[FFmpeg-devel] [PATCH] avcodec/av1dec: attempt to set SAR even if no hardware decoding is available

[James Almer]

On 1/9/2024 11:30 PM, Andreas Rheinhardt wrote:
> James Almer:
>> Should fix fate failures in avif tests when no external decoder is
>> available.
>>
>> Signed-off-by: James Almer <jamrial at gmail.com>
>> ---
>>   libavcodec/av1dec.c | 12 +++++++++++-
>>   1 file changed, 11 insertions(+), 1 deletion(-)
>>
>> diff --git a/libavcodec/av1dec.c b/libavcodec/av1dec.c
>> index da05a0f039..43b5aef687 100644
>> --- a/libavcodec/av1dec.c
>> +++ b/libavcodec/av1dec.c
>> @@ -1227,7 +1227,7 @@ static int av1_receive_frame_internal(AVCodecContext *avctx, AVFrame *frame)
>>                       av_log(avctx, AV_LOG_ERROR,
>>                              "Failed to get pixel format.\n");
>>                       s->raw_seq = NULL;
>> -                    goto end;
>> +                    continue;
>>                   }
>>               }
>>   
>> @@ -1248,6 +1248,16 @@ static int av1_receive_frame_internal(AVCodecContext *avctx, AVFrame *frame)
>>           case AV1_OBU_FRAME:
>>           case AV1_OBU_FRAME_HEADER:
>>               if (!s->raw_seq) {
>> +                // Attempt to set dimensions in all cases
>> +                if (s->pix_fmt == AV_PIX_FMT_NONE) {
>> +                    ret = update_context_with_frame_header(avctx, &obu->obu.frame.header);
>> +                    if (ret < 0) {
>> +                        av_log(avctx, AV_LOG_ERROR, "Failed to update context with frame header\n");
>> +                        return ret;
>> +                    }
>> +                    ret = AVERROR(ENOSYS);
>> +                    goto end;
>> +                }
>>                   av_log(avctx, AV_LOG_ERROR, "Missing Sequence Header.\n");
>>                   ret = AVERROR_INVALIDDATA;
>>                   goto end;
> 
> You should instead implement FF_CODEC_CAP_SKIP_FRAME_FILL_PARAM for this
> decoder so that one does not get a wall of error logs when probing
> without proper hardware.
> 
> - Andreas

I fixed the problem in a different way already, so this patch can be 
dropped.
I might look at FF_CODEC_CAP_SKIP_FRAME_FILL_PARAM later.