One sub_bitmaps struct could contain either a libass ASS_Image list, or
a mplayer native list of sub-bitmaps. This caused code duplication in
vo_vdpau.c and bitmap_packer.c.
Avoid this by creating such a sub_bitmap array even with libass. This
basically copies the list and recreates it in mplayer's native format.
It gets rid of the code duplication, and will make implementing extended
subtitle and OSD rendering in other VOs easier.
Also do some cosmetic changes and other preparations for the following
commits.
This removes the alternative values like "off", "0", "false" etc., and
also the non-English versions of these.
This is done for general consistency. It's better to have a single way
of doing things when multiple ways don't add singificant value.
Also update some choices for consistency.
Now it actually aborts, even if the abort command is not the first
command.
Make a policy change: commands before the abort command are silently
thrown away. Previously, normal commands were run after the abort
command finished (so they were run out of order). I'm not sure which
way is the best, all things considered, but the new way is simpler.
Basically, the encoding code path wanted to set osdlevel=0 as default,
while normal playback needs osdlevel=1. For this purpose, osdlevel was
set to -1 (i.e. invalid) initially to detect whether the --osdlevel
option was explicitly set. When encoding was not configured
(CONFIG_ENCODING undefined), the osdlevel value was not set from
-1 to 1 properly, and the OSD remained invisible by default.
Fix this by getting rid of this logic. It shouldn't be needed, since
osdlevel=1 never shows any OSD messages without user interaction.
Should this ever change, we could still check whether encoding is in
progress, or add another option to allow OSD rendering during encoding.
Normally, video/audio/sub track selection is persistent across files
played in the same mplayer instance. This is wanted, because settings
should not be reset across files in general. However, if the track
layout of a file is completely different from the previous, this will
essentially select random tracks. In this case, keeping the track
selection is confusing and unwanted.
Reset the track selection to default if the track layout changes. The
track layout is determined by number of tracks, track order, default
flag, whether the track is an external subtitle, and track language.
If a track layout change is detected when playing a new file, the -sid,
-aid and -vid options are reset to "auto".
This behavior is enabled only if the user selects tracks manually (via
keybinds and the "switch_audio" slave properties etc.). If no user
interactions take place, options specified on the command line will
follow the old behavior.
Previously, Matroska source files other than the initially opened one
were always accessed without caching. Enable cache for extra files
too. A separate cache process/thread is started for each file, which
is less than optimal but probably better than no caching if the user
explicitly enabled cache. This commit only implements caching for
Matroska ordered chapters (not for EDL timeline).
To build the timeline we need to demux the files in the current
directory to look for segments with matching uuid. This first demux is
done with no cache since we don't need to read a lot of the stream. If
the file is recognized as one of the needed sources it's reopened with
cache enabled.
Also move the stream_cache_size global variable to the options struct.
Conflicts:
cfg-mplayer.h
mplayer.c
stream/stream.h
timeline/tl_matroska.c
Remove subtitle selection code setting osd->ass_track directly and
vf_ass/vf_vo code rendering the track directly with libass. Instead,
do track selection and rendering with dec_sub.c functions.
Before, mpctx->set_of_ass_tracks[] contained bare libass tracks
generated from external subtitle files. For use with dec_sub.c, it now
contains struct sh_sub instances with decoder already initialized.
This commit breaks the sub_step command ('g' and 'y' keys) for
libass-rendered subtitles. It could be fixed, but it's so useless -
especially as with the existing implementation there's no practical
way to get subtitle delay back to normal after using it - that I
didn't bother.
Conflicts:
command.c
mp_core.h
mplayer.c
To draw libass subtitles, the code used ASS_Renderer objects created
in vf_vo (VO rendering) or vf_ass. They were destroyed and recreated
together with the video filter chain. Change the code to use a single
persistent renderer instance stored in the main osd_state struct.
