[FFmpeg-devel] asf.c: fix for "random" ftell() reponse

[Ronald S. Bultje]

Hi,

On Sun, Dec 14, 2008 at 2:36 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Sun, Dec 14, 2008 at 02:08:22PM -0500, Ronald S. Bultje wrote:
>> for RTSP-MS, each packet has its own ByteIOContext, and thus the
>> ftell() returned by it is invalid compared to the data read so far
>> (because it resets to zero at each new ASF packet). This means that
>
> and why does it do that?

For RTSP, each RTP (or RDT) packet contains a single ASF (or RM, or
whatever) packet. I send those to the demuxer using a local
ByteIOContext variable (since RTSP doesn't use buffered I/O and thus
has no ByteIOContext). I guess in a way it is also extra safe since an
error in the demuxer wouldn't screw up the RTSP demuxer state, but
anyway, that's debatable. The point is that I have a local
ByteIOContext which I init using the RTP/RDT packet data. This means
that after a header of 5kb and 2 packets of 2kb each, the demuxer
would expect a value of 9kb for url_ftell(), but it will return 2kb at
the end of the second packet and 0kb at the beginning of the third. I
can fix that by manually setting ByteIOContext->pos to whatever value
the demuxer wants, but it seemed kind of hackish to me (although it
would solve the problem also).

Of course, if you can think of a better way to solve the fact that
off=0 (rather than >=3), I'm open to any such solution.

Ronald