrueType scale. So
let's use it in libass, too.
Grigori Goronzy [Thu, 18 Aug 2011 04:19:33 +0000 (06:19 +0200)]
Restore fontconfig runtime configuration
Pass the fontconfig configuration file option and enable switch
through into the font selector. This restores some of the old
functionality related to fontconfig.
However, the functionality to delay the fontconfig database update will
not come back. This is not a big problem. Later it will be possible to
manually add the fontconfig provider, which will delay the update in a
comparable way.
Grigori Goronzy [Thu, 18 Aug 2011 03:33:00 +0000 (05:33 +0200)]
Add reference to font provider in font database
This provides more flexibility than just referencing the callbacks:
we can identify the font provider (useful for removing fonts when a
provider is freed) and possibly access the font provider private data.
Grigori Goronzy [Wed, 17 Aug 2011 22:15:57 +0000 (00:15 +0200)]
Add glyph coverage map for embedded fonts
Introduce a simple glyph coverage map (created when the font is added)
and use it for checking glyph coverage in font selection. This uses a
simple linear search at the moment.
Grigori Goronzy [Wed, 17 Aug 2011 16:59:00 +0000 (18:59 +0200)]
Never add a face twice to an ASS_Font
Introduce a unique ID per font face and check it in add_face to make
sure we never add a font face twice. This is useful in case the glyph
coverage report is unreliable.
Grigori Goronzy [Sun, 7 Aug 2011 00:21:09 +0000 (02:21 +0200)]
Custom font matching and font sources
Implement a simple font sorter (FontSelector) and an interface to deal
with multiple font sources (FontProvider). Unfinished business,
but works for the most part. Currently the only implemented FontProvider
uses fontconfig.
Dr.Smile [Thu, 25 Jun 2015 23:14:21 +0000 (02:14 +0300)]
Switch to virtual function table
Use one pointer to table of functions instead of scattered
bunch of function pointers. Different versions of these tables
can be constructed in compile time.
Also, bitmap memory alignment now depends only on SSE2/AVX2 support
and is constant for every width. That simplifies code without
noticeable performance penalty.
Oleg Oshmyan [Sun, 24 May 2015 22:00:38 +0000 (01:00 +0300)]
string2timecode: don't truncate to int
The timecode is a long long, but it is computed as a product whose
all multiplicands are (unsigned) ints and so effectively has the value
of an (unsigned) int. Fix this, and use the full long long range,
by explicitly making one of the first two multiplicands a long long.
Grigori Goronzy [Sun, 7 Jun 2015 19:42:42 +0000 (21:42 +0200)]
Check possible NULL dereference
Reported by clang scan-build static analysis. This cannot happen with
the current code, but the check might make this more robust in case
anything changes.
Oleg Oshmyan [Fri, 22 May 2015 21:06:40 +0000 (00:06 +0300)]
Parse and animate all colors and alpha values like VSFilter
* Allow exactly one of these prefixes in header values:
0x, 0X, &h, &H. Note that "0x0xFFFFFF" is a correct value,
as the first 0x is consumed by the parser and the second by
the string-to-number conversion following strtol semantics.
* Allow arbitrary numbers of leading & and H
(and not h) in any order in override tag values.
* Reduce header values modulo 2**32 instead
of saturating them to LLONG_MIN/MAX.
* Saturate override tag values to INT32_MIN/MAX
rather than to LLONG_MIN/MAX.
* Don't fiddle with bytes in alpha override tag values.
(They can be outside of the 0..255 range.)
Also change the byte swapping code to be more sensible.
Fixes #80.
Fixes #145.
Fixes #178.
Also fixes our behavior in the case described in
https://code.google.com/p/xy-vsfilter/issues/detail?id=80.
wm4 [Mon, 16 Mar 2015 20:02:47 +0000 (21:02 +0100)]
Don't use margins for events that should not be overridden
libass already does not use the margins for events using \pos. This is
not quite complete: there are other tags that we consider as "not
dialogue", and which should not be overridden. These tags do not use
EVENT_POSITIONED, and thus use the margins, which can mess up rendering.
wm4 [Mon, 9 Mar 2015 21:19:38 +0000 (22:19 +0100)]
Ignore extra coordinates in drawings
If there's an odd number of values, ignore the extra value, instead
of shifting all values after the next command. While such drawings are
boken strictly speaking, VSFilter handles them gracefully, and what
libass did before this commit made no sense anyway.