Because libass seems to misbehave if fonts are changed while a
renderer exists (even if ass_set_fonts() is called on the renderer
afterwards), the renderer is recreated after adding embedded fonts.
The known benefits are simpler code and avoiding delays when switching
between timeline parts from different files (libass fontconfig
initialization, needed when creating a new renderer, can take a long
time in some cases; switching between files rebuilds the video filter
chain, and this required recreating the renderers). On the other hand,
I'm not sure whether this could cause inefficient bitmap caching in
libass; explicitly resetting the renderer in some cases could be
beneficial. The new code does not keep the distinction of separate
renderers for vsfilter munged aspect vs normal; this means that
changing subtitle tracks can lose cache for the previous track.
The new code always sets some libass parameters on each rendering
call, which were previously only set if they had potentially changed.
This should be harmless as libass itself has checks to see if the
values differ from previous ones.
Conflicts:
command.c
libmpcodecs/vf_ass.c
libmpcodecs/vf_vo.c
mplayer.c
sub/ass_mp.c
Change libavcodec subtitle decoding code (used for some bitmap
subtitle types) to use the same decoding framework as sd_ass. The
functionality that was previously in av_sub.c and was directly called
from mplayer.c is now in sd_lavc.c.
Conflicts:
mplayer.c
sub/av_sub.h
sub/sd_lavc.c
Merged from mplayer2. The remaining use of is_av_sub() is replaced by
a check whether a subtitle decoder is active, which should give the
same results.
Add all subtitle tracks as reported by libdvdread at playback start.
Display language for subtitle and audio tracks. This commit restores
these features to the state when demux_mpg was default for DVD playback,
and makes them work with demux_lavf and the recent changes to subtitle
selection in the frontend.
demux_mpg, which was the default demuxer for DVD playback, reordered
the subtitle streams according to the "logical" subtitle track number,
which conforms to the track layout reported by libdvdread, and is what
stream_dvd expects for the STREAM_CTRL_GET_LANG call. demux_lavf, on
the other hand, adds the streams in the order it encounters them in
the MPEG stream. It seems this order is essentially random, and can't
be mapped easily to what stream_dvd expects.
Solve this by making demux_lavf hand out the MPEG stream IDs (using the
demuxer_id field). The MPEG IDs are mapped by mplayer.c by special
casing DVD playback (map_id_from/to_demuxer() functions). This mapping
is essentially the same what demux_mpg did. Making demux_lavf reorder
the streams is out of the question, because its stream handling is
already messy enough.
(Note that demux_lavf doesn't export stream IDs for other formats,
because most time libavformat demuxers do not set AVStream.id, and we
don't know which demuxers do. But we know that MPEG is safe.)
Another major complication is that subtitle tracks are added lazily, as
soon as the demuxer encounters the first subtitle packet for a given
subtitle stream. Add the streams in advance. If a yet non-existent
stream is selected, demux_lavf must be made to auto-select that subtitle
stream as soon as it is added. Otherwise, the first subtitle packet
would be lost. This is done by DEMUXER_CTRL_PRESELECT_SUBTITLE.
demux_mpg didn't need this: the frontend code could just set ds->id to
the desired stream number. But demux_lavf's stream IDs don't map
directly to the stream number as used by libdvdread, which is why this
hack is needed.
When switching editions, it would be nice to display an OSD message to
notify the user which edition is being played. This would be consistent
with visual feedback like on tracking switching, seeking, and so on.
Rather than introducing awkward hacks and special cases to determine
whether a file is being "restarted" (as it happens on edition switching)
to avoid clearing the OSD messages, simply never clear the OSD stack.
This is more consistent with handling of OSD bars too: they didn't
disappear when going to a new file (especially noticable when seeking
past the end of the file).
Add a new slave property which switches the current Matroska edition.
Since each edition can define an entirely new timeline, switching the
edition will simply restart playback at the beginning of the file with
the new edition selected.
Add 'E' as new keybinding to step the edition property.
