[FFmpeg-devel] [PATCH v1] avformat/mux: Set AV_PKT_FLAG_KEY for is_intra_only packet

[James Almer]

On 4/16/2020 10:23 PM, Limin Wang wrote:
> On Thu, Apr 16, 2020 at 05:16:35PM +0200, Nicolas George wrote:
>> Sorry, forgot to reply to that.
>>
>> Limin Wang (12020-04-04):
>>> the md5 framecrc is generated by libavformat/tests/movenc.c, I have no
>>> clue how the movenc testing work and how to get the real input to check why
>>> the frame isn't bitexact as I haven't see any subtitle track. 
>>
>> Well, check: re-run the test manually, change it to have an actual
>> output, examine the file. The possibilities are many.
>>
>> But tests are there for a reason. If a patch changes the output of
>> tests, then we cannot accept it unless somebody did examine the changes
>> very carefully and confirmed they are for the best. Ideally, an
>> explanation of the changes goes in the commit message.
> 
> Thanks, it's difficult to examine how it works for the tests/movenc.c without any
> documenets. As I can't give good explanation for the md5 changed so I had to stop
> to update the patch. 

The change is effectively just the keyframe flag being added to audio
packets in all of the movenc tests.

To verify, simply apply

diff --git a/libavformat/tests/movenc.c b/libavformat/tests/movenc.c
index 1d15d97ad9..0ff87da7d6 100644
--- a/libavformat/tests/movenc.c
+++ b/libavformat/tests/movenc.c
@@ -256,6 +256,7 @@ static void mux_frames(int n, int c)
             pkt.dts = pkt.pts = audio_dts;
             pkt.stream_index = 1;
             pkt.duration = audio_duration;
+            pkt.flags |= AV_PKT_FLAG_KEY;
             audio_dts += audio_duration;
         } else {
             if (frames == end_frames)

without your patch, and you'll get the same changes in the test.

Feel free to update your patch with what Andreas suggested.