This used to work; I'm not sure when or why it regressed. When setting
AVProbeData.filename to NULL, libavformat will crash in rtp_probe() by
unconditionally accessing the string.
We used to set the filename to NULL to prevent probing by file extension
when we don't deem it as necessary. Using an empty string also works for
this purpose.
This generally affects mp3 files that don't have any (or many) mp3
frames in the first 2 MB. 2 MB is the maximum probe size, and
libavformat returns a low probescore even if we give it the full 2 MB.
Trying to probe a larger buffer (or even the full file) doesn't work for
mysterious reasons.
The workaround consists in accepting a very weak probescore if the
format is detected as mp3 and we probed already 2 MB.
Restructure it a bit, so we can use the format hack list even if no mime
type applies. Shouldn't change anything functionally yet. Preparation
for the next commit.
MicroDVD files _can_ contain real timestamps instead of frame timestamps
if they declare a FPS. But this seems to be rare, so ignore that if the
FPS happens to match with the libavformat microdvd parser's default FPS.
This might actually break files that declare 23.976 FPS, but the video
file is not 23.976 FPS, but the chance that this happens is probably
very low, and the commit fixes the more common breakage with 25 FPS
video.
The TV code pretends to be part of stream/, but it's actually demuxer
code too. The audio_in code is shared between the TV code and
stream_radio.c, so stream_radio.c needs a small hack until stream.c is
converted.
Since m_option.h and options.h are extremely often included, a lot of
files have to be changed.
Moving path.c/h to options/ is a bit questionable, but since this is
mainly about access to config files (which are also handled in
options/), it's probably ok.
So, FFmpeg/Libav requires us to figure out video timestamps ourselves
(see last 10 commits or so), but the methods it provides for this aren't
even sufficient. In particular, everything that uses AVI-style DTS (avi,
vfw-muxed mkv, possibly mpeg4-in-ogm) with a codec that has an internal
frame delay is broken. In this case, libavcodec will shift the packet-
to-image correspondence by the codec delay, meaning that with a delay=1,
the first AVFrame.pkt_dts is not 0, but that of the second packet. All
timestamps will appear shifted. The start time (e.g. the time displayed
when doing "mpv file.avi --pause") will not be exactly 0.
(According to Libav developers, this is how it's supposed to work; just
that the first DTS values are normally negative with formats that use
DTS "properly". Who cares if it doesn't work at all with very common
video formats? There's no indication that they'll fix this soon,
either. An elegant workaround is missing too.)
Add a hack to re-enable the old PTS code for AVI and vfw-muxed MKV.
Since these timestamps are not reorderd, we wouldn't need to sort them,
but it's less code this way (and possibly more robust, should a demuxer
unexpectedly output PTS).
The original intention of all the timestamp changes recently was
actually to get rid of demuxer-specific hacks and the old timestamp
sorting code, but it looks like this didn't work out. Yet another case
where trying to replace native MPlayer functionality with FFmpeg/Libav
led to disadvantages and bugs. (Note that the old PTS sorting code
doesn't and can't handle frame dropping correctly, though.)
Bug reports:
https://trac.ffmpeg.org/ticket/3178https://bugzilla.libav.org/show_bug.cgi?id=600
This was needed to determine PTS from DTS, but the previous commits
make it unnecessary.
The builtin genpts hack was used for DVD, because libavformat's genpts
essentially went amok on DVD timestamp resets. See commit 65d87091 for
details.
Having the DTS directly can be useful for restoring PTS values.
The avi file format doesn't actually store PTS values, just DTS. An
older hack explicitly exported the DTS as PTS (ignoring the [I assume]
genpts generated non-sense PTS), which is not necessary anymore due to
this change.
This used to be needed to access the generic stream header from the
specific headers, which in turn was needed because the decoders had
access only to the specific headers. This is not the case anymore, so
this can finally be removed again.
Also move the "format" field from the specific headers to sh_stream.
demuxer->filepos contains the byte offset of the last read packet. This
is so that the player can estimate the current playback position, if no
proper timestamps are available. Simplify it to use demux_packet->pos in
the generic demuxer code, instead of bothering every demuxer
implementation about it.
(Note that this is still a bit incorrect: it relfects the position of
the last packet read by the demuxer, not that returned to the user. But
that was already broken, and is not that trivial to fix.)
This was originally added for better seeking where libavformat's seek
function won't work well: files with timestamp resets. In these cases,
the code tried to calculate an average bitrate, and then do byte based
seeks by multiplying the seek target time with the bitrate.
Apparently this was unreliable enough that the code was just commented
(and other parts became inactive). Get rid of it.