DVD titles are still separate. Apparently they work similarly, but I
don't have any multi-title DVDs for testing. Also, cdda (for audio CDs)
uses the same mechanism as DVDs to report a number of titles, so there
seems to be confusion what exactly this mechanism is supposed to do.
That's why the edition code is completely separate for now.
Remove demuxer.num_titles. It was just a rather useless cache for the
return value of the DVD titles related STREAM_CTRL.
One rather obscure corner case isn't taken care of: if the ordered
chapters file has file local options set, they are reset on playback
restart. This is unexpected, because edition switching is meant to
behave like seeking back to the beginning of the file.
Usually failing to initialize the video chain does not stop playback,
and continues with audio-only. But when VF/VO config failed (the second
step in initializing them), playback was canceled and advanced to the
next file.
Make it more consistent and just switch off video if it fails.
Also, if no audio track is selected, go to next file anyway. The player
core goes into some sort of idle mode if no video and audio track is
selected, and we don't want to trigger it here. (It's unclear whether
that is a bug or feature: it's meant to work around some corner cases,
but maybe it shouldn't be triggered when switching off video & audio
tracks in normal playback.)
Introduce a general track struct for every audio/video/subtitle track
known to the frontend. External files (subtitles) are now represented
as tracks too. This mainly serves to clean up the subtitle selection
code: now every subtitle is simply a track, instead of using a messy
numbering that goes by subtitle type (as it was stored in the
global_sub_pos field). The mplayer fontend will list external subtitle
files as additional tracks.
The timeline code now tries to match the exact demuxer IDs of all
tracks. This may cause problems when Matroska files with different
track numberings are used with EDL timelines. Change demux_lavf not
to set demuxer IDs, since most time they are not set.
Whenever the demuxer is accessed to retrieve metadata, use the newly
introduced master_demuxer field instead. If a timeline is used, the
master_demuxer will still refer to the main file, instead to segments.
Instead of mpctx->demuxer->stream, access mpctx->stream; even though
they are the same, the code becomes shorter. For the TV code, introduce
a function get_tvh() to access the TV handle, instead of duplicating the
code all over the place. Often the demuxer field is checked to determine
whether something is currently played; prefer other methods over that.
Note that the code before always accessed the current timeline segment,
and would e.g. read file metadata of the current segment. Now it always
returns metadata for the master file. This may have various wanted and
unwanted effects.
The timeline code kept pairs of stream and demuxer references around.
The reference to the stream is redundant, because it can be accessed
through the demuxer.
Simplify the code by removing the redundant stream reference.
Also, set mpctx->stream to the current segment when using timeline.
Fix a small memory leak in tl_matroska.c introduced earlier.
The code for showing OSD progression ('P' key) set a flag in a
mp_osd_msg_t to do its work. OSD messages can be re-used for completely
unrelated purposes (it's unclear why), so this has to be reset if an
old OSD progression message is reused for something different.
Be sure to reset the full message. Remove the messy code for searching
the OSD stack. Use the existing rm_osd_msg() function to remove the old
message (if there was one), and always create a new message. The new
code should be functionally equivalent to the old code. The "started"
flag wasn't reset before, but since the time is always overwritten, this
might be actually more correct.
The osd_show_progression command ('P' key) created text similar to
what --osdlevel=3 shows, and set that as an OSD message. This message
was static and wasn't updated while visible, even if video position
changed (very noticeable during fast forward, when real OSD position
changes rapidly). Instead of setting a static message, create a new
message type that makes the OSD update code behave as if osd level was
set to 3 for the duration of the message.
The OSD progress bar isn't updated while the message is active.
Based on mplayer2 commit 458001463b2252fc by uau.
Add a flags parameter to mp_input_set_section(). Add a flag that defines
whether bindings in the default section are used or not. This is useful
for special functionality, where the normal key bindings may have
unwanted effects.
