On Fri, 20 Nov 2015 13:30:50 +0100
Stefano Sabatini <stefa...@gmail.com> wrote:
> On date Wednesday 2015-11-18 15:21:15 +0100, Nicolas George encoded:
> > Le septidi 27 brumaire, an CCXXIV, Stefano Sabatini a écrit :
> > > Fix demuxing of files for which the demuxer returns AVERROR(EAGAIN) at
> > > some point. Also returns error code to the caller in case of different
> > > non EOF error.
> > > ---
> > > ffprobe.c | 11 ++++++++++-
> > > 1 file changed, 10 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/ffprobe.c b/ffprobe.c
> > > index c304a6d..481ff0b 100644
> > > --- a/ffprobe.c
> > > +++ b/ffprobe.c
> > > @@ -2027,7 +2027,16 @@ static int read_interval_packets(WriterContext *w,
> > > AVFormatContext *fmt_ctx,
> > > ret = AVERROR(ENOMEM);
> > > goto end;
> > > }
> > > - while (!av_read_frame(fmt_ctx, &pkt)) {
> >
> > > + while (1) {
> > > + ret = av_read_frame(fmt_ctx, &pkt);
> > > + if (ret == AVERROR(EAGAIN))
> > > + continue;
> >
> > This is a busy wait, this is one of the evilest constructions possible.
> >
>
> > For real EAGAIN cases, a sleep would be needed, until we have a real event
> > loop. Other cases are bugs in the demuxers.
>
> Yes, updated patch with EAGAIN (if this is fine, I'll send patches to
> fix the examples).
>
> In my case the EAGAIN is triggered by the flv demuxer code in
> flvdec.c:935:
>
> /* skip empty data packets */
> if (!size) {
> ret = AVERROR(EAGAIN);
> goto leave;
> }
>
> introduced in 527c2e64295671bfcd9c86ca2cfd75e0f4d79f73.
>
> Not sure if this is a defect.
Isn't it actually fcb4228c863ec80c9498a6c5e815d79ee9f89a7f and
a308737bd66a1c669b9be71a1b834d8419fe70ca?
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
