SVN 11084: playback of edited MPEG4 recording consumes all memory and crashes on cutlist point

[oa@…]

A recording made from a DVB source, auto-transcoded to MPEG4, cutlist edited and lossless transcoded to remove commercials halts playback on each cutpoint. The playback goes into a busy loop which allocates all memory and finally crashes to out of memory. MPlayer is able to play the same file, and so is mythfrontend if I manage to skip past the cut point with fast forward.

[oa@…]

log of playback (this over the network/nfs, but local playback exhibits same problem)

[oa@…]

backtrace of the process (while suspended)

[oa@…]

experimenting with several files, it seems this problem might be related to aspect ratio changes in the original stream when entering the commercial break, or otherwise "unclean" frames in the stream at the transition point. However, I have not (yet) noticed these causing the described playback problem prior to the file being transcoded.

[Isaac Richards]

I need a sample video to fix this.

[oa@…]

Hmm. I've been trying to cut a manageable piece out of a recording that exhibits this, but mythtranscode crashes to a seg fault as well. I can put a full recording available if that is acceptable (around 770 megs), or try to catch some program during a commercial break, but the latter will take some more time.

[anonymous]

fwiw, you can grab the full recording at ​http://htpc.home.fishpool.org/tmp/crash-at-0:14:36.nuv -- and like the name says, it crashes at 14 minutes, 36 seconds into the recording.

[oa@…]

Please take a look at #2360 - while the symptoms are different, l have a feeling these are closely related

[cpinkham]

I believe this was fixed in [11314]. It sounds like the same exact symptoms especially the comment about this being related to aspect ratio changes.

[oa@…]

I'm afraid my test clip still crashes at the same point. However, I applied [11314] to 0.20-fixes, so if there's some other changeset in the trunk that affects this, that might explain it. The patch applied cleanly, though.

[cpinkham]

I believe this input file was corrupted by the bug in the transcoder that was fixed in [13647]. When I run this file through nuvscan, the info contained in the fileheader at the resolution change is corrupt. This is what is crashing the players now. I just checked and this fix was never copied over to the -fixes branch, so I have just merged [13647] into 0.20-fixes as [14548]. I'm going to close this ticket for now unless someone can reproduce this issue with current SVN trunk or -fixes version [14548].