For example, it shouldn't be possible to seek during encoding. However,
you want to be able to cancel the encoding process gracefully. For that
purpose, the "encode" section of input.conf could be made exclusive:
mp_input_set_section(mpctx->input, "encode", MP_INPUT_NO_DEFAULT_SECTION);
And input.conf could contain this definition:
RIGHT seek 10
q {encode} quit
Then only the key "q" would be bound during encoding.
The functionality enabled with the --leak-report option is very useful
for debugging. Introduce a check for the MPLAYER_LEAK_REPORT environment
variable, and enable talloc leak report if it's set to "1".
The environment variable encourages enabling leak report permanently
during development. It's also a bit harder to get wrong: if the
--leak-report option is not the first option, it's silently ignored.
You can't put this option into a config file either. Enabling this
with --enable-debug in configure is not an option, because the leak
report code doesn't seem to be thread-safe, and thus is a bit dangerous.
Also, move the code to the very beginning to make sure leak report is
enabled before any other talloc function is used. Otherwise, these
allocations could be missed.
In commit 94782e464d, code was added to remove the first
command line argument. (Because that is essentially useless.) The code
for printing with command line on -v still assumed the first argument
should be skipped.
When the internal mplayer MPEG demuxer was removed (commit 1fde09db),
the default demuxer when using dvdnav was set to libavformat. Now it
turns out that this doesn't work with libavformat. It will terminate
playback right after the audio runs out (instead of looping it like the
video, or whatever it's supposed to do). I'm not sure what exactly the
problem is, but since 1. even mplayer-svn can't handle DVD menus
directly (missing highlights), 2. DVD menus are essentially worthless,
and 3. I don't directly watch DVDs, don't bother with it and remove it.
For basic playback, there's still libdvdread support.
Also, use pkg-config for libdvdread, and drop support for in-tree
libdvdread. Remove support for in-tree libdvdcss as well.
Remove the win32 loader - the win32 emulation layer, as well as the
code for using DirectShow/DMO/VFW codecs. Remove loading of xanim,
QuickTime, and RealMedia codecs.
The win32 emulation layer is based on a very old version of wine.
Apparently, wine code was copied and hacked until it was somehow able
to load a limited collection of binary codecs. It poked around in the
code segment of some known binary codecs to disable unsupported win32
API calls to make them work. Example from module.c:
for (i=0;i<5;i++) RVA(0x19e842)[i]=0x90; // make_new_region ?
for (i=0;i<28;i++) RVA(0x19e86d)[i]=0x90; // call__call_CreateCompatibleDC ?
for (i=0;i<5;i++) RVA(0x19e898)[i]=0x90; // jmp_to_call_loadbitmap ?
for (i=0;i<9;i++) RVA(0x19e8ac)[i]=0x90; // call__calls_OLE_shit ?
for (i=0;i<106;i++) RVA(0x261b10)[i]=0x90; // disable threads
Just to show how utterly insane this code is. You wouldn't want even
your worst enemy to have to maintain this. In fact, it seems nobody
made major changes to this code ever since it was committed.
Most formats can be decoded by libavcodecs these days, and the loader
couldn't be used on 64 bit platforms anyway. The same is (probably)
true for the other binary codecs.
General note about how support for win32 codecs could be added back:
It's not possible to replace the win32 loader code by using wine as
library, because modern wine can not be linked with native Linux
programs for certain reasons. It would be possible to to move DirectShow
video decoding into a separate process linked with wine, like the
CoreAVC-for-Linux patches do. There is also the mplayer-ww fork, which
uses the dshownative library to use DirectShow codecs on Windows.
The terminal status line (showing playback status etc.) was too long
for the standard 80 column width in some cases. Shorten the output by
abbreviating some fields with single letters.
Change "(PAUSED!)" to "(Paused)". This looks nicer.
Move the speed field forward and omit the explicit "header". It's
probably intuitively clear that "x2.00" means double speed.
The field showing the playback time in seconds was padded with spaces.
This just takes space away and wasn't really needed.
