Per FFmpeg commit 337f777f378c [1], FFmpeg removed nvenc_h264_encoder
and nvenc_hevc_encoder after deprecation in FFmpeg commit 888a5c794778
[2]. The names to be used are ff_h264_nvenc_encoder and
ff_hevc_nvenc_encoder. So we must allow alternative search of codec as
h264_nvenc or nvenc_h264 in obs-ffmpeg.c.
[1]: 337f777f37
[2]: 888a5c7947
AVFormatContext::oformat was made const on April 27, 2021 [1]. If we
respect the constness of AVOutputFormat and do not cast results from
FFmpeg functions to non-const, we cannot modify the results after the
fact. Our choices are either to cast them to non-const (and presumably
have them implicitly casted back to const on later function calls), or
only try to modify the results in versions of FFmpeg where these are not
expected to be const.
Instead of relying on casts, we can set the encoder values in the
ffmpeg_cfg struct, which are later passed to new_stream.
Also modifies deps/media-playback. Removes compiler warnings.
Some avformat functions return const AV(In/Out)putFormat per [1], so
ifdef as needed.
[1]: 56450a0ee4
lavf 59.0.100 avformat.h
avformat: Constify the API wrt AV(In|Out)putFormat
Also constify AVProbeData.
avcodec.h stopped including channel_layout.h per FFmpeg commit
1be3d8a0cb77 [1]. avformat.h stopped including avcodec.h per FFmpeg
commit e67e02d15672 [2]. As a result, we need to explicitly include
avutil/channel_layout.h when needed. Fixes compilation error against
FFmpeg later than the two mentioned commits.
[1]: 1be3d8a0cb
[2]: e67e02d156
FFmpeg commit e67e02d15672 [1] drops avcodec.h from avformat.h includes.
Therefore we need to include it explicitly.
[1]: e67e02d156
lavf/avformat.h: drop the avcodec.h include
AVFrame.pkt_pts was deprecated in FFmpeg commit 32c8359093d1 dated June
21, 2016 for lavc 57.24.0 (FFmpeg 3.2). AVFrame.pkt_pts was replaced by
AVFrame.pts, and it was removed in FFmpeg commit 6e30b35b85b8 [2] on
April 27, 2021. Ubuntu 18.04 has the oldest FFmpeg version we support,
FFmpeg 3.4. We should be able to safely migrate from AVFrame.pkt_pts to
AVFrame.pts.
[1]: 32c8359093
[2]: 6e30b35b85
obs-websocket's `GetInputDefaultSettings` request uses
obs_get_source_defaults() in order to fetch a source id's default
settings. This moves the default settings to the `.get_defaults`
callback so that our request does not return an empty object.
Does not affect existing functionality
As Intel continues to add discrete graphics cards, maintaining a long
list of discrete device IDs to check against isn't wise. When i+I
platforms are detected, default to integrated graphics.
Instead of trying to open devices /dev/video<0-63> check /dev for
all /dev/videoX devices that are existent in the tree. This allows
finding devices with higher names than 63. E.g. /dev/video100 can
now be found and opened which previously failed.
_GNU_SOURCE #define is introduced for versionsort() which is a GNU
extension.
Fixes#4347.
The commit 4a5ac3974 fixed the bug the first PTS is not zero. This
commit fixed the same issue in replay-buffer.
A variable `video_dts_offset` is renamed to `video_pts_offset` and the
variable is initialized by `pts` instead of `dts`.
This adds RIST to the list of protocols supported by the obs
ffmpeg_mpegts_muxer.
RIST is container and codec agnostic. But this implementation relies
on ffmpeg avformat integration and uses mpeg-ts container.
Intel committed an NDA disclaimer on each source file. The stated
intention was that the NDA "added to OBS doesn't apply to open source
code once it's been accepted by the community. You can remove it for
your modifications". This quote is from an email chain involving
Intel's legal team and developers. The NDA in the source files
mistakenly triggers source code scanners that look for license
violations. I have removed the comments that contain the NDA.
When video is disabled, no audio was sent to the container so that the
audio was not saved. Prior to this change, all audio was discarded until
the first video packet arrives. This change limits to discard audio only
if video is available.
At least one V4L2 device reports a menu control when queried with
QUERYCTRL yet does not report supporting a single value in its legal
range when queried with QUERYMENU. Such devices are arguably
ill-behaved, but the Linux UVC driver takes no special care to ignore
such pathological controls, so we have to do it ourselves.
Without this patch, a menu control with no valid values would cause
v4l2_control_changed to go into an infinite loop, since every S_CTRL
request on such pathological control properties would always fail,
triggering an additional modification event. By ignoring these
properties altogether, we avoid that problem, and the user benefits by
not having the UI cluttered by a confusing and useless menu control.
Remove lastNotifyTime throttle. Could ignore last device notification.
Will bring back if necessary, but seemed like overkill.
Make isInputDevice const for safety.
Make isDefaultDevice atomic. Allows it to be accessed simultaneously
from settings and notification callback threads. Race between setting
change in OBS and Windows is not a problem, a signal will be sent after
both to consolidate. Worst that should happen is a redundant reconnect
cycle.
Only read/write default_id from notification callback for thread safety.
Do not store IAudioClient and IAudioCaptureClient onto the source object
until initialization is almost complete, right before the capture thread
is created. We don't to retain objects from failed attempts. Also clear
them when stopping.
