I attached the file via the separate attachment page (if you try to attach when writing up a bug, it says limited to 500K, and IIRC, it doesn't give a progress bar for feedback).
Known issue in the mp2v decoder included in both libav and ffmpeg and it's still not fixed since 8 months despite reported repeatedly.
Sorry, it appears we forgot about the issue: Could you point me to the reports?
{{{
(lldb) bt
thread #44 (closed): tid = 0x592249, 0x00326920 VLC for iOS`mpeg_decode_slice [inlined] mpeg2_fast_decode_block_non_intra + 461 at mpeg12dec.c:453, stop reason = EXC_BAD_ACCESS (code=1, address=0x1240601b)
}}}
Thank you for the backtrace, I didn't know if my guess from ticket #9713 (closed) was correct or not (and I couldn't test)! Please note that even with the current fixes, CODEC_FLAG2_FAST may still crash for invalid input streams, so I strongly suggest not to enable it by default (no matter the platform).
It's fixed in both libav and FFmpeg using different approaches now.
Thanks for coming back to me. Disabling CODEC_FLAG2_FAST is not an option, sorry, as we need every bit of speed possible on mobile devices and it's probably a good idea to resolve those crashes one by one. Now that we have sufficient crash reporting facilities in place, this will be quite fast.
Thanks for coming back to me. Disabling CODEC_FLAG2_FAST is not an option, sorry, as we need every bit of speed possible on mobile devices and it's probably a good idea to resolve those crashes one by one.
I believe you misunderstand: The fast option avoids some checks to produce faster decoding assuming that the input stream is valid, if these checks are added, the gain (which is very useful for users who know their input) would be lost. In this specific case, Michael found a (faster) work-around but there are cases for which the fast option leads to invalid memory accesses.
Now that we have sufficient crash reporting facilities in place, this will be quite fast.
These crashes have been reported very, very often in the past, and many patches were sent that were supposed to "fix" them, this is why the code now contains comments to explain the usage of the functions.
To avoid such delays in the future: Could you point me to your original report?