| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
| |
ass_font_clear() can now free family string even in a failure case.
Fixes https://github.com/libass/libass/issues/414.
|
|
|
|
|
|
| |
While FreeType uses FT_Pos for 32-bit values,
the real underlying type is signed long,
so labs() is more appropriate here.
|
| |
|
|
|
|
|
| |
This commit removes prefilters altogether at the cost of
enlarged main filter kernel.
|
|
|
|
|
|
|
| |
This commit enforces strict invariant on ASS_Outline
to contain point coordinates into predetermined range.
Fixes https://github.com/libass/libass/issues/431.
|
|
|
|
|
|
|
|
| |
* Bidirectional isolates
* Arabic letter mark
* Invisible plus
* Variation selectors
* Mongolian control characters
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Fixes https://github.com/libass/libass/issues/418.
Fixes https://github.com/libass/libass/issues/427.
Fixes https://github.com/mpv-player/mpv/issues/7712.
A longstanding problem we had was that we ignored leading line breaks
in events. We somewhat accidentally assigned line breaks zero height,
and the code in `measure_text` that halved the height of empty lines
explicitly ignored leading empty lines for some reason. #285 fixed both
of these things: the former in e8d98dafb8d4d1fd80e9d398cfbef7db1e2ccb73,
the latter in 5d03af99c6d8f43be973cb6dacb5d6dd0ada33b1.
But life is not so simple!
It turns out that VSFilter [discards metrics of the line break and
leading/trailing spaces for lines that stay nonempty after whitespace
trimming][1]. So we actually handled nonblank (trimmable-space-less)
lines correctly *before* #285 and have been *mishandling* them
since #285 landed, adding the line break's metrics when they should
be ignored. This is not noticeable in plain text, but it affects cases
when a line's contents have a different height from the line breaks:
{\fs96}foo\N{\p1}m 0 0 l 0 1 1 1 1 0{\p0}\Nbar
{\fs96}foo\N{\fs1}bar{\fs96}\Nbaz
(the middle line should be 1px high, not 96px)
[1]: https://github.com/Cyberbeing/xy-VSFilter/blob/3.0.0.306/src/subtitles/RTS.cpp#L1401-L1422
More complicated cases with trimmable spaces have never been
handled correctly, but this is nevertheless a regression.
To fix this:
* move the `trim_whitespace` call before the `measure_text` call
(they seem independent, so this is easy),
* mark trimmed whitespace with a new dedicated flag in addition
to `.skip` so it can be distinguished from invisible glyphs,
* clear accumulated (line-leading-whitespace) metrics
when each line's first non-trimmed-whitespace glyph is found,
* skip trimmed-whitespace glyphs when the line is already proven
nonblank, because they are certainly line-trailing whitespace.
Note: line breaks themselves do have `.skip` and `.is_trimmed_whitespace`.
Note: our `.linebreak` is set on the glyph *after* the line break.
As a result, a nonblank line will include the metrics only of the glyphs
that (would) remain after trimming leading and trailing whitespace
(including the line break), while a blank line includes the metrics of all
glyphs from its first in-line glyph up to and including the terminating
line break (if any).
At the same time, line height is only halved for lines that are truly empty
and don't even contain any whitespace. (Before the halving, the line height
comes from the line break glyph.)
This matches VSFilter.
|
| |
|
|
|
|
|
| |
VSFilter compatibility: VSFilter does no special handling
of invisible control characters and sees the line as nonempty.
|
|
|
|
|
|
|
|
| |
This reverts commit 6c2120a7cb4a6fd648ef37ad8f0d961cbd60f500.
This turned out never to actually be necessary, wasn't actually consistent
with vsfilter, and could result in script authors relying on
accidentally-introduced deviations from vsfilter font matching.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Providing a negative acceleration to \t could result in undefined behavior due
to overflow in float->int conversion. This codifies the same behavior we
currently have on x86, which matches vsfilter's, without actually invoking UB.
Additionally, vsfilter color animations work subtly differently than ours did.
We used to lerp each individual color component's byte value, while vsfilter
performs the lerp on the component _in place within a larger int_. This didn't
result in major issues for most cases, but could probably result in subtle
rounding errors, and gave vastly different results for \t with negative
acceleration.
Negative \t acceleration is probably completely useless, but our behavior was
wacky in a different way from vsfilter's, and I'd rather have portable
wackiness than libass-specific wackiness.
It might still be possible to invoke UB in negative-acceleration \t using tags
other than colors; we should fix those cases as well.
|
|
|
|
|
| |
This provides higher precision in reported weights when using
the fontconfig font provider.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The assertion in commit 66cef6774386d558e1e39096db926d677dad6882
fires on the following ASS code:
\t(\t(foobar,)
(where "foobar" is any nonblank sequence that does not contain a backslash)
The outer \t is parsed as having a single argument "\t(foobar,"
including the comma. The inner \t is parsed as having a single
argument "foobar" *excluding* the comma. As a result, in the inner
parse_tags, args[cnt].end == end - 1 && nested, and the assert fires.
