[FFmpeg-devel] [PATCH 1/4] avcodec/h264dec: add missing flags to is_avc and nal_length_size AVOptions

[James Almer]

On 4/11/2021 10:25 AM, James Almer wrote:
> On 4/11/2021 6:56 AM, Anton Khirnov wrote:
>> Quoting James Almer (2021-04-10 23:00:20)
>>> Signed-off-by: James Almer <jamrial at gmail.com>
>>> ---
>>>   libavcodec/h264dec.c      | 5 +++--
>>> Â  tests/ref/fate/mov-zombie | 2 +-
>>> Â  2 files changed, 4 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/libavcodec/h264dec.c b/libavcodec/h264dec.c
>>> index 47b9abbc5c..f44c8c8175 100644
>>> --- a/libavcodec/h264dec.c
>>> +++ b/libavcodec/h264dec.c
>>> @@ -1027,9 +1027,10 @@ static int h264_decode_frame(AVCodecContext 
>>> *avctx, void *data,
>>> Â  #define OFFSET(x) offsetof(H264Context, x)
>>> Â  #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
>>> +#define VDE VD | AV_OPT_FLAG_EXPORT
>>> Â  static const AVOption h264_options[] = {
>>> -Â Â Â  { "is_avc", "is avc", OFFSET(is_avc), AV_OPT_TYPE_BOOL, {.i64 = 
>>> 0}, 0, 1, 0 },
>>> -Â Â Â  { "nal_length_size", "nal_length_size", OFFSET(nal_length_size), 
>>> AV_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0 },
>>> +Â Â Â  { "is_avc", "is avc", OFFSET(is_avc), AV_OPT_TYPE_BOOL, {.i64 = 
>>> 0}, 0, 1, VDE },
>>> +Â Â Â  { "nal_length_size", "nal_length_size", OFFSET(nal_length_size), 
>>> AV_OPT_TYPE_INT, {.i64 = 0}, 0, 4, VDE },
>>
>> Not objecting to the patch, but does anyone know how these fields are
>> useful to callers?
> 
> When i asked if these two were meant to be exported at all on IRC, Jan 
> Ekström argued that as exported avoptions they could be useful for 
> callers to know if the stream being decoded is annexb or size delimited, 
> so i figured i might as well keep them as is.
> As for being input options, if there's no extradata during init or 
> propagated as packet side data, situation where decoder will not 
> overwrite them but still use them when splitting NALUs, the caller can 
> set them manually and decoding could in theory succeed.
> 
> I don't know if the user that opened ticket #9176 cared about and used 
> them, or just noticed ffprobe printed different values in his own tests, 
> and reported it.

Will apply.