501289 - Really fast videos (393 frames/millisecond) are exciting to watch...

Status: VERIFIED FIXED

(Core :: Audio/Video, defect)

Description

[Matthew Gregan [:kinetik]]

Attachment: testcase

...but livelock the browser.

715	            mon.Wait(PR_MillisecondsToInterval(PRInt64(mCallbackPeriod*1000)));
(gdb) p mCallbackPeriod
$11 = 2.54312499e-06
(gdb) p mCallbackPeriod*1000
$12 = 0.002543124992371304
(gdb) p (long long)(mCallbackPeriod*1000)
$13 = 0

Comment 1

[Matthew Gregan [:kinetik]]

Forgot to note: this could be a dupe of bug 501285, depending on how we decide to fix it.

Comment 2

[Matthew Gregan [:kinetik]]

So what happens is the main thread is waiting on the decoder lock (e.g. to call NotifyDownloadEnded), and the decoder never drops the lock because the wait interval is 0.

Comment 3

[Matthew Gregan [:kinetik]]

Attachment: testcase 2

I wasn't sure if it was worth filing another bug for this testcase, so I'm erring on the side of not, since it's pretty similar.

We end up with an extremely small mCallbackPeriod value, but this time it's caused by the sample rate of the Vorbis track being 184MHz.

Comment 4

[Matthew Gregan [:kinetik]]

Mark Banner sent me a private video showing that this bug happens in the wild.  With a video produced by ffmpeg2theora 0.25, a video ended up with a framerate of 15750 FPS.  Hopefully Mark will add a few details on how this video was produced, as it's likely we'll run into this problem again in the future.

When we eventually fix this bug, it'll either cause the video to be rejected, or to play it at some "sane" default framerate.

Comment 5

[Chris Pearce [:cpearce (Not reading bugmail)]]

Matthew: do you still have the 15750 fps video you mention in comment 4? Does keyframe skipping come to our rescue here?

Comment 6

[Matthew Gregan [:kinetik]]

No, but there's an equivalent testcases attached to this bug.

Neither of the attached testcases cause problems with the current code as far as I can tell, so this bug is effectively fixed.

Comment 7

[u279076]

I'm not sure what exactly the problem is I should be looking for to verify this fixed.  Can someone please explain the issue in better detail?  Thanks.

Comment 8

[Matthew Gregan [:kinetik]]

If the two attached testcases don't hang the browser, we're treating the bug as fixed.

Comment 9

[u279076]

(In reply to comment #8)
> If the two attached testcases don't hang the browser, we're treating the bug as
> fixed.

Thanks.