Test case:
{\an7\pos(0.01,0.7)\c&H5A493D&\p1\iclip(11,m 857379 -744112 l 517759 -744112 517759 -942768 857379 -942768 m 851235 -422238 l 689105 -422238 689105 -620546 851235 -620546 m 679889 -421552 l 517411 -421552 517411 -620894 679889 -620894 -55984 m 860113 425634 l 519459 425634 519459 226640 860113 226640 m 868305 992080 l 465361 992080 465361 793936 868305 793936 m 841343 1317200 l 520145 1317200 520145 1118206 841343 1118206 m 411662 -245762 b 411662 -242373 408918 -239618 405518 -239618 l 342931 -239618 b 339532 -239618 336787 -242373 336787 -245762 l 336787 -308226 b 336787 -311626 339532 -314370 342931 -314370 l 405518 -314370 b 408918 -314370 411662 -311626 411662 -308226 m 414929 -907440 b 414929 -904051 412174 -901296 408785 -901296 l 346198 -901296 b 342798 -901296 340054 -904051 340054 -907440 l 340054 -969904 b 340054 -973294 342798 -976048 346198 -976048 l 408785 -976048 b 412174 -976048 414929 -973294 414929 -969904 m 414417 -585904 b 414417 -582515 411652 -579760 408252 -579760 l 345450 -579760 b 342041 -579760 339286 -582515 339286 -585904 l 339286 -648368 b 339286 -651758 342041 -654512 345450 -654512 l 408252 -654512 b 411652 -654512 414417 -651758 414417 -648368 m 414929 -44997 b 414929 -41597 412174 -38832 408785 -38832 l 346198 -38832 b 342798 -38832 340054 -41597 340054 -44997 l 340054 -107676 b 340054 -111075 342798 -113840 346198 -113840 l 408785 -113840 b 412174 -113840 414929 -111075 414929 -107676 m 415932 261456 b 415932 264845 413178 267600 409788 267600 l 347191 267600 b 343802 267600 341047 264845 341047 261456 l 341047 198992 b 341047 195602 343802 192848 347191 192848 l 409788 192848 b 413178 192848 415932 195602 415932 198992 m 414929 564048 b 414929 567437 412174 570192 408785 570192 l 346198 570192 b 342798 570192 340054 567437 340054 564048 l 340054 501584 b 340054 498194 342798 495440 346198 495440 l 408785 495440 b 412174 495440 414929 498194 414929 501584 m 414929 781136 b 414929 784525 412174 787280 408785 787280 l 346198 787280 b 342798 787280 340054 784525 340054 781136 l 340054 718672 b 340054 715282 342798 712528 346198 712528 l 408785 712528 b 412174 712528 414929 715282 414929 718672 m 414929 1133525 b 414929 1136914 412174 1139669 408785 1139669 l 346198 1139669 b 342798 1139669 340054 1136914 340054 1133525 l 340054 1071061 b 340054 1067661 342798 1064917 346198 1064917 l 408785 1064917 b 412174 1064917 414929 1067661 414929 1071061)}m 0 0 l 1280 0 1280 720 0 720 0 0{\p0}
wm4 [Fri, 6 Mar 2015 12:28:48 +0000 (13:28 +0100)]
Make timestamp parsing more lenient
Fixes "[ARR] Musashi no Ken - 01 [AVC][5697986B].ssa", which has
timestamps like "00:02:30.85". It starts with 2 zeros instead of one,
which probably means it's an invalid file, but it's accepted by various
other parsers (VSFilter, MPlayer, vlc, ffmpeg).
wm4 [Fri, 6 Mar 2015 12:14:39 +0000 (13:14 +0100)]
Refine list of tags that prevent selective style overrides
Somewhat stolen from:
https://github.com/Cyberbeing/xy-VSFilter/blob/xy_sub_filter_rc3/src/subtitles/RTS.cpp#L2004
(xy-VSFilter started work on this in commit 014da6d9766417d7886eb867c9f2c14038f2a226)
wm4 [Thu, 26 Feb 2015 19:49:18 +0000 (20:49 +0100)]
Allow more fine grained control over style overrides
Add tons of ASS_OVERRIDE_ flags, which control whether certain ASS_Style
fields are copied when doing selective style overrides with
ass_set_selective_style_override_enabled().
This comes with some cleanup. It should be fully backwards-compatible.
wm4 [Thu, 26 Feb 2015 19:49:10 +0000 (20:49 +0100)]
Always use the new style with \r and style overrides
I don't know why it was done this way, but some fields were always taken
from the event's style, instead of the (possibly different) \r style.
Undo this, as it seems like an unnecessary complication.
Oleg Oshmyan [Mon, 9 Feb 2015 22:26:59 +0000 (00:26 +0200)]
Use correct types in be_blur_c
Also fix a related sort-of-bug: a multiple of sizeof(uint16_t)
was being added to a pointer that already pointed to uint16_t.
This was not causing any harm given enough space in the buffer.
Fixing the above also lets us combine the two memsets.
Oleg Oshmyan [Thu, 13 Mar 2014 03:15:18 +0000 (03:15 +0000)]
Calculate \be using [0..64] value range (like VSFilter)
To avoid banding in the output, the full [0..255] value range
is restored before the last \be pass, which then uses the full
range and hides the bands by virtue of being a blur.
With this, our \be finally closely matches VSFilter's.
The only visible difference (other than the lack of banding) is
in clipping: we add proper padding and output the whole blurred
image, while VSFilter does not add any padding and hence clips
the blurred image too early.
Dr.Smile [Fri, 23 Jan 2015 03:32:16 +0000 (06:32 +0300)]
Clean up bitmap combining
Now pre- and post-combining operations are clearly separated,
many parameters in cache keys are no longer necessary due to that.
Also an ambiguous (in case of multiple fonts) text string is replaced
with a list of direct bitmap references in composite cache key.
Oleg Oshmyan [Fri, 16 Jan 2015 11:26:13 +0000 (13:26 +0200)]
Add extern "C" guard to public header
Our public headers can be #included from C++. Hence, ass.h needs
an extern "C" guard to declare the correct linkage for functions.
ass_types.h only defines types, so it does not need a guard.