[x264-devel] Updated (and hopefully final) threaded slicetype patch

[Leon Woestenberg]

On Tue, Apr 7, 2009 at 11:40 PM, David DeHaven <dave at sagetv.com> wrote:
>
>>> No, it does not need to be (and may not behave properly if the mutex is
>>> still locked when the broadcast is done).
>>>
>> Explain please why it would not behave properly.
>
> I'm not sure... faint memory about it not working right, it could have been
> a logical error. It's been years since I debugged that code. Now that I
> think about it, you're probably right and it might actually be safer inside
> the critical section.
>
I tried both:

- changing "if"  to "while" - this is a must-fix item anyway.
- putting the broadcast inside the lock region, I think this is no
longer necessary with the fix above. (I have to verify this on a piece
of paper).

The deadlock stays.

Further debugging shows that one of the lists get full, and it never get() from.

Can someone give a high-level overview of the lookahead design, i.e.
how it's supposed to work. Thanks.

Leon