The various ass_* functions were created when libass was part of the
MPlayer tree and the distinction between MPlayer-specific and other
functions was less clear. Now that libass is a clearly separate
library, using the same ass_* namespace for player functions is ugly.
Rename the functions to use mp_ass_ prefix instead.
Rendering of ASS subtitles tries to be bug compatible with VSFilter
and stretches fonts when the video is anamorphic (some scripts try to
compensate for this VSFilter behavior, so trying to render them
"correctly" would give the wrong result). However this behavior is not
appropriate for subtitles we converted to ASS format ourselves for
libass rendering, as they certainly don't have VSFilter bug
workarounds. Change the code to use different behavior for "native"
ASS tracks and converted ones. It's questionable whether the
VSFilter-compatible behavior is appropriate for external .ass files
either, as there could be anamorphic and non-anamorphic versions of
the same video and the bug-compatible behavior can only be correct for
one alternative at most. However it's probably better to keep it as a
default at least, so that extracting a muxed subtitle track and using
that does not give behavior different from the original muxed one.
The aspect ratio setting is per ASS_Renderer, and changing it resets
libass caches. For that reason this commit adds separate renderer
instances to use for the "correct" and "VSFilter bug compatible"
cases.
SubRip subtitles have no "official" spec for any styling support, but
various tags are in common use; previous code filtered out text
between <> to remove HTML-style tags. Add support for those tags and
for MicroDVD subtitle styling. The style display is implemented by
converting the subtitles to the ASS subtitle format and displaying
them with libass, so libass needs to be enabled.
Original patch by Clément Bœsch <ubitux@gmail.com>.
mp_property_do() takes the value to set a property to through a
pointer. The calling code used '&mpctx->global_sub_pos' as the
pointer; however that variable could be changed during the
mp_property_do() call. Use a pointer to a copy of the original value
instead.
I think this only caused problems if you switched subtitle tracks from
a real one to "disabled" and then switched to a timeline part from
another source.
Add a framework for subtitle decoder modules that work more like
audio/video decoders do, and change libass rendering of demuxed
subtitles to use the new framework.
The old subtitle code is messy, with details specific to handling
particular subtitle types spread over high-level code. This should
make it easier to clean things up and fix some bugs/limitations.
Add a special option "-leak-report" that enables talloc leak
reporting. It only works if it's given as the first argument.
The code abuses the CONF_TYPE_PRINT option type to make main option
parsing ignore the option. The parser incorrectly consumed the
following commandline argument as a "parameter" for options of this
type when they had the flag to not exit after printing the message.
Fix this. It makes no difference for any previously existing option I
think.
The avsub implementation tries to fall back to MPlayer's other text
subtitle decoding if libavcodec returns text as the 'decoded'
subtitle. The code implementing this is buggy, and as far as I can see
it should not be triggered normally (libavcodec decoding is only
used for xvid, pgs and dvb subtitles, and for those libavcodec should
return bitmaps). Remove the buggy code (don't try to support
non-bitmap results) and simplify things a bit.
The contents of mpcommon.c were quite arbitrary; the most common
reason to place some functions in this file had been "MEncoder happens
to need similar code as MPlayer and we want to share some parts, but
we have no clue whatsoever how to organize things in a sensible way,
so we'll just dump those parts we want to share in mpcommon.c". As a
result of containing an essentially random subset of top-level player
functionality the mpcommon.h header required access to central structs
and was unsuitable for inclusion in lower-level code, but was
nonetheless included there for the mplayer_version symbol.
Move almost all contents from mpcommon.c to mplayer.c. mplayer.c is
already big and should perhaps be split further, but keeping a few
random functions in mpcommon.c would not be an improvement.
PulseAudio could keep reporting high delay values after a reset of
playing audio. This broke playback after seeking in some cases. Add a
workaround that should make things more robust against such
misbehavior.
Trying to do a framestep while playing an audio-only file would play
the file until the end, then start the next file in paused state. Make
framestep state enter pause again immediately if there is no video.
Also reset framestep state when switching files.
* hr-seek:
input: add default keybindings Shift+[arrow] for small exact seeks
input: support bindings with modifier keys for X input
core: audio: make ogg missing audio timing workaround more complex
core: add support for precise non-keyframe-limited seeks
core: add struct for queued seek info
commands: add generic option -> property wrapper
options: add "choice" option type, use for -pts-association-mode
core: remove looping in update_video(), modify command handling a bit
core: seek: use accurate seek mode with audio-only files
core: avoid using sh_video->pts as "current pts"
libvo: register X11 connection fd in input event system
core: timing: add special handling of long frame intervals
core: move central play loop to a separate function
Conflicts:
DOCS/tech/slave.txt
After the addition of exact seeking the code to work around missing
audio timestamps with ogg/ogm needs improvement. Now it's normal to
need adjustment at stream start time 0 (seeking to a position after
start of video but before second keyframe) with any video format, and
for exact seeks with ogg it's now more important not to skip the
sync. Make the check to detect the problem case more precise to avoid
affecting most other formats, and try to decode a second of audio
(hoping to get timestamps for those packets) before giving up.
Add support for seeking to an arbitrary non-keyframe position by
decoding video starting from the previous keyframe. Whether to use
this functionality when seeking is controlled by the new option
-hr-seek and a new third argument to the "seek" command. The default
is to use it for absolute seeks (like chapter seeks) but not for
relative ones. Because there's currently no support for cutting
encoded audio some desync is expected if encoded audio passthrough is
used. Currently precise seeks always go to the first frame with
timestamp equal to or greater than the target position; there's no
support for "matching or earlier" backwards seeks at frame level.
To prepare for the addition of exact seek support, add a struct for
queued seek state and a helper function to update its state. It would
have been cumbersome to update additional state (showing whether the
seek is forced to be exact or non-exact) manually at every point that
handles seeks.
