Romain Vimont authored 5 years ago and Thomas Guillem committed 5 years ago

In ThreadDisplayPicture(), when "refresh" was true, the output parameter
deadline was not written and the function returned a non-zero value.

As a consequence, in video_output.c:Thread(), the next loop iteration
waited for the max deadline (100ms). When the following frame target
date was before this deadline, the video was stuttering.

To avoid the problem, write the deadline before returning from
ThreadDisplayPicture(), so that Thread() does not wait more than
expected.

Since an existing frame is refreshed only every 80ms
(VOUT_REDISPLAY_DELAY), this happened only on low framerate videos
(<12.5 fps). Otherwise, "refresh" was always false and the problem never
occurred.

Signed-off-by: Thomas Guillem <thomas@gllm.fr>

From the es_out, times are a triplet: pos, time, length. So they should be
sent via a single event and atomically.

This also suppress one player lock when times are updated.

cf. https://github.com/sahlberg/libsmb2/pull/94

Group every struct/enum/functions by group instead of having all structs, all
enums then all functions.

Split the header by the following doxygen groups:
 - Player instance
 - Playback control
 - Title and chapter control
 - Program control
 - Tracks control
 - Tracks synchronisation (delay)
 - Teletext control
 - External renderer control
 - Audio output control
 - Video output control
 - Player events

No functional changes.

Michael Young authored 5 years ago and Thomas Guillem committed 5 years ago

Since the underlying library supports them (since 3.0.0).

Signed-off-by: Thomas Guillem <thomas@gllm.fr>

That support port specifications.

That was added after the 3.0.0 release.

Felix Paul Kühne authored 5 years ago and Thomas Guillem committed 5 years ago

Signed-off-by: Thomas Guillem <thomas@gllm.fr>

The medialibrary now handles various mrls for the same device, so we can
remove this logic as it doesn't belong here anymore.

We don't need it anymore.

For non-opaque we copy in a local surface during prepare. It was already using
a single surface exposed as a pool. We do not use a pool in this case anymore.

We can remove is_slow as the core allocates its own "fast" pool in this case.

We can get it from the chroma description.

That includes Linux targets, in addition to Windows, iOS and Android.

If video_update_format_decoder() is called when loading the filter it works
just like before. If video_update_format_decoder() is called later when the
stream starts decoding then we can use the proper format.

It is not needed to load the symbols using dlsym here, they will be
automatically available if the OS version is recent enough or NULL if
not.

This is no longer needed and should not be done for a whole file.

- Fix needlessly strict cdat parsing and inverted logic
  (regression from 4.0.0-dev-8529-gb2b157076d)
- Fix cdt2 not parsed at all due to inverted logic
  (regression from 4.0.0-dev-8530-g8e8e0d7244)
- Fix allocation twice as large as needed.
- Remove bogus 64-bits boundary checks.

Abel Tesfaye authored 5 years ago and Jean-Baptiste Kempf committed 5 years ago

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Abel Tesfaye authored 5 years ago and Jean-Baptiste Kempf committed 5 years ago

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Abel Tesfaye authored 5 years ago and Jean-Baptiste Kempf committed 5 years ago

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

No need to.

unreadable

Alexandre Janniaux authored 5 years ago and Jean-Baptiste Kempf committed 5 years ago

It was failing on the following assertion in vlc_playlist_GoTo:
`index == -1 || (size_t) index < playlist->items.size'

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>