[vlc-devel] commit: Mostly revert "libvlc: add mouse events as libvlc_MediaPlayerEvents " ( Rémi Denis-Courmont )

[Rémi Denis-Courmont]

On Thu, 18 Feb 2010 15:09:53 +0100, Jean-Paul Saman <jpsaman at gmail.com>
wrote:
> On Thu, Feb 18, 2010 at 8:58 AM, Rémi Denis-Courmont <remi at remlab.net>
> wrote:
>> ----- Message d'origine -----
>>> On Wed, Feb 17, 2010 at 11:07 PM, git version control
> <git at videolan.org> wrote:
>>> You could have
>>> suggested another way of implementing this during the review of this
>>> code.
>>
>> You seem to assume that I (a hobbyist) have a duty to review and provide
> solutions. I do not. If something does not work it is the author's duty
to
> reimplement it in a correct way. Not the reviewer.
> 
> You seem to have the impression that I asked you for an
> implementation. That was not my intention.

You asked how to do it. I don't want to waste my time thinking about that,
especially as I really do not like the basic concept of delivering mouse
events at all. I already spent enough of my free time reviewing it three
times, re-vectoring the code for VLC_VAR_COORDS, diagnosing the dead lock
bug and removing the code.

>> Making LibVLC not dead lock is helpful. Not publishing a bad API that we
> would need to support I do not know how long is helpfull too.

> Where does the deadlock occur?

There was a lock order inversion between the media player object lock and
the variable lock.

> so that I can fix it, How can I trigger the deadlock ?

Dead locks are timing-dependent. I don't know how to reproduce it reliably.

> If the implementation of remembering the vout in the media_player is
> the wrong approach, then what would be an alternative?

That is yours to figure out.

I wasted enough of my time on this bad idea already.

-- 
Rémi Denis-Courmont
http://www.remlab.net
http://fi.linkedin.com/in/remidenis