[FFmpeg-devel] [PATCH v4] avformat/hlsenc: merge mpegts and fmp4 workflow to one workflow

[Andreas Rheinhardt]

Liu Steven:
> 
> 
>> 在 2019年8月9日，下午5:30，Liu Steven <lq at chinaffmpeg.org> 写道：
>>
>>
>>
>>> 在 2019年8月5日，上午10:29，Steven Liu <lq at chinaffmpeg.org> 写道：
>>>
>>> just remove the 'i' of the v3 mail subject.
>>> write mpegts or fmp4 context into buffer, and flush the buffer into
>>> output file when split fragment. merge two format split workflow into
>>> one workflow
>>>
>>> Signed-off-by: Steven Liu <lq at chinaffmpeg.org>
>>
>> ping?
>>
> Pushed

This patch both created and fixed memleaks: Before this patch,
generating hls_segment_size.m3u8 (needed for fate-hls-segment-size)
leaks 792 B directly and 787122 B indirectly. With this patch,
Generating hls_segment_size.m3u8 and hls_init_time.m3u8 leaks 264 B
directly and 1056 B indirectly.
The leaks happen in hls_write_trailer: The last dynamic buffer that is
opened in flush_dynbuf is not freed. I wanted to add a call to
ffio_free_dyn_buf, but then I noticed that it seems that if something
gets written in av_write_trailer(oc) it never leaves the dynamic
buffer, although it probably should. So I leave fixing this to those
who know this code better than I do.

- Andreas

PS: For the fate test run with Valgrind this means that generating the
necessary test files fails. But the tests itself are not counted as
failing on the website [1]. Instead, the number of tests decreased.

[1]: http://fate.ffmpeg.org/history.cgi?slot=x86_64-archlinux-gcc-valgrind