[FFmpeg-devel] [PATCH 2/2] avcodec/mmaldec: fix deprecation warning

[Andreas Rheinhardt]

Ming Shun Ho:
> On Fri, Sep 24, 2021 at 12:52 AM Andreas Rheinhardt
> <andreas.rheinhardt at outlook.com> wrote:
>>
>> Ho Ming Shun:
>>> Signed-off-by: Ho Ming Shun <cyph1984 at gmail.com>
>>> ---
>>>  libavcodec/mmaldec.c | 2 ++
>>>  1 file changed, 2 insertions(+)
>>>
>>> diff --git a/libavcodec/mmaldec.c b/libavcodec/mmaldec.c
>>> index 5b75a1e74d..96140bf53d 100644
>>> --- a/libavcodec/mmaldec.c
>>> +++ b/libavcodec/mmaldec.c
>>> @@ -772,7 +772,9 @@ static int ffmmal_decode(AVCodecContext *avctx, void *data, int *got_frame,
>>>
>>>      if (avctx->extradata_size && !ctx->extradata_sent) {
>>>          AVPacket pkt = {0};
>>> +FF_DISABLE_DEPRECATION_WARNINGS
>>>          av_init_packet(&pkt);
>>> +FF_ENABLE_DEPRECATION_WARNINGS
>>>          pkt.data = avctx->extradata;
>>>          pkt.size = avctx->extradata_size;
>>>          ctx->extradata_sent = 1;
>>>
>>
>> You did not fix the underlying issue (av_init_packet() is deprecated and
>> its use should therefore be discontinued), but just hid the warning.
> 
> I was under the impression that it is deprecated to remove sizeof(AVPacket)
> from libavcodec's ABI. Internally however it should be ok.
> 

Some devs actually want to move AVPacket to libavutil (because it is
supposed to be cleaner) in the long term, so we already try to avoid
AVPackets on the stack in general. Note that get_packet_defaults() (the
actual successor to av_init_packet()) is internal to libavcodec/avpacket.c.

>> This patch here:
>> https://ffmpeg.org/pipermail/ffmpeg-devel/2021-March/277278.html intends
>> to fix the underlying issue (and also the warning). I have just not
>> found anyone to actually test whether it works. I presume you could do
>> it. Would you be so kind and do it?
> 
> Had to patch it to build:
> 
> diff --git a/libavcodec/mmaldec.c b/libavcodec/mmaldec.c
> index 0ea07ea787..2bed7309bb 100644
> --- a/libavcodec/mmaldec.c
> +++ b/libavcodec/mmaldec.c
> @@ -779,7 +779,7 @@ static int ffmmal_decode(AVCodecContext *avctx,
> void *data, int *got_frame,
>             return ret;
>     }
> 
> -    if ((ret = ffmmal_add_packet(avctx, avpkt, pkt->data, pkt->size)) < 0)
> +    if ((ret = ffmmal_add_packet(avctx, avpkt, avpkt->data, avpkt->size)) < 0)

How embarrassing. Good that it has never been applied in that form.

>         return ret;
> 
>     if ((ret = ffmmal_fill_input_port(avctx)) < 0)
> 
> Seems to be working fine with the above patch applied.
> 

Thanks for testing.

> Will this series likely be applied? I am asking because I am working on another
> patch to convert this to the new dataflow API, and your patches look like they
> might conflict.
> 

What new dataflow API? While it is very likely to lead to conflicts when
applying, I think it is unlikely to lead to actual deep conflicts that
can't be easily resolved. Do you already have a scetch of your patches
for the dataflow API for me to take a look?

- Andreas