Instead of using a callback to "capture" the image next time the filter
function is called, do it the other way around: on every filter
invocation, create a reference to the image, and return it if a
screenshot is requested. This also fixes the 1-frame delay when taking
screenshots with the filter.
This also allows simplifying screenshot.c.
VFCAP_TIMER disables any additional waiting done by mpv in the
playloop. Remove VFCAP_TIMER, but re-use the idea for vo_image and
vo_lavc.
This means --untimed doesn't have to be passed when using --vo=image.
Change the entire filter API to use reference counted images instead
of vf_get_image().
Remove filter "direct rendering". This was useful for vf_expand and (in
rare cases) vf_sub: DR allowed these filters to pass a cropped image to
the filters before them. Then, on filtering, the image was "uncropped",
so that black bars could be added around the image without copying. This
means that in some cases, vf_expand will be slower (-vf gradfun,expand
for example).
Note that another form of DR used for in-place filters has been replaced
by simpler logic. Instead of trying to do DR, filters can check if the
image is writeable (with mp_image_is_writeable()), and do true in-place
if that's the case. This affects filters like vf_gradfun and vf_sub.
Everything has to support strides now. If something doesn't, making a
copy of the image data is required.
Deprecate the hardware specific video codec entries (like ffh264vdpau).
Replace them with the --hwdec switch, which requests that a specific
hardware decoding API should be used. The codecs.conf entries will be
removed at a later time, but for now they are useful for testing and
compatibility.
Instead of --vc=ffh264vdpau, --hwdec=vdpau should be used.
Add a fallback if hardware decoding fails. Most hardware decoders
(including vdpau) support only a subset of h264, and having such a
fallback is supposed to enable a better user experience.
Simplify the decoder pixel format handling by making it handle only
the case vd_lavc needs: a video stream always decodes to a single
pixel format.
Remove the handling for multiple pixel formats, and remove the
codecs.conf pixel format declarations that are left.
Remove the handling of "ambiguous" pixel formats like YV12 vs. I420 (via
VDCTRL_QUERY_FORMAT etc.). This is only a problem if the video chain
supports I420, but not YV12, which doesn't seem to be the case anywhere,
and in fact would not have any advantage.
Make the "flip" flag a global per-codec flag, rather than a pixel format
specific flag. (Some ffmpeg decoders still return a flipped image, so
this has to be done manually.) Also fix handling of the flip operation:
do not overwrite the global flip option, and make the --flip option
invert the codec flip option rather than overriding it.
Slices allowed filtering or drawing video in horizontal bands or
blocks. This allowed working on the video in smaller units. In theory,
this could bring a performance win by lowering cache pressure, as you
didn't have to keep the whole video frame in cache while filtering,
only the slice.
In practice, the slice code path was barely used for the following
reasons:
- Multithreaded decoding with ffmpeg didn't use slices. The ffmpeg
slice callback was disabled, because it can be called from another
thread, and the mplayer video chain is not thread-safe.
- There was nothing that would turn "full" images into appropriate
slices, so slices were rarely used.
- Most filters didn't actually support slices.
On the other hand, supporting slices lead to code duplication and more
complex code in general. I made some experiments and didn't find any
actual measurable performance improvements when using slices. Even
ffmpeg removed slices based filtering from libavfilter in favor of
simpler code.
The most broken thing about the slices code path is that slices can't
be queued, like it is done for images in vo.c.
This function sucks and apparently is not very portable (at least on
mingw, the configure check fails). Also remove the emulation of that
function from osdep/strsep*, and remove the configure check.
Ensure that even if a seek is inaccurate it will not show video from
outside the defined timeline. Previously, seeking to the beginning of
a segment could show frames from before the start of the segment if
the seek was done in inaccurate mode and the demuxer seeked to an
earlier position. Now hr-seek machinery is used to skip at least the
frames that should not be part of playback timeline at all.
Now external subtitles essentially use the playback time, instead of
the segment time.
This is more useful when using external subtitles with mkv ordered
chapters. The previous behavior is not necessarily incorrect, and e.g.
makes it easier to use subtitles directly extracted from ordered
chapters segments. But we consider the new behavior more useful.
Also see commit 06e3dc8.
This is simpler and more useful. We could add a new switch for the old
functionality, but that would probably be more confusing than helpful.
When passing only a single file to the command line, this commit
shouldn't change behavior.
(Classic mplayer provided both features by duplicating the loop
functionality in the "playtree".)
When the last frame is displayed, and a frame step command is issued,
playback ands and advances to the next file. But before this commit,
the next file was played unpause. Fix this, and make sure pause is
kept.
