]> git.sesse.net Git - ffmpeg/commit
avformat/mp3dec: improve junk skipping heuristic
authorwm4 <nfxjfg@googlemail.com>
Tue, 20 Oct 2015 10:17:21 +0000 (12:17 +0200)
committerwm4 <nfxjfg@googlemail.com>
Tue, 20 Oct 2015 10:17:21 +0000 (12:17 +0200)
commitde1b1a7da9e6ddf42447271e519099a88b389e4a
tree9e7e04924ab14e733f66fd729a05dc451f0ee840
parent35564318ad5a679afed14a1c88b517aa2d21298d
avformat/mp3dec: improve junk skipping heuristic

Commit 2b3e9bbfb529e6bde238aeb511b55ebe461664c8 caused problems for a
certain API user:

https://code.google.com/p/chromium/issues/detail?id=537725
https://code.google.com/p/chromium/issues/detail?id=542032

The problem seems rather arbitrary, because if there's junk, anything
can happen. In this case, the imperfect junk skipping just caused it to
read different junk, from what I can see.

We can improve the accuracy of junk detection by a lot by checking if 2
consecutive frames use the same configuration. While in theory it might
be completely fine for the 1st frame to have a different format than the
2nd frame, it's exceedingly unlikely, and I can't think of a legitimate
use-case.

This is approximately the same mpg123 does for junk skipping. The
set of compared header bits is the same as the libavcodec mp3 parser
uses for similar purposes.
libavformat/mp3dec.c