When adding to the queue, this triggers rowCountChanged which then
calls canClearFinished, reading undefined memory for state if this is
not set. The correct state is set immediately after by checkInputPath.
Detected by UBSan.
When an external display is disconnected with a fullscreen projector
attached to it, Qt will trigger a resize of the window twice, which
makes the renderer queue 2 resize event blocks.
At the time when those blocks are run, Qt's window destructor will have
reset all pointers, and the block directly accesses pointers within
structures identified by pointers, which are invalid by that point.
This commit makes a block return early if the associated window has
been destroyed already and also explicitly checks for valid pointers
after.
Receiving buffers flagged as corrupted, or empty, is a casualty of
how things are implemented in various compositors, and it's not
a critical situation to be afraid of. We can expect a few buffers
to be flagged as corrupted here and there, and that's fine.
Demote these warnings to debug messages, as they're still useful
for debugging.
The crash reporter made with Qt was introduced in 824c7b0, but then was
quickly replaced by a native OS implementation in d42a7ce. This code has
been lying around ever since.
A trivial cosmetic change. This groups the constructor and
destructor together. Two sides of the same coin, yin and yang,
bound by their nature, and linked by their position in code.
This commit introduces an important distinction in the initialization
process that is essential to future camera and audio work: it splits
creating and connecting to the PipeWire socket, from connecting to
specific PipeWire nodes.
Right now, for the only consumer in existence - the ScreenCast portal
code - this distinction is irrelevant, but from an API perspective it
just makes sense to model it this way.
We don't actually need to restore the node id anymore, since this
is handled by the screencast-portal.c code now.
Remove the pipewire_node field, and just create the stream directly
from the passed file descriptor now.
This separation of obs_pipewire_create() and play_pipewire_stream()
was an artifact of how the original code was written, and there is
no reason to keep this separation anymore. Inlining it will help
future commits too.
And let each portal pass the stream properties relevant to them.
This makes the pipewire.c more independent of the actual portal
by providing a wrapper to create a stream. This will be relevant
later, when linux-pipewire introduces more portals like the Camera
one.
When the linux-pipewire plugin is unloaded, make sure to cleanup
the D-Bus proxy for the ScreenCast portal too. This effectively
doesn't change anything, but it's always good to keep up with the
code hygiene.
Previously we would actually initialize a texture memory that would then
also have to be deleted when we bound the EGLImage to the texture during
dmabuf import.
Instead simply do not create the dummy texture memory. One odd thing is
that we must still query the texture to ensure its initialized or
binding the EGLImage will not work. So we leave the TEXTURE_MAX_LEVEL
check.
This makes screencapture up to 100x faster on discrete intel cards and
likely has some performance benefit for amd/integrated cards. Without
this dmabufs are actually slower than shared memory for these intel
cards.
Copying the encoder list takes a while which blocks the main thread.
Doing the copying asynchronously removes instead of blocking the main
thread improves startup performance by about 60ms.
If an user exits the source tree edit mode, without changing the name,
the scene item would become undraggable until the scene is refreshed.
We were setting the focus to just the label, not the whole widget.
Qt 6.2 introduced an issue where SingleSelection item views would
deselect the current item if the user clicked on empty area in the
widget.
This was very confusing in the scene tree as it was now possible to
unselect the current scene. A workaround for this in OBS was added in
08e4ee6 and expanded on in dc30cf0, but being quite hacky it never was
the perfect solution.
I since dug into Qt and fixed the issue upstream in qt/qtbase@f11e5435c7
for Qt 6.4.3 and newer, so we can remove the workaround when using those
versions (with the long term goal of removing the code altogether).
In some cases, `QToolbar::widgetForAction` may return `nullptr`. This
causes a crash on affected system.
To mitigate this, we check for `widgetForAction` result and skip
operating on a NULL widget.
When a variable `pos` became larger than `AUDIO_OUTPUT_FRAMES`, `count`
will get overflowed number. To avoid the overflow, continue the loop
when `AUDIO_OUTPUT_FRAMES - pos` is not positive.
The aac spec has a list of default channel layouts that are implemented
by all aac encoders / decoders.
Unfortunately 2.1 is not in that list. When I wrote the surround sound
support into OBS, as a workaround I decided to encode 2.1 as
AV_CH_LAYOUT_SURROUND = FL+FR+FC.
The LFE channel is encoded as a regular audio channel which uses more
bits but this allows compatibility with all aac decoders.
Recently we updated to the new channel layout API of FFmpeg, but the
mapping of 2.1 to AV_CH_LAYOUT_SURROUND was forgotten in the FFmpeg
native aac encoder. This is remedied here.
Signed-off-by: pkv <pkv@obsproject.com>
The `srt_bstats()` function is not guaranteed to zero the stats object
before filling in stats. If a socket was connected, then disconnected,
without any data being sent over it, the stats object would remain
uninitialized, and weird results could occur.