Before this commit, the --osd-* options (like --osd-font-size etc.)
configured both the OSD and subtitle font. Make them separate, and add
--sub-text-* options (like --sub-text-size etc.). Now --osd-* affects
the OSD font only, and --sub-text-* unstyled text subtitles only.
This is better than having just the operating system type decide the
wakeup period, as e.g. when compiling for Win32/cygwin, a wakeup period
of 0.5 would work perfectly fine.
Instead, the default wakeup period is now only decided by availability
of a working select() system call (which is the case on cygwin but not
mingw and MSVC) AND a vo that can provide an event file descriptor or a
similar hack (vo_corevideo). vos that cannot do either need polling for
event handling and now can set the wakeup period to 0.02 in the vo code.
Add `mp_find_config_file` to search different known paths and use that in
ass_mp to look for the fontconfig configuration file.
Some incidental changes spawned by this feature where:
* Buffer allocation for the strings containing the paths is now performed
with talloc. All of the allocations are done on a NULL context, but it still
improves readability of the code.
* Move the OSX function for lookup inside of a bundle: this code path was
currently not used by the bundle generated with `make osxbundle`. The plan
is to use it again in a future commit to get a fontconfig config file.
Even if this is not so bad as other files, I need to add some stuff so...
why not!?
`uncrustify -l C -c TOOLS/uncrustify.cfg --no-backup --replace core/path.h`
`uncrustify -l C -c TOOLS/uncrustify.cfg --no-backup --replace core/path.c`
The header was unchanged by the tool.
Keep the currently displayed subtitles even when the user cycles through
subtitle tracks, and the subtitle is decoded by libavcodec (such as
vobsubs). Do this by not clearing the subtitles on reset(). reset() is
also called on seek, so check the start PTS whether the subtitle should
really be displayed (there's already an end PTS). Note that sd_ass does
essentially something similar.
The existing code has checks for whether the PTS reported by the demuxer
is invalid (MP_NOPTS_VALUE). I don't know under what circumstances this
can happens, so fall back to the old behavior if the PTS is invalid.
This slightly improves display of the current playback time in files
with sparse video packets (like video tracks containing a slow MJPG
slideshows as in [1]), or audio files with cover art image attachments.
While the video PTS is always "stuck" at the last frame displayed or
the last seek, audio is usually continuous. Given sane samplerates and
working audio drivers (to query how much of the current audio buffer has
been played), the audio PTS should always be more reliable.
[1] http://www.podtrac.com/pts/redirect.mp3/traffic.libsyn.com/rtpodcast/Rooster_Teeth_Podcast_191.m4a
ffmpeg pretends that image attachments (such as contained in ID3v2
metadata) are video streams. It injects the attached pictures as packets
into the packet stream received with av_read_frame().
Add the --audio-display option to allow configuring whether attached
pictures should be displayed. The default behavior doesn't change
(images are displayed).
Identify video streams, that are actually image attachments, with "[P]"
in the terminal output.
Modify the default stream selection such that real video streams are
preferred over attached pictures. (This is just for robustness; I do not
know of any samples where images are added before actual video streams
and could lead to bad default stream selection with the old code.)
This caused errors like:
core/mplayer.c:4308:5: error: implicit declaration of function 'pthread_win32_thread_detach_np' [-Werror=implicit-function-declaration]
It turns out a pthread.h include was missing. It's not clear why this
used to work (or rather, why it happens only sometimes). Possibly some
libraries or system headers recursively include pthread.h under certain
circumstances or configurations.
Fix missing quoting in configure, which led to broken terminal output.
Closes#6.
libavformat wants to read a full ~400KB of data to determine whether
it's really AAC. This causes slow startup with AAC web radio streams [1]
(possible due to a broken initial packet). There are similar issues
with other file formats.
Make the probe "score" (libavformat's mechanism for testing file
formats) configurable with the -lavfdtops:probescore option. This allows
lowering the amount of data read on probing. If the probe score is below
the probescore option value, demux_lavf will try to get a higher score
by feeding more data to libavformat, until the required score or the
max. probe size is reached.
Remove the lavf_preferred demuxer entry. This had a purpose in
mplayer-svn, but now there doesn't seem to be any good reason for it
to exist. Make sure that our native "good" demuxers are above
demux_lavf in demuxer_list[] instead (so that they are preferred).
[1] http://lr2mp0.latvijasradio.lv:8000
ffmpeg recently added a demuxer that can read vobsubs (pairs of .sub and
.idx files). Get rid of the internal vobsub reader, and use the ffmpeg
demuxer instead.
Sneak in an unrelated manpage change (autosub default).
This affects streams loaded with -subfile and -audiofile. They could get
out of sync when they were deselected, and the main file was seeked. Add
code to seek external files when they are selected (see
init_demux_stream()).
