[FFmpeg-devel] [PATCH v2] avcodec/pthread_frame, decode: allow errors to happen on draining

[Ronald S. Bultje]

Hi,

On Fri, Apr 28, 2017 at 6:19 AM, Muhammad Faiz <mfcc64 at gmail.com> wrote:

> So, all frames and errors are correctly reported in order.
> Also limit the numbers of error during draining to prevent infinite loop.
>
> This fix fate failure with THREADS>=4:
>   make fate-h264-attachment-631 THREADS=4
> This also reverts a755b725ec1d657609c8bd726ce37e7cf193d03f.
>
> Suggested-by: wm4, Ronald S. Bultje, Marton Balint
> Signed-off-by: Muhammad Faiz <mfcc64 at gmail.com>
> ---
>  libavcodec/decode.c        | 21 +++++++++++++++++++--
>  libavcodec/internal.h      |  3 +++
>  libavcodec/pthread_frame.c | 15 +++++++--------
>  3 files changed, 29 insertions(+), 10 deletions(-)
>
> diff --git a/libavcodec/decode.c b/libavcodec/decode.c
> index 6ff3c40..edfae55 100644
> --- a/libavcodec/decode.c
> +++ b/libavcodec/decode.c
> @@ -568,8 +568,24 @@ FF_ENABLE_DEPRECATION_WARNINGS
>          avctx->time_base = av_inv_q(av_mul_q(avctx->framerate,
> (AVRational){avctx->ticks_per_frame, 1}));
>  #endif
>
> -    if (avctx->internal->draining && !got_frame)
> -        avci->draining_done = 1;
> +    /* do not stop draining when got_frame != 0 or ret < 0 */
> +    if (avctx->internal->draining && !got_frame) {
> +        if (ret < 0) {
> +            /* prevent infinite loop if a decoder wrongly always return
> error on draining */
> +            /* reasonable nb_errors_max = maximum b frames + thread count
> */
> +            int nb_errors_max = 20 + (HAVE_THREADS &&
> avctx->active_thread_type & FF_THREAD_FRAME ?
> +                                avctx->thread_count : 1);
> +
> +            if (avci->nb_draining_errors++ >= nb_errors_max) {
> +                av_log(avctx, AV_LOG_ERROR, "Too many errors when
> draining, this is a bug. "
> +                       "Stop draining and force EOF.\n");
> +                avci->draining_done = 1;
> +                ret = AVERROR_BUG;
> +            }
> +        } else {
> +            avci->draining_done = 1;
> +        }
> +    }


Hm... I guess this is OK, it would be really nice to have a way of breaking
in developer builds (e.g. av_assert or so, although I guess technically
this could be enabled in prod builds also).

Also, Marton suggested to return AVERROR_EOF, maybe handle that here also
in addition to ret=0?

Ronald