[FFmpeg-devel] [PATCH] set AVFrame decode_error_flags in case of decoding error by h264dec

[Marton Balint]

On Wed, 12 Jun 2019, Michael Niedermayer wrote:

> On Wed, Jun 12, 2019 at 10:09:08AM +0200, Marton Balint wrote:
>>
>>
>> On Wed, 12 Jun 2019, Michael Niedermayer wrote:
>>
>>> On Tue, Jun 11, 2019 at 03:21:41PM -0500, Amir Z wrote:
>>>> Thanks Michael Niedermayer for looking into this
>>>>
>>>> What I am trying to solve is having a way to detect concealed decoding
>>>> errors by the caller to avcodec_receive_frame.
>>>>
>>>> Should I add a general value e.g. #define
>>>> FF_DECODE_ERROR_DECODE_ERROR_OCCURRED 4 ?
>>>
>>> I suggest
>>> FF_DECODE_ERROR_CONCEALMENT_ACTIVE or something similar and then always
>>> set this for all cases of error concealment
>>> Its more informative than just knowing there was an error
>>
>> Concealment is a consequence. Error_flags should refer to the cause. A
>> generic UNKNOWN error seems much better to me if it is not feasible to
>> determine the cause.
>
> Concealment is the consequence generally, still the error in the frames
> differs between concealment or no concealment. A user application may
> want to treat these differently.
> concealemnt is not supported by all codecs currently and also not by
> all variants, for example interlaced material tends to be less supported
> in concealment. A user app might choose to discard a frame that
> contains errors but no concealemnt if the following frame is fine.
>
> Also in a very pedantic view, concealemnt itself is an error too.
> Its rarly known exactly where the damage starts so concealment often needs
> to cover more and by doing so adds errors in a minority of locations
> that is if you just want a formal argument why this would fit in here.
> Not an argument against the principle that concealemnt differs here in what
> it is, you are certainly correct about that.

OK, FF_DECODE_ERROR_CONCEALMENT_ACTIVE is fine then.

Thanks,
Marton