Use avformat_seek_file() under certain circumstances. Both av_seek_frame()
("old" API) and avformat_seek_file() ("new" API) seem to be broken with
some formats. At least the vobsub demuxer doesn't implement the old API
(and the old API doesn't fallback to the new API), while the fallback
from new API to old API gives bad results. For example, seeking forward
with small step sizes seems to fail with the new API (tested with
Matroska by trying to seek 1 second forward relative to priv->last_pts).
Since only subtitle demuxers implement the new API anyway, checking
whether iformat->read_seek2 is set to test whether the old API is not
supported gives best results. This is a hack at best, but makes things
work.
Remove backwards seeking on seek failure. This was annoying, and only
was there to compensate for obscure corner cases (see 1ad332). In
particular, files with completely broken seeking that used to skip back
to the start on every seek request may now terminate playback.
Do this only if demux_lavf is used. Using demux_mpg and the ffmpeg DVD
subtitle decoder doesn't work. The problem is probably that demux_mpg
doesn't join split sub packets, while demux_lavf does. The internal
DVD sub decoder (spudec.c) can, while ffmpeg's dvdsub can't. I do not
know whether this is the actual problem.
If DVD playback is used, create "fake" vobsub-style text extradata
(like .idx files) to pass resolution and palette information to the
ffmpeg decoder. We could use the "palette" AVOpt and avcodec_set_dimensions()
instead, but it's actually simpler this way. Note that the decoder
doesn't parse any other fields. Also note that DVD playback still uses
demux_mpg by default, so this code is inactive unless -demuxer lavf is
specified. This is mainly preparation for the case when we manage to get
rid of demux_mpg for DVD playback.
The option sub-forced-only was accidentally renamed back to
forcedsubsonly in commit 72205635ab, causing a segfault in
mp_property_generic_option due to missing option.
Button 3 and 4 are mapped to the mouse wheel. Double click events for
them annoying and not useful at all.
I don't know about buttons 5-19; for all I know these could be mapped
to wheels as well. Even if not, double click events are probably not
very important for these. Disable double clicks for these as well.
Remove the options --utf8 and --unicode which had no effect any more
(what they once did should be doable with --subcp). The only use of
corresponding variables left in code was subreader.c code using
sub_utf8 as a flag indicating whether iconv conversion was active.
Change the code to test the existence of iconv context instead.
Conflicts:
DOCS/man/en/options.rst
core/cfg-mplayer.h
sub/sub.c
sub/sub.h
sub/subreader.c
Merged from mplayer2 commit ea7311.
Note: --unicode was already removed
When the cache fill status goes below a certain threshold, automatically
pause the player. When the cache is filled again, unpause again.
This is intended to help with streaming from http. It's better to pause
a while, rather than exposing extremely crappy behavior when packet
reads during decoding block the entire player.
In theory, we should try to increase the cache if underruns happen too
often. Unfortunately, changing the cache implementation would be very
hard, because it's insane code (forks, uses shared memory and "volatile"
etc.). So for now, this just reduces the frequency of the stuttering if
the network is absolutely too slow to play the stream in realtime.
This commit is separate from the previous one to separate our own
changes from changes merged from mplayer2 (as far as that was possible).
Make it easier for stream implementations to request being cached. Set
a default cache size in stream.c, and remove them from various stream
implementations. Only MS streaming support sets a meaningful cache size.
Make querying cache size saner. This reduces the amount of #ifdefs
needed.
Code enabling the cache by default for network streams did that by
modifying the value of the "cache" option. This wasn't sane, as
multiple streams may be created and all share the same options. Change
the code to not modify options but store data in the stream instance
instead.
Conflicts:
core/mplayer.c
demux/demux.c
stream/cache2.c
stream/network.c
stream/network.h
stream/pnm.c
stream/stream.c
stream/stream_rtp.c
Merged from mplayer2 commit e26070. Note that this doesn't solve any
actual bug, as the playlist crashing bug has been fixed before.
Since the global cache size option value is not overwritten anymore, the
option doesn't need to be restored on end of playback (M_OPT_LOCAL).
Based on a patch by qyot27. Add export LC_ALL=C on top of version.sh to
make the output locale independent.
Note that the build time will not be updated on every "make" invocation,
but only when the git revision is updated. This is a good thing, as
repeated make invocations should not rebuild the binary. (This would
break "sudo make install" too.)
libavdevice supports various "special" video and audio inputs, such
as screen-capture or libavfilter filter graphs.
libavdevice inputs are implemented as demuxers. They don't use the
custom stream callbacks (in AVFormatContext.pb). Instead, input
parameters are passed as filename. This means the mpv stream layer has
to be disabled. Do this by adding the pseudo stream handler avdevice://,
whose only purpose is passing the filename to demux_lavf, without
actually doing anything.