This was a hack for .mov reference files. The mov demuxer, which
triggered this code, has been removed in commit 1fde09db6f.
The code serves no purpose anymore, and it was bogus in the first
place. (This mov feature should have been handled either by the core's
timeline support, or as normal playlist.)
It's not clear why this video filter supported OSD rendering.
The manpage says:
"Can be used for placing subtitles/OSD in the resulting black bands."
But every single VO already does this if vf_expand adds black borders.
This feature is 100% pointless.
Commit 7484ae8e2e attempted to introduce two ass_library handles
(as it was needed to deal with how ass_library manages fonts), but the
commit was completely bogus: it assumed osd_state->ass_library would be
used by osd_libass.c only, which is not the case. As result, some of the
subtitle code used the wrong ass_library handle.
We need two ass_library handles in osd_state. The one from the mplayer
core for subtitles (osd_state->ass_library), and one for OSD rendering
(osd_state->osd_ass_library).
osd_libass.c used the same ASS_Library object as the player core. This
caused a problem: when playing a new file, all fonts loaded by the
ASS_Library object were unloaded, including the OSD font. Parts of the
OSD would stop being rendered correctly.
Solve this by creating a separate ASS_Library, with its own set of
fonts.
Commit 168293e0ae assumed the OSD drawing routines (which have the
functions osd_draw_text/_ext as entrypoint) would always be called, and
relied on that to reset the change flag.
Some VOs, such as vo_null, didn't do this. Pausing could turn into
endless framestepping in some cases. Restore the part of the OSD drawing
logic that dealt with this. (Alternatively, the VOs could be obliged to
always call the OSD drawing routines, even if the VO doesn't actually
draw the OSD. But it seems even more messy to rely on that.)
The commit 74df1d8e05 (and f752212c62) replaced the configure
endian check with byte order macros defined by standard headers. It
turns out that MinGW-w64 actually doesn't define these macros in the
sys/types.h system header. (I assumed it does, because a quick test
seemed to work. But that was because gcc -W -Wall doesn't warn against
undefined macros. You need -Wundef for that.) MinGW-w64 has a
sys/params.h header defining these macros, but sys/types.h doesn't
include it, so it's useless without special casing the mplayer code.
Add a hack top configure instead. Define the macros directly, and
assume MinGW-w64 only works on little endian machines.
The other changes are basically random typos and superficial oversights.
Options parsing used to be ambiguous, as in the splitting into option
and values pairs was ambiguous. Example:
-option -something
It wasn't clear whether -option actually takes an argument or not. The
string "-something" could either be a separate option, or an argument
to "-option". The code had to call the option specific parser function
to resolve this.
This made everything complicated and didn't even have a real use. There
was only one case where this was actually used: string lists
(m_option_type_string_list) and options based on it. That is because
this option type actually turns a single option into a proxy for several
real arguments, e.g. "vf*" can handle "-vf-add" and "-vf-clr". Options
suffixed with "-clr" are the only options of this group which take no
arguments.
This is ambiguous only with the "old syntax" (as shown above). The "new"
option syntax always puts option name and value into same argument.
(E.g. "--option=--something" or "--option" "--something".)
Simplify the code by making it statically known whether an option takes
a parameter or not with the flag M_OPT_TYPE_OLD_SYNTAX_NO_PARAM. If it's
set, the option parser assumes the option takes no argument.
The only real ambiguity left, string list options that end on "-clr",
are special cased in the parser.
Remove some duplication of the logic in the command line parser by
moving all argument splitting logic into split_opt(). (It's arguable
whether that can be considered code duplication, but now the code is a
bit simpler anyway. This might be subjective.)
Remove the "ambiguous" parameter from all option parsing related code.
Make m_config unaware of the pre-parsing concept.
Make most CONF_NOCFG options also CONF_GLOBAL (except those explicitly
usable as per-file options.)
This also requires that the OSD stack related functions carry a pointer
to MPContext.
