[vlc-devel] [RFC] core: player: make the lock optionally reentrant

[Romain Vimont]

On Mon, May 13, 2019 at 05:53:37PM +0300, Rémi Denis-Courmont wrote:
> Le maanantaina 13. toukokuuta 2019, 17.27.58 EEST Romain Vimont a écrit :
> > On Mon, May 13, 2019 at 05:06:45PM +0300, Rémi Denis-Courmont wrote:
> > > Hi,
> > > 
> > > Recursive locking only addresses backward compatibility in the corner case
> > > than the call is from the same thread. Since that was never a requirement
> > > before, that's not backward compatible.
> > In practice, I think this is always the case. Anyway, this is the case I
> > want to continue to support with this change.
> 
> You can argue all you want that that was unreasonable design, but that's what 
> it was, and there is no shortage of "unreasonable" uses of LibVLC.
> 
> I don't care if we break backward compatibility, but I do care if it's done 
> silently.

Just to clarify, when I suggested using recursive mutex, I didn't
thought about the "blocking-dispatched to another thread" case, so thank
you for having pointed it out.

Now, I think this is acceptable to break this case, since in practice,
it probably won't hurt (and we're allowed to break).