fix parsing error at the end of stream

fix parsing error at the end of stream

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=48988

[wjia(left Chromium), 2010-06-04 05:29:28]

When demuxer hits the end of stream and has some pending read requests, it
creates EOS buffer to feed decoder. But it should bypass bitstream_converter.

[jiesun, 2010-06-04 15:35:30]

thanks for doing this.
I think the bitstream are set at initialize time. if resetted, and people want
to replay the video, it "may" not call init task, just seek task. therefore I
think the safe way to do it is to exclude the EOS packet from converter. how do
you think?

[wjia(left Chromium), 2010-06-04 16:39:26]

On 2010/06/04 15:35:30, jiesun wrote:
> thanks for doing this.
> I think the bitstream are set at initialize time. if resetted, and people want
> to replay the video, it "may" not call init task, just seek task. therefore I
> think the safe way to do it is to exclude the EOS packet from converter. how
do
> you think?

I think replaying video requires re-init filters since some of them have state
variable indicating that EOS has been received and will not processing incoming
buffers any more.

If we can guarantee that packet->data==NULL indicates end of stream, then the
better fix is to bypass bitstream_converter inside
FFmpegDemuxerStream::EnqueuePacket.

[scherkus (not reviewing), 2010-06-04 19:03:25]

http://codereview.chromium.org/2635002/diff/3001/4001
File media/filters/ffmpeg_demuxer.cc (right):

http://codereview.chromium.org/2635002/diff/3001/4001#newcode121
media/filters/ffmpeg_demuxer.cc:121:
!bitstream_converter_->ConvertPacket(packet)) {
I would check for an empty packet here as EOS is a recoverable state

destroying the bitstream filter would be bad :)

[wjia(left Chromium), 2010-06-04 20:10:11]

http://codereview.chromium.org/2635002/diff/3001/4001
File media/filters/ffmpeg_demuxer.cc (right):

http://codereview.chromium.org/2635002/diff/3001/4001#newcode121
media/filters/ffmpeg_demuxer.cc:121:
!bitstream_converter_->ConvertPacket(packet)) {
On 2010/06/04 19:03:25, scherkus wrote:
> I would check for an empty packet here as EOS is a recoverable state
> 
> destroying the bitstream filter would be bad :)

Done.

[jiesun, 2010-06-04 20:14:50]

LGTM. thx

[scherkus (not reviewing), 2010-06-04 20:15:58]

LGTM