[FFmpeg-devel] [PATCH] seeking in asf files where audio stream is found before video

[Michael Niedermayer]

Hi

On Sun, Jul 08, 2007 at 12:00:10AM +0200, elupus wrote:
> Hi,
> 
> Currently asf demuxer reads the index provided in the file on first seek 
> and adds it to the stream that the seek occured in. Any seeks on other 
> streams in the file will then not use the provided index.
> 
> This turns out to be an issue since av_find_stream_info will seek to pts 0 
> for all available streams, if audio is found first, then index get's loaded 
> on the audio stream. When a seek is tried later, the default stream seeked 
> is the video stream wich doesn't have an index loaded.
> 
> If an asf stream doesn't have a index, it defaults to binary seek, wich 
> will then add some index points to the stream, on the next seek those will 
> be used without interpolation.
> 
> This patch changes the demuxer to only load index for the video stream, and 

this is just wrong


> fall back to the general seeks in utils.c if its internal seek doesn't work.

good but why does this make a difference? it shouldnt, its a welcome cleanup
but i dont see how this would change anything


> 
> Sadly seem to be a bug in av_gen_search, wich happens to search the whole 
> damn file when av_find_stream_info tries to seeks to pts 0, so there is 
> also a hack in this patch to handle seeks to time 0.
> 
> On another note, shouldn't the demuxer set AVFormatContext->data_offset 
> instead of just setting data_offset in ASFContext? If so i can provide a 
> patch for that.

maybe


[...]

> Index: asf.c
> ===================================================================
> --- asf.c	(revision 9451)
> +++ asf.c	(working copy)
> @@ -989,6 +989,9 @@
>          }
>          asf->index_read= 1;
>      }
> +    else
> +        asf->index_read= -1;
> +
>      url_fseek(&s->pb, current_pos, SEEK_SET);
>  }

ok but this belongs into a seperate patch with the related changes and
the } and else should be on the same line



> @@ -999,15 +1002,24 @@
>      int64_t pos;
>      int index;
>  
> +    if (pts == 0) {
> +      // this is a hack since av_gen_search searches the entire file in this case
> +      av_log(s, AV_LOG_DEBUG, "SEEKTO: %"PRId64"\n", pos);
> +      url_fseek(&s->pb, s->data_offset, SEEK_SET);      
> +      return 0;
> +    }
> +

indention is wrong and there is trailing whitespace


>      if (asf->packet_size <= 0)
>          return -1;
>  
> +    if (st->codec->codec_type != CODEC_TYPE_VIDEO)
> +        return -1;
> +

why should we fail on all non video streams this is just a ugly
hack

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

Complexity theory is the science of finding the exact solution to an
approximation. Benchmarking OTOH is finding an approximation of the exact
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20070714/c53c0b3e/attachment.pgp>