This is because arguments that contain backslashes are parsed
differently from arguments that do not. But if the argument to \t
contains no backslashes, why bother parsing it at all?
It clearly has no override tags and affects nothing.
Rather than try to make the assert more clever (and more convoluted),
this commit skips parsing the last \t argument if it has no backslash.
The assert is now valid. This probably does not significantly affect
parsing speed in either direction.
Fixes https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=25796.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Before commit 6835731c2fe4164a0c50bc91d12c43b2a2b4e799,
parse_tags used to recurse for each nested \t(). The depth
of this recursion was not limited, and each \t in \t(\t(\t(...
added another level. This could lead to stack overflow.
Since that commit, parse_tags still recurses, but at most once:
it is called with nested=false at the top level and recurses with
nested=true for the outermost \t() (except rare cases in which
even this one level of recursion is avoided). Parsing stops at
the closing ) both for the outermost \t() and for any inner \t()
nested inside it, so the inner recursive call cannot recurse further.
This was not immediately obvious when reading the code,
and therefore it was not obvious that stack overflow is avoided.
Make it so by adding an assertion.
|
|
|
|
| |
GCC, Clang and MSVC now issue a warning when deprecated API is used.
|
|
|
|
|
| |
Also makes a measurable difference on float-parse-heavy scripts,
and it's not like these had a reason to not be in the header.
|
|
|
|
|
|
|
|
|
|
| |
Yes, this actually makes a major difference on some tag-heavy scripts.
This lets the whole function get inlined, rather than making a call out
to the libc's strcmp implementation. The libc's version is likely
much faster on longer strings, but we're only ever comparing against
strings that are a few characters long, so that doesn't really matter.
It also avoids making an extra pass for the strlen.
|
|
|
|
|
|
|
| |
Some releases rely on this.
See corresponding VSFilter code:
https://github.com/Cyberbeing/xy-VSFilter/blob/cf8f5b27de77fe649341bfab0fdfd498e1ad2fa6/src/subtitles/RTS.cpp#L1270
https://github.com/Cyberbeing/xy-VSFilter/blob/cf8f5b27de77fe649341bfab0fdfd498e1ad2fa6/src/subtitles/RTS.cpp#L1291
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bracket matching is incompatible with VSFilter (even on modern
Windows), so disable it by default. But as it's generally
a good thing (and 100% more compliant with current Unicode),
keep it available as an ASS_Feature.
It can be toggled individually or enabled as part of the
catch-all ASS_FEATURE_INCOMPATIBLE_EXTENSIONS feature.
If libass is compiled against FriBidi older than 1.0,
bracket matching is impossible. Signal this at runtime
by failing to recognize the ASS_FEATURE_BIDI_BRACKETS
feature. This way, clients who want to use bracket matching
can set the feature without any compile-time checks for
FriBidi and can be freely linked against libass that is itself
compiled against any version of FriBidi; and yet they can
detect at runtime whether the feature is actually enabled.
Fixes https://github.com/libass/libass/issues/374.
|
| |
|
| |
|
|
|
|
|
|
| |
The only prealloc value actually used is 0, which is not useful
and invokes implementation-defined (and potentially obsolescent
as per C11 DR400) behavior.
|
|
|
|
|
|
| |
This more forgiving behaviour resembles VSFilter's more closely and will
fix #117 ; trailing whitespace should not be an issue.
Though as pointed out in #19 VSFilter is still far more lenient.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Fixes https://github.com/libass/libass/issues/141.
Fixes https://github.com/libass/libass/issues/300.
VSFilter's original intention with this seems to have been to transform the
event as a whole, including the shadows as an integral part, as opposed to
transforming the shadows separately: imagine the event being rasterized into
a single image including the shadows, and then that image being transformed.
Unfortunately, due to a caching bug, what actually ends up happening is that
the shadow is transformed the intended way, but the main body is then simply
shifted back by \shad from the transformed & rasterized shadow, instead of
being transformed & rasterized on its own. The result seems sensible if you
look at the shadow only but incomprehensible if you look at the main body.
Transforms with \shad are actually used and this behavior is relied upon,
as evidenced by https://github.com/libass/libass/issues/300 (in which
the main body is made invisible and the shadow is used instead of it
due to having \t-animatable position and full-area blur despite \bord).
|
|
|
|
| |
This is *VSFilter compatible and fixes #287.
|
|
|
|
|
|
|
| |
ffmpeg/libav did rely on *VSFilter incompatible libass defaults for years.
Avoid breaking them when changing libass' default to match *VSFilter.
To this end detection of the generated by ffmpeg signature is added.
|
|
|
|
|
|
|
|
| |
Keeps backwards compatibility as libass is currently the only renderer accepting
custom format lines and is currently defaulting SBAS to 'yes' (VSF incompatible)
This commit also changes the default/fallback ASS Event format to use 'Name'
instead of 'Actor'.
|
|
|
|
| |
And warn about duplicate headers
|
|
|
|
|
|
|
|
| |
closes #304
With this commit the padding of the BorderStyle=4 box, given by \shad, is no
longer measured from the text without borders, but from the border of the text.
