No need to reset mp_pass_perf which is only read when desc.len > 0.
This avoids zeroing >16384 uint64_t's every time it's called.
Profiling shows that this reduces CPU usage for frame rendering by ~4%.
Since d41f0a54b0, enabling a stream will
no longer perform a refresh seek, to fix issues with switching video
tracks. Additionally, 62e9a0c5f6 also
prevents refresh seek on track switching when it happens right after a
seek.
Unfortunately, when external audio files are loaded, preventing refresh
seeks can cause A-V sync issues. Since external tracks have separate demuxer
instances from the internal tracks, the demuxer instances from both
internal and external tracks are explicitly sought to the same pts when
seeking is performed. When enabling an external audio track for the first
time, the existing logic prevents refresh seek because no packets have ever
been read, so the track ends up not being sought. This means that switching
the track in the middle of playback results in a huge A-V desync.
To make it worse, unlike one demuxer instance with multiple tracks,
tracks from multiple demuxer instances are not synced in any meaningful
way beyond the separate explicit seeking. The only thing which keeps these
tracks synced is the generic A-V sync logic, which is unfit for such large
desync above. This means the audio is at the start position after the track
is switched, behind the video, so the audio is not considered ready, and
audio is continuously filtered sequentially until it is synced to the video.
While this is happening, the audio filtering exhausts all CPU resources.
There is also a case with cache enabled: if a seek causes some new data to
be cached, the demuxer is sought to the end of joined cache range.
If track is switched after this, the existing logic prevents a refresh seek,
so the demuxer is at a wrong position after the track is switched.
Fix this by allowing refresh seek for non-video streams, even when no packets
have been read yet or immediately after a seek.
Preselect and seek to the correct subtitle line when there is a
sub-delay.
Also slightly increase the offset without a video track else it
occasionally doesn't seek to the selected line with non-0 sub-delay and
no video. This now uses the same offset as sub/sd.h.
There is no need to compare sub-start now that we don't check for exact
matches with ffmpeg's timestamps. time-pos can always be used.
Comparing time-pos removes the need for the workaround of sub-start of
embedded subtitles being earlier than ffmpeg's timestamps, as time-pos
is always slightly after sub-start.
Removing this workaround also fixes a bug present since 4059d1832b,
which started comparing numerical values of timestamps instead of
strings to determine the preselected subtitle line, where the line after
the current one is preselected when sub-start is greater than ffmpeg's
timestamps because they have been rounded, e.g. sub-start 10.001 is >
than ffmpeg's 10.00.
This will also allow considering --sub-delay in the comparisons, as
comparing sub-start - sub-delay would preselect the wrong lines.
Since 995c47da9a, setting --mute=auto has
been equivalent to --mute=no. It was formally documented later in
79e20ff485. This is an old legacy relic
and the auto choice popping up during auto completion could be confusing
to users who aren't aware of the history. Remove it for good and convert
the option to a proper boolean.
Since 7fc4bac29f, playback-time and
time-pos will now always have exactly the same values. Both properties
are commonly used, so just make playback-time an alias for
time-pos and document it.
withLock is available for macOS versions starting from 10.10 but can
only be used on xcode 14+. xcode 14 introduced swift 5.7, so we guard it
behind a swift version check.
Hashbangs are meant for scripts that are executed, but a bash completion
script is meant to be sourced and therefor shouldn't have a hashbang.
Remarked by Debian's ``lintian`` tool with the
``bash-completion-with-hashbang`` tag.
Limit list entries to 100. obj_settings_list is not designed to hold
more items, and it quickly starts taking ages to add all items. 100 is
more than enough.
Fixes 30s timeout on OSS-Fuzz and generally fixes possible DoS on mpv.
"file" is implicit key, it is always available, so we have to check if
it is not empty.
Found by OSS-Fuzz.
Fixes: 96ef62161ac8edb3c111bde7a79cb07dd6db813c
If the output display reports high bitdepth, 12-bit and 10-bit swapchain
is used the value returned from d3d11_color_depth needs to take it into
account.
The options parameter here can be NULL. It's NULL checked a few lines down,
but not for the propagate_opts loop. This results in null derefence with
--demuxer-lavf-format=image2.
Fixes: b6413f82b2
VapourSynth introduced their version 4 API in R55, 3 years ago. mpv is
still using the deprecated version 3. I think it is good to migrate
before VapourSynth completely removes it.
The most impacted area is the video format. Previously we have a fixed
table of supported formats, each represented by an integer. In v4, the
integer is replaced by pointer to the full struct of the format.
Second, the way to create video filter is changed. Previously caller
needs to supply a "initialization" callback to `createFilter()`, which
sets up video info etc. In v4, video info is prepared first, passed to
the `createVideoFilter2()` and we get back the node.
Also, previously mpv was using the `fmSerial` filter mode, which means
VapourSynth 1) can only request one frame from mpv at a time, and 2) the
order of frames requested must be sequential. I propose to change it to
the parallel request mode, which requests multiple frames at a time in
semi random order. The reasons are, for 1), the get frame function,
`infiltGetFrame()`, unlocks the mutex during output image generation,
thus can benefit from parallel requests. For 2) thanks to the frame
buffer, unordered frame requests are acceptable. Just make sure the
buffer is large enough.
Third, the way VapourSynth scripting environment works change. In v4,
the scripting API is operated through struct pointer, much like the
exist `vsapi` pointer. The "finalize" function is also no longer needed.
There are jpg files out there that have extra embedded metadata
(pictures from smartphones commonly have an embedded gain map). ffmpeg
doesn't currently support this, so mpv currently sees the extra metadata
as essentially another frame. This results in weird video-like behavior.
Until ffmpeg support this more properly, we can work around this by
simply discarding extra packets and not sending the new frame to the
internal playloop. If demux_lavf detects that the stream is a
single-frame jpg, then we only accepts the packet at pos 0. Anything
else is discarded. Ref #13192.
Passing jpg filenames to ffmpeg is actually quite bad. This causes all
jpg images (at least in my testing) to be probed as image2 which is
blacklisted in demux_lavf since it's completely broken (see commit
74e62ed2d1 for some details). What happens
in practice is that the lavf demxuer fails and the it opens in the mf
demuxer instead. This is OK for simple viewing, but because that demuxer
is limited, many specific file properities and other things are
completely unavailable which breaks any script that may depend on them
(e.g. width, height, etc.) For the small subset of files that this
commit appeared to "fix" (in reality, it just fell back to the mf
demuxer) is not worth breaking property usage in the vast majority of
normal proper files. Ideally ffmpeg should fix this but some other
workaround on our end can be used instead.
This reverts commit d0aeca5918.
