[vlc-devel] [PATCH] Revert "video_output: translate the mouse coordinates to display coordinates early"

[Alexandre Janniaux]

Hi,

On Wed, Nov 04, 2020 at 01:42:11AM +0800, Zhao Zhili wrote:
>
>
> > On Nov 3, 2020, at 10:09 PM, Rémi Denis-Courmont <remi at remlab.net> wrote:
> >
> > Le tiistaina 3. marraskuuta 2020, 13.56.58 EET Steve Lhomme a écrit :
> >> This reverts commit 5b2129eda58a4a19d37a9fb33e64af3777555233.
> >>
> >> This causes deadlocks with display modules where the display and windowing
> >> is tied.
> >
> > As I *already* pointed out earlier, no, this patch did not introduce any
> > deadlock. Pretty much all (other) window events also take the display lock.
> > That lock ordering constraint on window events is preexisting and remains with
> > or without this revert. Conversely, any deadlock would be preexisting and
> > remaining unfixed.
> >
> > Indeed, in the case of the MacOS vout, that deadlock bug is even explicitly
> > documented in the code already (look for FIXME).
> >
> > All this revert really does on (not already broken setups) is increase the
> > latency and thus the error on the mouse coordinate translation. With the
> > revert, translation is bottlenecked by whatever the video output control
> > thread does.
>
> Without the revert, I can't make any progress on any part of the code.
> I can keep the revert on my local branch for a period of time.

Which parts are you modifying? If that's only linked to thee video
output, you might want to give a try with the new window module
sent by Marvin a few days ago.

I'd prefer that we do not break old code when new one is being
written and even actually finalized to update it to the newer
paradigm. That's much like the «do not break the build during
week-end» request usually made here and on IRC.

Regards,
--
Alexandre Janniaux
Videolabs