Use av_rescale_q() for converting FFmpeg timestamps to base::TimeDelta (second attempt).

Use av_rescale_q() for converting FFmpeg timestamps to base::TimeDelta (second attempt).

Previously we were using integer math to convert to microseconds, but depending on the frame rate and packet size we could introduce enough error that could accumulate and introduce audio/video synchronization drift.  av_rescale_q() is a simple function but is designed to minimize error as much as possible.

Second attempt since I forgot to manually resolve av_rescale_q() for unittests.

[scherkus (not reviewing), 2009-05-20 06:08:02]

Windows was resolving my delay loaded symbols hence I didn't catch the break
locally.  Try servers don't work with this patch either :\

Anyway, the solution is to implement a cheap version of av_rescale_q() manually
in ffmpeg_demuxer_unittest.cc

[awong, 2009-05-20 18:17:42]

lgtm

http://codereview.chromium.org/113619/diff/1/7
File media/filters/ffmpeg_demuxer_unittest.cc (right):

http://codereview.chromium.org/113619/diff/1/7#newcode137
Line 137: int64 av_rescale_q(int64 a, AVRational bq, AVRational cq) {
Hah...cute.  We do need a way to better mock these things. :-/

[scherkus (not reviewing), 2009-05-20 18:21:27]

http://codereview.chromium.org/113619/diff/1/7
File media/filters/ffmpeg_demuxer_unittest.cc (right):

http://codereview.chromium.org/113619/diff/1/7#newcode137
Line 137: int64 av_rescale_q(int64 a, AVRational bq, AVRational cq) {
On 2009/05/20 18:17:42, awong wrote:
> Hah...cute.  We do need a way to better mock these things. :-/

You should see the actual implementation.  It's an explosion of shifts and bit
mask operators!!

This implementation will easily overflow hehe