[FFmpeg-devel] [PATCH] attachments support in matroska demuxer

[Aurelien Jacobs]

On Sun, 20 Jan 2008 02:56:20 +0100
Michael Niedermayer <michaelni at gmx.at> wrote:

> On Thu, Jan 17, 2008 at 01:00:26AM +0100, Aurelien Jacobs wrote:
> > Evgeniy Stepanov wrote:
> > 
> > > On Thursday 03 January 2008 04:23:09 Michael Niedermayer wrote:
> > > > On Thu, Jan 03, 2008 at 01:38:39AM +0100, Aurelien Jacobs wrote:
> > > > > On Sat, 29 Dec 2007 22:38:24 +0300
> > > > > Evgeniy Stepanov <eugeni.stepanov at gmail.com> wrote:
> > > >
> > > > [...]
> > > >
> > > > > Conclusions:
> > > > > - Attachment really looks like a stream itself. You can select
> > > > >   the one you want to remux. For example the film poster would
> > > > >   just be a jpeg stream with a single picture. The specificity
> > > > >   of those streams would be:
> > > > >
> > > > >     * contain one single packet
> > > > >     * not pts
> > > >
> > > > Contain no packets, just one global extradata, this was proposed
> > > > already. If you use normal packets there will be problems with them
> > > > disapearing if a stream is split timewise or if you seek before
> > > > demuxing the first packet.
> > > > Iam not completely happy with this use of streams but we will need
> > > > some solution and maybe its the least ugly, we will see ...
> > > >
> > > > at least iam pretty sure extradata to be a better choice than 1
> > > > normal packet
> > > >
> > > > >     * has a filename (important so that an ASS decoder can
> > > > >       select the proper ttf based on it's name)
> > > > >
> > > > >     * always demuxed before other (normal) streams (IIRC)
> > > >
> > > > extradata should be available before packets ...
> > > >
> > > > >   It seems to me that attachment would fit pretty well, each one
> > > > >   in its own AVStream. Does this sound reasonable ?
> > > > > - Here is how remuxing to other formats (avi, mov...) could work:
> > > > >     * supported codec (jpeg poster) could be muxed as a normal
> > > > >       JPEG stream with only one picture
> > > > >     * ttf streams should be written in a separate file (that's
> > > > >       the way various players expect to find ttf fonts when
> > > > >       demuxing AVI, IIRC)
> > > > >
> > > > > - I think the only required API modifications to support this
> > > > >   would be:
> > > > >     * adding AVStream.filename and maybe
> > > > > AVStream.id_of_link_stream.
> > > > >     * adding CODEC_ID_TTF ? (ugly, but shouldn't cause much
> > > > > trouble)
> > > > >     * adding a way to mux ttf font in a separate file (maybe use
> > > > >       a separate muxer for this, ie. call 2 different muxers from
> > > > >       ffmpeg. seems ugly, but I've no better idea right now).
> > > >
> > > > Well, there are a few more things i think
> > > > We need name / (mime) type to remux attachments
> > > > (CD Cover / ? / tiff)
> > > > (Fan art / image/x-photoshop / psd)
> > > > There also could be other things like Author, ...
> > > >
> > > > Also we need a way to get the attachments to the decoders, yes we
> > > > dont have a ASS/SSA decoder but lets assume we had.
> > > >
> > > > And while we can just store the streams in avi/mov, where does the
> > > > filename/name/type go? if its dropped you could end up with a huge
> > > > number of images and not know what each is.
> > > >
> > > > > What do you think about this proposal ? Does it sounds like a
> > > > > reasonable base ?
> > > > >
> > > > > If it's not, the only viable alternative seems to be
> > > > > AVFormatContext.attachments just like in the original patch,
> > > > > but with some additionnal code to allow "remuxing", etc...
> > > >
> > > > yes, these seem to be the 2 options we have
> > > 
> > > This patch implements attachments as separate streams with extradata.
> > 
> > OK. Here are things issues in this patch:
> >  - CODEC_TYPE_NB must stay at the end of the enum.
> >  - Adding attachments to AVFormatContext is probably a leftover from
> >    your previous patch.
> >  - I don't like the fake MATROSKA_TRACK_TYPE_ATTACHMENT and everything
> >    depending on it.
> >  - I don't think CODEC_ID_ATTACHMENT is of any use if it's used for
> >    every kind of attachement. CODEC_TYPE_ATTACHMENT already gives the
> >    same information.
> > 
> > I fixed all those issues in the attached patch.
> > 
> > Moreover, I tried to associate a specific CODEC_ID_* to different kind
> > of attachements. To me, this seems like a good idea, but I would like
> > to know what others think about it ?
> > 
> [...]
> 
> > Index: libavformat/avformat.h
> > ===================================================================
> > --- libavformat/avformat.h	(revision 11544)
> > +++ libavformat/avformat.h	(working copy)
> > @@ -21,8 +21,8 @@
> >  #ifndef FFMPEG_AVFORMAT_H
> >  #define FFMPEG_AVFORMAT_H
> >  
> > -#define LIBAVFORMAT_VERSION_INT ((52<<16)+(4<<8)+0)
> > -#define LIBAVFORMAT_VERSION     52.4.0
> > +#define LIBAVFORMAT_VERSION_INT ((52<<16)+(5<<8)+0)
> > +#define LIBAVFORMAT_VERSION     52.5.0
> >  #define LIBAVFORMAT_BUILD       LIBAVFORMAT_VERSION_INT
> >  
> >  #define LIBAVFORMAT_IDENT       "Lavf" AV_STRINGIFY(LIBAVFORMAT_VERSION)
> > @@ -347,6 +347,9 @@
> >  
> >  #define MAX_REORDER_DELAY 4
> >      int64_t pts_buffer[MAX_REORDER_DELAY+1];
> > +
> > +    char *filename;
> 
> > +    char *type;
> 
> iam against some undocumented char *type
> please document precissely what it repressents and how it differs from
> codec_tag, stream_codec_tag and codec_id
> or even better get rid of it and use codec_tag or explain why the type
> here should be special cased relative to these funny codec id strings in
> matroska

Currently, *type stores the standard mime type of the attachment.
It is only useful to stream-copy from mkv to mkv, preserving attachments
which have a mime-type not listed in ff_mkv_mime_tags[].
So I guess it's not really useful as long as ff_mkv_mime_tags[] is
reasonably complete (which it is not, right now, but I will improve it).
I will remove it for now.

> filename should be documented as well, and in stream type independant way like
> the "source filename" of the stream or so

Sure. I will add doxy comments in next version of this patch (probably
tomorrow).

Aurel