[FFmpeg-devel] Revert non-blocking protocol

[Ronald S. Bultje]

Hi,

On Sat, Mar 12, 2011 at 4:51 AM, Nicolas George
<nicolas.george at normalesup.org> wrote:
> Le primidi 21 vent?se, an CCXIX, Ronald S. Bultje a ?crit?:
>> >> As I stated I'd rather have that fixed/discussed, I'm having strange
>> >> issues doing strange stuff (sctp isn't exactly common) and Apparently
>> >> reverting those fixed that part.
>> > Does ffplay mms://... work, as per issue 2657? Does pressing "q" in
>> > ffmpeg work with local and remote files? Right now it doesn't.
>> And issue2659 also. I expect several of the next 10 to be the same bug.
>
> I think I understand the bug.
>
> There was a design problem before I started tinkering with the code:
> url_interrupt_cb provoked the return of the EINTR error code, which is
> completely at odds with its usual Unix semantic.
>
> The usual Unix semantic of EINTR is that you want to ignore it. One of the
> things that "3ce3b49 Non-blocking protocol: core wrapper functions"
> introduced was to ignore it systematically.
>
> Unfortunately, while this is correct when EINTR comes from Unix, this is
> wrong when EINTR comes from a nested protocol.
>
> The obvious solution seems to be to fix the design problem:
>
> ? ? ? ? if (url_interrupt_cb())
> - ? ? ? ? ? ?return AVERROR(EINTR);
> + ? ? ? ? ? ?return AV_ERROR_INTERRUPTED;
>
> (and of course the corresponding change on other wrong uses of EINTR).
>
> If you agree on the principle, I shall start to implement it as soon as
> possible.

It's fine with me as long as it solves the bug. Having ffmpeg or
ffplay hang is under no circumstances ever acceptable.

Ronald