Note that the player still does byte based seeks in these cases when
doing percent-seeks.
Instead of having each demuxer do it (only demux_mkv actually did...),
let generic code determine whether the file is seekable. This requires
adding exceptions to demuxers where the stream is not seekable, but the
demuxer is.
Sort-of try to improve handling of unseekable files in the player. Exit
early if the file is determined to be unseekable, instead of resetting
all decoders and then performing a pointless seek.
Add an exception to allow seeking if the file is not seekable, but the
stream cache is enabled. Print a warning in this case, because seeking
outside the cache (which we can't prevent since the demuxer is not aware
of this problem) still messes everything up.
Pointless, using stream->start_pos/end_pos instead.
demux_mf was the only place where this was used specially, but we can
rely on timestamps instead for this case.
This one really did bite me hard (see previous commit), so enable it by
default.
Fix some cases of shadowing throughout the codebase. None of these
change behavior, and all of these were correct code, and just tripped up
the warning.
Change talloc destructor so that they can never signal failure, and
don't return a status code. This makes our talloc copy even more
incompatible to upstream talloc, but on the other hand this is
preparation for getting rid of talloc entirely.
(The talloc replacement in the next commit won't allow the talloc_free
equivalent to fail, and the destructor return value would be useless.
But I don't want to change any mpv code either; the idea is that the
talloc replacement commit can be reverted for some time in order to
test whether the talloc replacement introduced a regression.)
By default, libavformat uses UDP for rtsp playback. This doesn't work
very well. Apparently the reason is that the buffer sizes libavformat
chooses for UDP are way too small, and switching to TCP gets rid of this
issue entirely (thanks go to Reimar Döffinger for figuring this out).
In theory, you can set buffer sizes as libavformat options, but that
doesn't seem to help.
Add an option to select the rtsp transport, and make TCP the default.
Also remove an outdated comment from stream.c.
Retrieve per-chapter metadata, but don't do much with it. We just make
the metadata of the _current_ chapter available as chapter-metadata
property. Returning the full chapter list with metadata would be no
problem, except that the property interface isn't really good with
structured data, so it's not available for now.
Not sure if it's worth it, but it was requested via github issue #201.
The really funny thing about this commit is that this code is added on
top of another work around. Basically, subtitle seeking in libavformat
is completely broken. To make it useful, we have to add yet another
workaround.
The basic problem is that libavformat's subtitle seeking code always
uses the stream time base, instead of AV_TIME_BASE if stream index -1 is
passed to the avformat_seek_file() function.
Fixes github issue #216. Hopefully this will be fixed in ffmpeg too at
some point.
This is completely useless, and in this particular case, it broke the
fallback for MLP2 subtitles (stored as .txt files) to demux_subreader.
(Yes, libavformat should be fixed to handle this, but for now this will
_always_ break playback of subtitle files stored in .txt.)
You can still force this demuxer, but by default we will just pretend
that the "tty" demuxer does not exist.
Perhaps not very useful, but reserved for situations when a user reports
awful latency and experimentation/debugging might be required to find
out why or to fix it (happens often).
avio_alloc_context() is documented to require an av_malloc'ed buffer. It
appears libavformat can even reallocate the buffer while it is probing,
so passing a static buffer can in theory lead to crashes.
I couldn't reproduce such a crash, but apparently it happened to
mplayer-svn. This commit follows the mplayer fix in svn commit r36397.
Remove the (now unused) code for determining correct-pts mode based on
the demuxer in use. Change its description in the manpage to reflect
what this option does now.
Gives really funky results with PNG attachments otherwise. The main
problem is that avcodec_flush_buffers() does not fully reset the
decoder, so passing multiple PNG packets without keyframe flags will
attempt to combine the new picture with the previously decoded
contents. (Makes no sense with proper PNG - maybe this codepath is
intended for MNG or APNG.)
DVD playback had some trouble with PTS resets: libavformat's genpts
feature would try reading until EOF (worst case) to find a new usable
PTS in case a packet's PTS is not set correctly. Especially with slow
DVD access, this would make the player to appear frozen.
Reimplement it partially in demux_lavf.c, and use that code in the DVD
case. This is heavily "inspired" by the code in av_read_frame from
libavformat/utils.c. The difference is that we stop reading if no PTS
has been found after 50 packets (consider this a heuristic). Also, we
don't bother with the PTS wrapping and last-frame-before-EOF handling.
Even with normal PTS wraps, the player frontend will go to hell for the
duration of a frame anyway, and should recover quickly after that.
The terribleness of this commit is mostly that we duplicate libavformat
functionality, and that we suddenly need a packet queue.
