[vlc-devel] Re: Google SoC

[Remi Denis-Courmont]

On Thu, 01 Mar 2007 10:48:18 +0100, Clément Stenac <zorglub at diwi.org> wrote:
>> Event polling without arbitrary (CPU waking up) timers

> Except in the interfaces where this is handled by the frameworks and maybe
> in the network code, the background threads of vlc should by now be almost
> timer-free.

Network code is one problem I am trying to deal with, but it is not at all
easy. Unfortunately, replacing b_die with a thread condition does not work
here, and I assume no one wants to do thread cancellation. One option might
be a per-instance pipe (or socket pair on Winsock) that is used to wake up
poll() when an object is killed; and a lock to force every poll()ing thread
to wake-up and check if they are "running" the killed object. One pipe per
object looks way too heavy. One pipe per thread is not do-able because we
currently have no way to track object-to-thread mapping.

IIRC, the input playlist code still has timers. The handler for pf_run-less
interface is an example of this.

--
Rémi Denis-Courmont
http://www.remlab.net/

-- 
This is the vlc-devel mailing-list, see http://www.videolan.org/vlc/
To unsubscribe, please read http://developers.videolan.org/lists.html