Change the logic how the filename is passed to libavformat. Remove
handling of the filename from demux_open_lavf() and move it to
lavf_check_file(). (This also fixes a possible bug when skipping the
"lavf://" prefix.)
libavdevice now can be invoked by specifying demuxer and args as in:
mpv avdevice://demuxer:args
The args are passed as filename to libavformat. When using libavdevice
demuxers, their actual meaning is highly implementation specific. They
don't refer to actual filenames.
Note:
libavdevice is disabled by default. There is one problem: libavdevice
pulls in libavfilter, which in turn causes symbol clashes with mpv
internals. The problem is that libavfilter includes a mplayer filter
bridge, which is used to interface with a set of nearly unmodified
mplayer filters copied into libavfilter. This filter bridge uses the
same symbol names as mplayer/mpv's filter chain, which results in symbol
clashes at link-time.
This can be prevented by building ffmpeg with --disable-filter=mp, but
unfortunately this is not the default.
This means linking to libavdevice (which in turn forces linking with
libavfilter by default) must be disabled. We try doing this by compiling
a test file that defines one of the clashing symbols (vf_mpi_clear).
To enable libavdevice input, ffmpeg should be built with the options:
--disable-filter=mp
and mpv with:
--enable-libavdevice
Originally, I tried to auto-detect it. But the resulting complications
in configure did't seem worth the trouble.
This is consistent with the demuxer/decoder info output mpv already has,
and is also generally useful to know, especially if using
--ao=codec1,codec2,... syntax.
This caused e.g. "--alang=" (without anything following) to be printed
in the terminal output when the file specified no language for the
track. Introduced by commit 9085b8.
MPlayer/mplayer2 still show DVD subtitles in gray. Depending on who you
ask, this can be considered a bug or a feature. Include rendering in
gray as explicit feature, so the user can decide what is better.
This affects all indexed sub bitmaps entering the OSD rendering path.
Currently, this means all image subs are affected by this option, but
nothing else.
Apparently the -spugauss option was popular. The code originally
implementing this is gone (scaler stuff in spudec.c). Reimplement it
using libswscale to scale and blur image subtitles if the --sub-gauss
option is set.
The code does some rather lazy padding to allow the blur to spread
pixels past the original image bounding box. (This problem exists with
normal bilinear scaling too, but is barely noticable.)
Technically, this doesn't just blur subtitles, but anything RGBA (or
indexed) that enters the OSD rendering path. But only image subtitles
produce these OSD formats currently, so no explicit check is done to
prevent blurring in other cases.
Until now, screenshots with the video filter didn't add subs (unclear
whether that was an oversight or feature). Fix this and make behavior
when taking screenshots with vf_screenshot more consistent with VO
screenshots.
The change in vf_screenshot is needed, because add_subs() checks this
flag to decide whether it's allowed to mutate the image.
This commit has another user visible side effect. When taking a
screenshot each frame (using the "each-frame" mode of the screenshot
command), a normal screenshot command will stop the each-frame mode.
mp_image has this confusing distinction between the w/h and width/height
fields. w/h are the actual width and height, while width/height have a
very special meaning inside the video filter code: it's the actually
allocated width, which is also used for stride padding.
Screenshot related code abused the w/h fields to store the aspect
corrected size. Some code confused the role of w/h and width/height.
Fix these issues. For aspect corrected size, display_w/h are used, while
width/height should never be used outside vf.c internals and related
code.
This also fixes an actual bug when taking screenshots of anamorphic
video with vf_screenshot, as well as using vo_image with such videos.
Enable printf format warnings for set_osd_[t]msg.
Remove the pointless assertion in mplayer.c (the assertion proved that
the following NULL check is probably pointless, but leave that check
anyway for robustness - it's not really clear whether it's needed).
The playback status symbol in the OSD status display on video (such as
displayed when seeking or with the show_progress input command)
sometimes kept displaying the last seek, without resetting the symbol.
(For example: disable the OSD, seek, enable the OSD, run show_progress;
but also other cases.)
The main reason for that was the code clearing the OSD bar is also
responsible for clearing the osd_function (which stores the playback
symbol). If no OSD bar was set, the osd_function was never reset.
Fix by always setting the timer for clearing the OSD bar and the
osd_function whenever the osd_function is set. Clearing the OSD bar
when it wasn't set is OK. If the OSD bar is set some time after
osd_function is set, the timer is overwritten - that's a good thing,
as it makes both disappear from the screen at exactly the same time.
Always reset osd_function to 0 and determine the playback status
explicitly from mpctx->paused when displaying the status on screen.
