[FFmpeg-devel] [RFC PATCH 2/2] ffmpeg_demux: make readrate stall warning input-specific
Gyan Doshi
ffmpeg at gyani.pro
Wed Apr 23 07:16:54 EEST 2025
On 2025-04-23 02:26 am, Marvin Scholz wrote:
> Given the readrate catchup is input-specific, it seems to make sense to
> also warn on a per-input demux basis.
We want to warn once each time reading is resumed but only for the first
stream resumption. That's the case right now.
With this change, if an input stalls thrice, only the resumption after
the first stall will get reported.
Regards,
Gyan
> ---
> fftools/ffmpeg_demux.c | 7 ++++---
> 1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/fftools/ffmpeg_demux.c b/fftools/ffmpeg_demux.c
> index 6e01efc420..1f7ec20a43 100644
> --- a/fftools/ffmpeg_demux.c
> +++ b/fftools/ffmpeg_demux.c
> @@ -101,6 +101,8 @@ typedef struct DemuxStream {
> int64_t resume_pts;
> // measure of how far behind packet reading is against spceified readrate
> int64_t lag;
> + // state for resume after stall warning
> + int resume_warned;
> } DemuxStream;
>
> typedef struct Demuxer {
> @@ -504,7 +506,6 @@ static void readrate_sleep(Demuxer *d)
> (f->start_time != AV_NOPTS_VALUE ? f->start_time : 0)
> );
> int64_t initial_burst = AV_TIME_BASE * d->readrate_initial_burst;
> - int resume_warn = 0;
>
> for (int i = 0; i < f->nb_streams; i++) {
> InputStream *ist = f->streams[i];
> @@ -523,12 +524,12 @@ static void readrate_sleep(Demuxer *d)
> ds->lag = lag;
> ds->resume_wc = now;
> ds->resume_pts = pts;
> - av_log_once(ds, AV_LOG_WARNING, AV_LOG_DEBUG, &resume_warn,
> + av_log_once(ds, AV_LOG_WARNING, AV_LOG_DEBUG, &ds->resume_warned,
> "Resumed reading at pts %0.3f with rate %0.3f after a lag of %0.3fs\n",
> (float)pts/AV_TIME_BASE, d->readrate_catchup, (float)lag/AV_TIME_BASE);
> }
> if (ds->lag && !lag)
> - ds->lag = ds->resume_wc = ds->resume_pts = 0;
> + ds->lag = ds->resume_wc = ds->resume_pts = ds->resume_warned = 0;
> if (ds->resume_wc) {
> elapsed = now - ds->resume_wc;
> limit_pts = ds->resume_pts + elapsed * d->readrate_catchup;
More information about the ffmpeg-devel
mailing list