[FFmpeg-devel] [PATCH] avformat: Allow forcing use of AVParsers

[Ronald S. Bultje]

Hi,

On Sep 5, 2016 12:05 PM, "Michael Niedermayer" <michael at niedermayer.cc>
wrote:
>
> On Mon, Sep 05, 2016 at 05:25:14AM -0400, Ronald S. Bultje wrote:
> > Hi,
> >
> > On Sep 5, 2016 11:18 AM, "Michael Niedermayer" <michael at niedermayer.cc>
> > wrote:
> > >
> > > TODO: version bump, docs
> > >
> > > Signed-off-by: Michael Niedermayer <michael at niedermayer.cc>
> > > ---
> > >  libavformat/avformat.h      | 8 ++++++++
> > >  libavformat/options_table.h | 1 +
> > >  libavformat/utils.c         | 3 +++
> > >  3 files changed, 12 insertions(+)
> > >
> > > diff --git a/libavformat/avformat.h b/libavformat/avformat.h
> > > index 3ee7051..33b921a 100644
> > > --- a/libavformat/avformat.h
> > > +++ b/libavformat/avformat.h
> > > @@ -1886,6 +1886,14 @@ typedef struct AVFormatContext {
> > >       * - decoding: set by user through AVOptions (NO direct access)
> > >       */
> > >      char *protocol_blacklist;
> > > +
> > > +    /**
> > > +     * Force parsing.
> > > +     * - encoding: unused
> > > +     * - decoding: set by user through AVOptions (NO direct access)
> > > +     */
> > > +    int force_parsing;
> > > +
> > >  } AVFormatContext;
> > >
> > >  int av_format_get_probe_score(const AVFormatContext *s);
> > > diff --git a/libavformat/options_table.h b/libavformat/options_table.h
> > > index 3b74d1b..359796c 100644
> > > --- a/libavformat/options_table.h
> > > +++ b/libavformat/options_table.h
> > > @@ -103,6 +103,7 @@ static const AVOption avformat_options[] = {
> > >  {"format_whitelist", "List of demuxers that are allowed to be used",
> > OFFSET(format_whitelist), AV_OPT_TYPE_STRING, { .str = NULL },
 CHAR_MIN,
> > CHAR_MAX, D },
> > >  {"protocol_whitelist", "List of protocols that are allowed to be
used",
> > OFFSET(protocol_whitelist), AV_OPT_TYPE_STRING, { .str = NULL },
 CHAR_MIN,
> > CHAR_MAX, D },
> > >  {"protocol_blacklist", "List of protocols that are not allowed to be
> > used", OFFSET(protocol_blacklist), AV_OPT_TYPE_STRING, { .str = NULL },
> > CHAR_MIN, CHAR_MAX, D },
> > > +{"forceparsing", "force use of AVParsers", OFFSET(force_parsing),
> > AV_OPT_TYPE_INT, { .i64 = -1 },  -1, AVSTREAM_PARSE_FULL_ONCE, D },
> > >  {NULL},
> > >  };
> >
> > Why int?
>
> what else ?
> a flag doesnt work as AVStreamParseType has more than 1 value

An enum would be nice.

> > At vdd, we discussed parsers, maybe we should sync on that because I
> > believe it makes this unnecessary.
>
> not sure i understand what you suggest ?

Anton expressed interest in merging parsers and bitstream filters. I
believe they are (other than semantically) identical in goal, and the
semantic difference is sometimes violated already. If you think of them as
bitstream filters, we could merge their functionality with auto-insertion
of BSF. It was proposed that we could go one step further and do parsing
not in the demuxer step (av_read_frame), but before the decoder (which is a
change) and in the muxer (which we already do). Auto-insertion mechanisms
would be shared and this would lead to interesting new features that
someone demuxing a mpeg system stream or vp9/divx-with-packed-b-frames
using external tools (which don't split/reframe) but using avcodec decoders
would still see it work. On the other hand, people using lavf to demux such
streams would not need to merge (divx/vp9) before muxing, making that
process more efficient (using lavf) or less buggy (using non-lavf muxers).

(Some people also proposed that maybe reframing - like in mpeg elementary
streams - and splitting/merging - like in packed-divx or vp9 - are perhaps
different enough that they should happen in different places, although I
believe majority agreed it could stay in the same place.)

If you follow this, forceparsing wouldn't be necessary and you could use
existing bitstream CLI options and API to achieve (on the input side) what
parsers do right now.

Ronald