ffmpeg2theora

[PATCH] ffmpeg2theora segfaults when resampling audio or converting VP9 input

Bug #1483861 reported by Brion Vibber on 2015-08-11

This bug affects 2 people

Affects		Status	Importance	Assigned to	Milestone
	ffmpeg2theora	New	Undecided	Unassigned

Bug Description

In upgrading our video transcode servers at Wikimedia, we found that the current versions of ffmpeg2theora are kinda broken with current ffmpeg.

Downstream bug notes:
* https://phabricator.wikimedia.org/T69953
* https://phabricator.wikimedia.org/T55863#1498329
* https://phabricator.wikimedia.org/T103335#1526759

Patch against current master attached, seems to get it working in my dev environment (MediaWiki-Vagrant set up with Ubuntu Trusty plus ffmpeg backported from Debian).

Revision history for this message

Brion Vibber (brion) wrote on 2015-08-11:

0001-Fix-regressions-segfaults-during-after-conversion.patch Edit (2.2 KiB, text/plain)

Revision history for this message

Steffen Pankratz (kratz00) wrote on 2016-10-19:

stacktrace_1483861.txt Edit (10.0 KiB, text/plain)

ffmpeg2theora 0.30 and even the latest git revision (49dea2a077f62925b62636758cc6c1f565b77535) suffer from this bug. For me this bug is triggered by adding the "-H" parameter, as I wanted to re-sample the audio to 44100 Hz. See stacktrace_1483861.txt (This is a debug build of the mentioned git revision).

The attached patch by Brion still applies against the mentioned git revision:

patch -p1 -i ../0001-Fix-regressions-segfaults-during-after-conversion.patch
patching file src/ffmpeg2theora.c
Hunk #1 succeeded at 611 (offset 68 lines).
Hunk #2 succeeded at 1065 (offset 68 lines).
Hunk #3 succeeded at 1896 (offset 70 lines).

With Brion's patch ffmpeg2theora does not crash anymore and the audio is re-sampled to 44100 Hz.

Is there any reason this patch was not accepted yet?