(Alternative description: padding changed from \shad to \shad+\bord)
|
| |
|
|
|
|
|
|
| |
closes #143
As libass doesn't support the 'Collsions' header, we are only concerned with the
default stacking direction of *VSF here
|
| |
|
|
|
|
|
|
|
| |
- Convert tabs to spaces
- Ensure one space between keywords and parenthesis
- Ensure space between ')' and '{'
- Trim trailing whitespace
|
| |
|
| |
|
|
|
|
|
| |
We used to do a lot of work to get freetype to imitate VSFilter's font-metrics
handling. Instead, we now just duplicate its behavior directly early on.
|
|
|
|
| |
device_x is in anamorphic coordinates, the product of x2scr (not x2scr_scaled).
|
|
|
|
|
|
| |
The ratio was accidentally flipped.
Use the actual video size, not the screen size that includes margins.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Based on a commit by wm4.
Nowadays, margins are used by players such as mpv to implement
video zoom & pan, although this was not expected when margins
were first implemented in libass. This results in unpleasant
rendering when panning too far, and it is argued that subtitles
should not change size or move when panning and zooming at all.
libass also makes an attempt to keep subtitles on screen even when
the use of margins is disabled. This is unintuitive and prone to break.
Fix this by strictly separating events which render as if they were
part of the video, and events which should use margins. The latter
will now use the entire screen as canvas, rather than using the video
frame. This actually simplifies the various y2scr functions.
To preserve scaling (mainly for styled subtitles where line breaks are
carefully chosen based on font/video size ratio) and to avoid badly
stretching out things like ASS Margins due to aspect ratio differences
between video and screen, estimate the unpanned & unzoomed video size
from the video aspect ratio and the screen size, and base all scaling
on that. This means that if the user plays a video in letterboxed mode
without extra margins, they get the same scaling as if they were playing
the same video with the same video rectangle size without any margins
at all (with some elements merely spaced out to make use of the black
bars); and when they zoom & pan afterwards, the subtitles don't move
or change size.
This changes behavior even with ass_set_use_margins(_, 0). Before this,
normal dialogue was forced into the visible video area (if negative
margins were set); now it renders it as if it were part of the video.
This also changes the behavior of left and right margins even with
ass_set_use_margins(_, 1). Before this, normal dialogue was forced
into the visible video width (with both positive and negative margins);
now it renders across the entire width of the screen/window.
For 4:3 video letterboxed on 16:9 screen, this means text will cross
the edges of the video, which may look worse than before.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Normal subtitles in use_margins mode, which do not have \clip tags or
similar, were clipped in a nonsensical way. It was especially visible
when moving subtitles up with ass_set_line_position().
This happened because state.clip_* is initialized with a clipping
rectangle for the video area. Later it tries to translate the clipping
rect accordingly, but this does not make much sense if you account for
the margins.
Just reset the clipping rect to the screen in these cases. explicit=0 is
enough to know that the clipping rect was never explicitly set.
|
| |
|
|
|
|
| |
Closes #397
|
|
|
|
|
|
|
|
| |
The while() checked the pointer for nullness, so the analyzer assumed it could
potentially be null, and thus warned on the reference to it later.
Using a do/while instead means we're only checking for subsequent linked-list
entries, which was the intent here anyway, and avoids the warning.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
| |
Since C does not allow empty enums, there is an "example", which doesn't
do anything. I think we might be able to make this change the default
bidi direction or so. As if this commit, the flag set by it is also not
available outside of ass.c, which should be solved by moving parser_priv
to an internal header.
|
|
|
|
|
|
|
|
| |
This just gives a minor hint that all fields are frozen, but that the
size of the structs are not part of the ABI.
Most importantly, ASS_Style and ASS_Event are completely ABI-frozen,
because ASS_Track has the "styles" and "events" arrays.
|
|
|
|
|
|
|
|
|
| |
Check for overflows that could happen with alignment and the
multiplication. The INT_MAX / 4 is somewhat approximate and assumes that
degenerate alignment values won't happen.
This still assumes that a possibly overflowing end_w/end_h calculation
doesn't make the compiler's optimizer destroy the overflow checks.
|
| |
|
| |
|
|
|
|
|
| |
This makes it a bit clearer that the struct's contents won't be reused
across multiple iterations
|
|
|
|
| |
This fixes a double-free in be0d1613f79a95073d18d96a60e1394abf9316a2
|
|
|
|
|
| |
This previously gave the pre-transition value; VSFilter's behavior is to give
the post-transition value.
|
|
|
|
|
|
|
|
| |
Fixes a crash in case a font does not has kCTFontURLAttribute,
which is the case for example on macOS 10.15.1 for the
".AppleSymbolsFB" font.
Fix #358
|
|
|
|
|
| |
This makes results much more consistent with other platforms,
particularly around cases where fonts have multiple conflicting names.
|
| |
|
| |
|