[FFmpeg-devel] Fwd: [PATCH] Automatically inserted bitstream filter 'h264_mp4toannexb' into hlsenc

[Michael Niedermayer]

On Wed, Jul 13, 2016 at 02:52:34PM +0200, Hendrik Leppkes wrote:
> On Wed, Jul 13, 2016 at 12:28 PM, Michael Niedermayer
> <michael at niedermayer.cc> wrote:
> > On Wed, Jun 29, 2016 at 01:07:13AM +0200, Hendrik Leppkes wrote:
> >> On Wed, Jun 29, 2016 at 12:30 AM, Steven Liu <lingjiujianke at gmail.com> wrote:
> >> >
> >> >
> >> > 2016-06-29 0:16 GMT+08:00 Hendrik Leppkes <h.leppkes at gmail.com>:
> >> >>
> >> >> On Tue, Jun 28, 2016 at 3:29 PM, Steven Liu <lingjiujianke at gmail.com>
> >> >> wrote:
> >> >> > before patched:
> >> >> > [root at localhost ffmpeg]# ./ffmpeg_g -re -i /root/facebook.mp4 -c copy -f
> >> >> > hls -v verbose -y aaaa.m3u8
> >> >> > ffmpeg version N-80779-gb18d6c5 Copyright (c) 2000-2016 the FFmpeg
> >> >> > developers
> >> >>
> >> >>
> >> >> The way I see it, the auto bsf from the mpegts muxer should already
> >> >> cover this without changes?
> >> >
> >> >
> >> > add the hls_check_bitstream into hls format and call the hls->avf to call
> >> > the mpegtsenc's check_bitstream,
> >> > don't do this operation,  it won't call the mpegtsenc's check_bitstream,
> >> > look at the message bellow:
> >> >
> >>
> >> This is the wrong approach to fixing this however, it should be
> >> understood first why its not doing this transparently right now.
> >> hlsenc uses the mpegts muxer like one would use any other muxer, so
> >> why does autobsf not get used?
> >>
> >> If we understand that, maybe it can be fixed in a more generic way,
> >> and not end up in ugly hacks in hlsenc.
> >
> > The issue is that hlsenc does call ff_write_chained() which causes
> > av_write_frame() instead of av_interleaved_write_frame() to be used,
> > bypassing the current auto bsf code
> >
> > Using av_interleaved_write_frame() would, i assume introduce a delay
> >
> 
> Maybe we should re-factor this BSF logic into a shareable function and
> make both av_write_frame and av_interleaved_write_frame use it then,
> instead of adding it to hlsenc (and after that even more
> playlist-muxers like dash, segment, etc)

patchset submited that does this
does this look ok to you ?

@Steven, does it work for you ?

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

I do not agree with what you have to say, but I'll defend to the death your
right to say it. -- Voltaire
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20160713/06183695/attachment.sig>