[MPlayer-dev-eng] threaded cache

[Reimar Döffinger]

On Sun, Nov 27, 2011 at 11:56:20AM -0500, compn wrote:
> On Sun, 27 Nov 2011 17:41:17 +0100, Reimar Döffinger wrote:
> >On Sun, Nov 27, 2011 at 06:32:03PM +0200, Andrej N. Gritsenko wrote:
> >late for most of our users.
> 
> some users want to only have one process.

Why? Also that is already possible by only changing configure.
It just will have worse performance than the current default
implementation without this patch.

> maybe it can be compile time option ?

I have no objections at all to adding a option to control
PTHREAD_CACHE directly, it should already work.
Though currently IMO more speaks against than for making it the
default.

> i dont care one way or the other, just trying not to bikeshed.

Maybe someone can come up with a better idea.
Explicitly doing a wakeup when reaching the end of cache or similar
should also reduce the issues.
I just can't understand why they came up with such a broken API.
Absolute timeouts have advantages, but if your implementation means
that timeouts _never_ work reliably (at least not portably) it's
quite irrelevant how many advantages they have in theory.