It was posted exactly once as the first control and never again. Instead,
this handles the event implicitly when the video output thread starts.

The control queue processing flag must now be initially set, so that the
first WaitEmpty() actually waits for the video output thread to
initialize and get ready to process events.

Generate the OSD title subpicture directly, and only then queue the
resulting SPU to the video output as an event.

Do not needlessly dispatch through the vout thread.

...rather than defer it to the video output thread.

Move the concerned initialization from the vout thread into the existing
non-threaded initialization code.

Juha-Pekka Heikkila authored Aug 10, 2018 and Rémi Denis-Courmont committed Aug 13, 2018

Address issues in KMS plugin pointed out by Steve Lhomme

Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

KO Myung-Hun authored Aug 07, 2018 and Rémi Denis-Courmont committed Aug 13, 2018

Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

fix mentions documentation which says that's... BGR

This reverts commit 1ed99377.

Zhao Zhili authored Aug 10, 2018 and Thomas Guillem committed Aug 10, 2018

This allows the following use case: the user of libvlc do render itself with
OpenGLES and interoperate with libvlc via SurfaceTexture, just as what VLC
gles2 module does.

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

This reverts commit 4594dedf.

This is the wrong way to do it. And breaks other things like magnify.
Either the visible area should account for the x/y offsets or there should be
and offset stored per plane.

We also need to decide in a picture copy copies only the visible area (and not
event the part in the x/y offset) or some of the decoder/crop padding too. The
current code leans towards the latter but still copies too much.

Romain Vimont authored Jul 18, 2018 and Thomas Guillem committed Aug 09, 2018

TAB_FIND() was implemented, but not ARRAY_FIND(). It may be useful.

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

Romain Vimont authored Jul 18, 2018 and Thomas Guillem committed Aug 09, 2018

Add some tests for ARRAY_* macros.

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

cf. videolan/vlc-ios#261

A lot of crashes happen when doing the CVPX HW -> SW conversion. Unfortunately,
I was unable to reproduce it.

My best guess is that this buffer overflow is caused by the i_visible_height
that is too big.  This value should be correct since it comes from the hxxx
packetizer, but it may be different from the actual height of the cvpx buffer.

regression from 4c8b741e

The old code does not take into account the index shifts if
dragging an item down the list.
Easy fix is to reload the whole model for now.

Olivier Fourdan authored Jul 31, 2018 and Hugo Beauzée-Luyssen committed Aug 08, 2018

commit 58155349 ("qt: make the time tooltip a tooltip") reintroduced the
bug fixed by previous commit ea992379 ("remove X11BypassWindowManagerHint
from time tooltip") meant to fix #17829.

The tooltip window is a shaped, override-redirect window which shows
when the pointer enters the slider and disappears when it leaves the
slider.

By placing the tooltip window too low, above the slider, vlc causes a
constant map/unmap of that window which will flood Xwayland and the
Wayland compositori.window manager and can up with a deadlock situation.

When the tooltip window is showing, the pointer will enter the tooltip
window and thus leave the slider, which will cause the tooltip to be
unmapped, which in turn causes the pointer to re-enter the slider area,
which will cause the tooltip to show again, so on and so forth.

On Wayland/Xwayland where the Wayland compositor (i.e. the display
server) is also the window manager, that can lead to a deadlock
situation because some X11 requests are blocking on rountrips to the
Xserver, which in the case of Xwayland is also a Wayland client, waiting
on the Wayland compositor for some operations, the Wayland compositor
being also a Window manager and an X11 client at the same time...

Avoid the issue by placing the tooltip so that it does not show up at
the same location where the slider is, without overlap, to avoid that
map/unmap requests storm.

Closes: https://gitlab.gnome.org/GNOME/mutter/issues/244


Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>