Free the OSD stack items (mp_osd_msg) at exit by making MPContext the
talloc parent. (E.g. when exiting while something is still displayed on
the OSD.)
The only place exit_player() should be called is the main() function.
exit_player() should be the only function allowed to call exit(). This
makes it easier to guarantee proper deinitialization, and allows using
the --leak-report flag without showing false positives.
The quit slave command now sets a flag only. It uses the same mechanism
that's normally used to advance to the next file on the playlist, so the
rest of the playback path should be able to react to the quit command
quickly enough. That is, the player should react just as fast to quit
requests in practice as before this commit.
In reinit_audio_chain(), the player was actually exited if
init_audio_filters() failed. Reuse the normal error handling path to
handle this condition.
The print_timeline() function actually had contained some code that
changed the MPContext state. Since you wouldn't expect that from a
function named print, move that code out of the function. The
misleading code structure was introduced in commit 6f564fe82b.
Commit 89a17bcda6 simplified the idle loop to run any commands
mplayer receives, not just playlist related commands. Unfortunately, it
turns out many slave commands always assume the presence of a demuxer.
MPContext->demuxer is assumed not to be NULL. This made the player
crash when receiving slave commands like pause/unpause, chapter
control, subtitle selection.
We want mplayer being able to handle this. Any slave command or
property, as long as it's backed by a persistent setting, should be run
successfully, even if no file is being played. If the slave command
doesn't make sense in this state, it shouldn't crash the player.
Insert some NULL checks when accessing demuxers. If sh_video or
sh_audio are not NULL, assume demuxer can't be NULL.
(There actually aren't that many properties which need to be changed. If
it gets too complicated, we could employ alternative mechanisms instead,
such as explicitly marking safe properties with a flag.)
There are different C types for each stream type: sh_video for video,
sh_audio for audio, sh_sub for sub. There is no type that handles all
stream types in a generic way. Instead, there's a macro SH_COMMON, that
is used to define common fields for all 3 stream structs. Accessing
the common fields is hard if you want to be independent from the stream
type.
Introduce an actual generic stream struct (struct sh_stream), which is
supposed to unify all 3 stream types one day. Once all fields defined
by SH_COMMON have been moved into sh_stream, the transition is complete.
Move some fields into sh_stream, and rewrite osd_show_tracks to use
them.
The SH_COMMON lang field seems to be blatantly unreliable and is not
always set by demux_lavf (at least not with dvdnav:// ).
Also fix the same for the show_tracks_osd slave command.
The structure is now as follows:
- main():
* basic initializations (e.g. init_libav() and more)
* pre-parse command line (verbosity level, config file locations)
* load config files (parse_cfgfiles())
* parse command line, add files from the command line to playlist
(m_config_parse_mp_command_line())
* call:
- handle_help_options():
* check each help-related option
* print help if requested
* main() exits if help was requested
* call function that works down the playlist:
- play_files():
* run idle loop (idle_loop()), until there are files in the
playlist or an exit command was given (slave mode only)
* actually load and play a file:
- play_current_file():
* run all the dozens of functions to load the file
and initialize playback
* run a small loop that does normal playback, until
the file is done or a slave command terminates
playback
(each iteration, run_playloop() is called)
* uninitialize playback
* determine next entry on the playlist to play
* loop
* call exit_player_with_rc() (there are many other places which
use this function, though)
This is about the vo_x11_init_state() call. It basically opens a X11
connection. It's called in the main() function once. It's not really
clear why this isn't done on VO creation instead. Maybe one reason was
that --no-fixed-vo used to be the default: when playing a new file, the
full VO state would be free'd and recreated. Keeping the X11 connection
possibly improved things, although the question is how. In summary,
there is no good reason to do this, and it only adds platform specific
details to the player frontend.
Do the X11 initialization in the respective VOs instead.
wm4
Rudolf Polzer
Stefano Pigozzi
Uoti Urpala