[FFmpeg-devel] [PATCHv2] avcodec/utils: do not reallocate packet buffer for AV_CODEC_ID_WRAPPED_AVFRAME

[wm4]

On Mon, 20 Feb 2017 21:11:50 +0100 (CET)
Marton Balint <cus at passwd.hu> wrote:

> On Sun, 19 Feb 2017, Hendrik Leppkes wrote:
> 
> > On Sun, Feb 19, 2017 at 2:41 PM, wm4 <nfxjfg at googlemail.com> wrote:  
> >> On Sun, 19 Feb 2017 14:35:42 +0100
> >> Marton Balint <cus at passwd.hu> wrote:
> >>  
> >>> Reallocating a wrapped avframe invalidates internal pointers, such as extended
> >>> data.
> >>>
> >>> FFmpeg has another way of passing AVFrames to muxers, but it seems the API
> >>> (av_write_uncoded_frame) is not implemented in the ffmpeg CLI yet.
> >>>
> >>> Signed-off-by: Marton Balint <cus at passwd.hu>
> >>> ---
> >>>  libavcodec/utils.c | 4 ++--
> >>>  1 file changed, 2 insertions(+), 2 deletions(-)
> >>>
> >>> diff --git a/libavcodec/utils.c b/libavcodec/utils.c
> >>> index f4085bf..184821a 100644
> >>> --- a/libavcodec/utils.c
> >>> +++ b/libavcodec/utils.c
> >>> @@ -1820,7 +1820,7 @@ int attribute_align_arg avcodec_encode_audio2(AVCodecContext *avctx,
> >>>      AVFrame *padded_frame = NULL;
> >>>      int ret;
> >>>      AVPacket user_pkt = *avpkt;
> >>> -    int needs_realloc = !user_pkt.data;
> >>> +    int needs_realloc = !user_pkt.data && avctx->codec_id != AV_CODEC_ID_WRAPPED_AVFRAME;
> >>>
> >>>      *got_packet_ptr = 0;
> >>>
> >>> @@ -1964,7 +1964,7 @@ int attribute_align_arg avcodec_encode_video2(AVCodecContext *avctx,
> >>>  {
> >>>      int ret;
> >>>      AVPacket user_pkt = *avpkt;
> >>> -    int needs_realloc = !user_pkt.data;
> >>> +    int needs_realloc = !user_pkt.data && avctx->codec_id != AV_CODEC_ID_WRAPPED_AVFRAME;
> >>>
> >>>      *got_packet_ptr = 0;
> >>>  
> >>
> >> I don't understand this logic in the first place. If nothing was
> >> encoded (!ret), and avpkt->data is set (why is it set?), then the
> >> AVPacket.buf is realllocated (why?) to the packet size (why???) - how
> >> does it make sense?  
> >
> > ret = 0 indicates successfull encode, ie. not an error code. What
> > result code did you expect?
> > AFAIK the realloc is performed to shrink over-sized pre-allocated
> > packets down and save memory.
> >  
> 
> So is it OK to apply the patch?

My suggestion doesn't work? (Would probably imply adding the padding in
wrapped_avframe.c.)