[FFmpeg-devel] [PATCH] avcodec/bsf: restructure the internal implementation of the bistream filter API

[James Almer]

On 4/17/2020 6:11 PM, James Almer wrote:
> On 4/17/2020 5:40 PM, Andreas Rheinhardt wrote:
>> James Almer:
>>> On 4/17/2020 3:49 PM, Andreas Rheinhardt wrote:
>>>> What is the caller now supposed to do with the packet he has? Submit it
>>>> again for filtering?
>>>
>>> In case of failure, you should abort the process or flush and attempt to
>>> resume filtering in a fresh state. Trying to send more data after you
>>> got an invalid data error is not going to get good results.
>>>
>> If one sends data different from the data that led to the invalid data
>> error one might very well get good results. That is what is being done
>> for ages in ffmpeg.c as well as mux.c and probably a few other places.
>> That's also why it is so important to know whether it was the data that
>> one just sent that caused the error or not.
>>
>> Your advice is btw against the current documentation (the part which you
>> want to remove, but not because of the error handling)
> 
> True that the CLI does not abort on error by default. But I'm not sure
> what part of my doxy change is against the above advice. I only
> documented the fact av_bsf_send_packet() returning EAGAIN must not be
> considered an error.

Nevermind, i realize now you were talking about the doxy change in this
patch. Sorry about that.

I removed it because it of course would no longer be required to call
av_bsf_receive_packet() until the filter is drained after a change like
this, when you could feed new packets as they are filtered. It would
still be possible, but no longer the only option.