Code in get_metadata() allocated too small a buffer for the text it
wrote (noticed by Clément Bœsch). Make the code cleaner and more
robust by changing it to use talloc_asprintf(). Also make it always
return non-NULL and remove checks on caller side.
Let higher-level code call update_video() again instead of looping
inside it until there's a frame ready to show. Change the conditions
for running user commands somewhat. Overall effect shouldn't be that
big. Now other commands can be executed after a seek before a video
frame is decoded; in this case the seek target time may be used as the
"current position".
build_afilter_chain is not safe to use directly, thus make it
static and instead use reinit_audio_chain which should have
better error handling.
Fixes a crash with -af hrtf and changing speed, audio will
still stop playing though.
git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@32648 b3059339-0415-0410-9bf9-f77b7e298cf2
git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@32630 b3059339-0415-0410-9bf9-f77b7e298cf2
Fix crash on path without directories.
Regression introduced in r32630. Patch by Yuriy Kaminskiy yumkam at mail ru.
git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@32631 b3059339-0415-0410-9bf9-f77b7e298cf2
Handle correctly paths with mixed '/' and '\' in it.
Patch by Yuriy Kaminskiy (yumkam at mail ru)
git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@32632 b3059339-0415-0410-9bf9-f77b7e298cf2
Handle ':' on systems with DOS paths: it allows paths like C:foo.avi.
git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@32642 b3059339-0415-0410-9bf9-f77b7e298cf2
Add a new field "video_pts" to mpctx. It records the time of the last
frame flipped visible on VO. Change various code which used
sh_video->pts to use either the new field or get_current_time(mpctx).
Add separate handling for the case where the time to flip the next
frame on the VO is more than 50 ms away. In that case don't update OSD
contents yet, but wait for possible changes until 50 ms before the
frame. Sleep until that time in mp_input_get_cmd(), so the sleep is
done in select() and input events can be responded to immediately.
Also raise the limit on audio out delay used to limit sleep.
Some Matroska files have inaccurate ordered chapter endpoints, and so
parts where one chapter should end and the next begin at the same
timestamp were not merged. This resulted in an unnecessary seek over a
minimal distance. Add a heuristic to merge parts with a minimal gap or
overlap between them.
Based on patch by Hector Martin <hector@marcansoft.com>.
ogg/ogm demuxers can give first audio packets without timestamp after
a seek. Due to some backwards compatibility code this results in the
sync code getting audio timestamp 0. In this case a lot of audio was
dropped unnecessarily when seeking to a position later in the file, as
the code saw audio starting from 0, video from something larger.
Make the code more robust in two ways. First, add a special case to
not try syncing if we get audio timestamp <= 0 (hopefully there aren't
many files where we'd really get audio starting from 0 and video from
a later timestamp). Second, when throwing audio away, make the code
recalculate from scratch the amount of bytes that still need to be
thrown away after every decode call. This limits the amount of damage
initial too-small timestamps can do, as the code will see the better
timestamps after a while.
Playing AVI files containing B-frames with demux_lavf printed two
"decreasing pts" info messages at the start of the file. We know the
timestamps from AVI won't be valid pts, so add a demuxer field to
convey that information to the timing code and make that not even try
to use the timestamps as valid pts.
Add code to enforce matching pts with video when (re)starting the
audio stream, by either cutting away the first samples or inserting
silence at the beginning. New option -noinitial-audio-sync can be used
to disable this and return to old behavior.
demuxer_get_current_chapter() accessed sh_video/sh_audio pts fields to
determine playback position. demux layer shouldn't access those and
the values used weren't quite correct anyway. Give the playback
position as a parameter to the demux layer function instead. Also
change the top-level get_current_chapter() to use get_current_time()
in the timeline case where it didn't refer to demux layer.
If the option is enabled and all audio has been buffered to the AO,
then the player will move to the next file without waiting for the
buffered audio to drain, while leaving the AO initialized. If the
playback of the next file starts quickly enough (before the AO buffer
empties) then it should continue writing audio to the same AO with no
gap in between.
At least with PCM it's possible to get an audio stream that doesn't
end at a multiple of whole sample per channel. At least ao_alsa
refuses to accept that part of input, and so EOF detection in
fill_audio_out_buffers didn't trigger until the 0.04 second sanity
check (as there "was still audio not sent to AO left"). Change the
logic to detect EOF if there's less than one sample per channel of
unsent data left.
When file format detection failed the output only said
"Exiting... (End of file)" after "Playing <file>." (or possibly error
messages triggered by format-specific check functions in between). Add
an explicit "Failed to recognize file format." error message.
When -alang / -slang was specified the numerically first matching
track (if any) was always chosen. This meant that specifying "-alang
eng" could change the track choice even if all tracks were in English,
because now the default flag of tracks was ignored. Change the logic
to take the default flag into account as a secondary sorting key.
The code also accepted prefix matches, so that "-slang g" would match
track language "ger". I think that was not intentional. Change it to
require exact matches.
Various code referred to "mpctx->demuxer" where it should really have
referred to the one used for audio/subtitles in case those differ. Fix
by using "mpctx->d_audio->demuxer" etc instead. Disable the copying of
streams in demux_demuxers; that was a partial workaround for things
referring to the main demuxer (and it wasn't enough anyway).
This fixes, among other things, switching audio tracks within the file
specified by -audiofile.
Move functions to query current playback position, percentage position
and total video length from from the demuxer layer to top level. The
functions need access to playback state that doesn't belong on the
demuxing level. Make the new functions more capable and simplify some
code that can now rely on them. This fixes some errors in displayed in
OSD and slave mode information when using timeline (ordered chapters).
