[FFmpeg-devel] [RFC][PATCH] libavformat: add librist protocol

[Paul B Mahol]

On Mon, Feb 22, 2021 at 11:02 PM Marton Balint <cus at passwd.hu> wrote:

>
>
> On Mon, 22 Feb 2021, Paul B Mahol wrote:
>
> > On Mon, Feb 22, 2021 at 8:27 PM Marton Balint <cus at passwd.hu> wrote:
> >
> >>
> >>
> >> On Mon, 22 Feb 2021, Paul B Mahol wrote:
> >>
> >> > On Mon, Feb 22, 2021 at 9:10 AM Gijs Peskens <gijs at peskens.net>
> wrote:
> >> >
> >> >> Hi, some feedback from one of the libRIST developers:
> >> >>
> >> >> Testing this patch (via diff found
> >> >> on:
> >> >>
> >>
> https://patchwork.ffmpeg.org/project/ffmpeg/patch/20210126205145.20448-1-onemda@gmail.com/
> >> >> ) I've noticed FFmpeg is not reading fast enough from the libRIST
> FIFO
> >> >> buffer (which is limited to 1024 packets) causing it to overrun.
> >> >> Unfortunately libRIST had a bug where it kept incrementing the buffer
> >> >> counter indefinitely, this has now been fixed and it will log an
> error
> >> >> whenever this happens (which is very often with FFmpeg).
> >> >> FFmpeg will need to read the FIFO faster, by calling the librist_read
> >> >> function more often. If that's not possible within the design of
> FFmpeg
> >> >> the module can either:
> >> >>
> >> >
> >> >
> >> > Unfortunately some other devs disagree with initial patch and thus
> this
> >> > issue.
> >>
> >> Could you pinpoint which change/discussion are you referring to here?
> >>
> >>
> > Probably one that is about longer polling time..
>
> I don't see how a larger polling time can cause such issue, unless
> something is very buggy. If there are queued packers in the fifo,
> rist_receiver_data_read should return immediately regardless of the
> polling time.
>

What about using some kind of for loop when reading?


>
> Regards,
> Marton
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".