All demuxers make a reasonable effort to set packet timestamps, and thus
support correct-pts mode. This commit also implicitly switches
demux_rawvideo to correct-pts mode.
We still allow demuxers to disable correct-pts mode in theory.
Get rid of the strange and messy reliance on DEMUXER_TYPE_ constants.
Instead of having two open functions for the demuxer callbacks (which
somehow are both optional, but you can also decide to implement both...),
just have one function. This function takes a parameter that tells the
demuxer how strictly it should check for the file headers. This is a
nice simplification and allows more flexibility.
Remove the file extension code. This literally did nothing (anymore).
Change demux_lavf so that we check our other builtin demuxers first
before libavformat tries to guess by file extension.
Before this commit, we tried to play along with libavformat and tried
to pretend that attached pictures are video streams with a single
frame, and that the frame magically appeared at the seek position when
seeking. The playback core would then switch to a mode where the video
has ended, and the "remaining" audio is played.
This didn't work very well:
- we needed a hack in demux.c, because we tried to read more packets in
order to find the "next" video frame (libavformat doesn't tell us if
a stream has ended)
- switching the video stream didn't work, because we can't tell
libavformat to send the packet again
- seeking and resuming after was hacky (for some reason libavformat sets
the returned packet's PTS to that of the previously returned audio
packet in generic code not related to attached pictures, and this
happened to work)
- if the user did something stupid and e.g. inserted a deinterlacer by
default, a picture was never displayed, only an inactive VO window)
- same when using a command that reconfigured the VO (like switching
aspect or video filters)
- hr-seek didn't work
For this reason, handle attached pictures as separate case with a
separate video decoding function, which doesn't read packets. Also,
do not synchronize audio to video start in this case.
The code touched by this commit makes sure that DVD subtitle tracks
known by libdvdread but not known by demux_lavf can be selected and
displayed properly. These subtitle tracks have the first packet
some time late in the packet stream, so that libavformat won't
immediately recognize them, and will add the track as soon as the
first packet is seen during normal demuxing.
demux_mpg used to handle this elegantly: you just set the MPEG ID of
the stream you wanted. demux_lavf couldn't do this, so it was emulated
with a DEMUXER_CTRL. This commit changes it so that new streams are
selected by default (if autoselect is enabled), and the playloop
simply can take appropriate action before the lower layer throws away
the first packet.
This also changes the demux_lavf behavior that subtitle packets are
always demuxed, even if not needed. (They were immediately thrown away,
so there was no advantage to this.)
Further, this adds the ability to demux.c to deal with demuxing more
than one stream of a kind at once. (Though currently it's not useful.)
AVDISCARD_DEFAULT is probably a bit better for normal decoding.
AVDISCARD_NONE would (as by documentation) include "useless" packets
too, while DEFAULT filters these.
Generally remove all accesses to demux_stream from all the code, except
inside of demux.c. Make it completely private to demux.c.
This simplifies the code because it removes an extra concept. In demux.c
it is reduced to a simple packet queue. There were other uses of
demux_stream, but they were removed or are removed with this commit.
Remove the extra "ds" argument to demux fill_buffer callback. It was
used by demux_avi and the TV pseudo-demuxer only.
Remove usage of d_video->last_pts from the no-correct-pts code. This
field contains the last PTS retrieved after a packet that is not NOPTS.
We can easily get this value manually because we read the packets
ourselves. Reuse sh_video->last_pts to store the packet PTS values. It
was used only by the correct-pts code before, and like d_video->last_pts,
it is reset on seek. The behavior should be exactly the same.
For now, we want to get rid of the demux->sub access, because this
field will become private to demux.c in a later commit. So replace the
current hack with another hack.
The need for the hack will be removed sooner or later. (Instead of
autoselecting a specific stream, all new streams will be enabled by
default, so that no packets can get lost. The frontend will then be
responsible to deselect unwanted streams.)
These separate arrays were used by the old demuxers and are not needed
anymore. We can simplify track switching as well.
One interesting thing is that stream/tv.c (which is a demuxer) won't
respect --no-audio anymore. It will probably work as expected, but it
will still open an audio device etc. - this is because track selection
is now always done with the runtime track switching mechanism. Maybe
the TV code could be updated to do proper runtime switching, but I
can't test this stuff.
Delete demux_avi, demux_asf, demux_mpg, demux_ts. libavformat does
better than them (except in rare corner cases), and the demuxers have
a bad influence on the rest of the code. Often they don't output
proper packets, and require additional audio and video parsing. Most
work only in --no-correct-pts mode.
Remove them to facilitate further cleanups.
wm4
Ben Boeckel
Stefano Pigozzi