[mythtv] New MPEG2 commercial-cut code ready for testing

[Geoffrey Hausheer]

On 11/16/05, Adam Egger wrote:
>
> STATE: 4 (b5e5a082) W: 720 H: 576 P: 1080000 4:24:30.6 T:2 F: 2
> STATE: 0 (8374595) W: 720 H: 576 P: 1080000 4:24:29.19 T:2 F: 2
> STATE: 0 (8543db2) W: 720 H: 576 P: 1080000 4:24:29.19 T:0 F: 2
> STATE: 0 (825214a) W: 720 H: 576 P: 1080000 4:24:29.19 T:1 F: 2
> STATE: 0 (82ee515) W: 720 H: 576 P: 1080000 4:24:29.19 T:5 F: 2
> STATE: 0 (850cb92) W: 720 H: 576 P: 1080000 4:24:29.19 T:3 F: 2
> STATE: 0 (842a4ea) W: 720 H: 576 P: 1080000 4:24:29.19 T:4 F: 2
> STATE: 0 (83b1255) W: 720 H: 576 P: 1080000 4:24:29.19 T:8 F: 2
> STATE: 0 (83310b2) W: 720 H: 576 P: 1080000 4:24:29.19 T:6 F: 2
> STATE: 0 (837d652) W: 720 H: 576 P: 1080000 4:24:29.19 T:7 F: 2
> STATE: 0 (829f3ad) W: 720 H: 576 P: 1080000 4:24:29.19 T:11 F: 2
> STATE: 0 (846fef2) W: 720 H: 576 P: 1080000 4:24:29.19 T:9 F: 2
>
> *** glibc detected *** double free or corruption (out): 0x083a1470 ***
>
> Program received signal SIGABRT, Aborted.
> [Switching to Thread -1225319584 (LWP 28235)]
> 0xb75047a7 in raise () from /lib/tls/libc.so.6
> (gdb) bt
> #0  0xb75047a7 in raise () from /lib/tls/libc.so.6
> #1  0xb750604b in abort () from /lib/tls/libc.so.6
> #2  0xb753b005 in __fsetlocking () from /lib/tls/libc.so.6
> #3  0xb7541657 in malloc_usable_size () from /lib/tls/libc.so.6
> #4  0xb7541af2 in free () from /lib/tls/libc.so.6
> #5  0x0804c9ea in MPEG2fixup::build_frame (this=0xbfea3bc0,
> pkt=0xbfea378c, fname=0xbfea37c0 "cnv0.yuv")
>     at mpeg2fix.cpp:956
> #6  0x0804d105 in MPEG2fixup::convert_to_i (this=0xbfea3bc0,
> frameNum=9, numFrames=3, headPos=9) at mpeg2fix.cpp:1222
> #7  0x080507d0 in MPEG2fixup::start (this=0xbfea3bc0) at mpeg2fix.cpp:1494
> #8  0x08051bfd in main (argc=14, argv=0xbfea4f34) at mpeg2fix.cpp:1806

This is odd.  1st I really need to see your full command line, as you
appear to be applying a commercial-cut here, but I don't know what you
actually chose.

2nd, while this debug is more interesting, it isn't very enlightening.
 There is really no way that I can see to get a double free on the
codecContext (the free atline 956), which implies, perhaps, a stack
overrun.  I wonder if this is a segfault from libavcodec similar to
what Bryan saw.  What version of SVN are you running?
You could try setting a break point at line 939 (at the
avcodec_encode_video call) and doing something like this before and
after the call:
print c
print outbuf
print picture
print out_size
print outbuf_size
print *c
print *picture

if any of the 1st 3 are different (the 0x????????, not any data
afterwards), that is definitely stack corruption.  The 6th and 7th
commands will dump out a lot of text.  I'm not particularly interested
in the values, but you should spot-check that there are no whacky
numbers.
after that step through and ensure that it still dies on the 'free(c)'

You could also try disabling all SSE, SSE2, etc by adding something
like this at line
938 (before the encode_video call):
c->dsp_mask = 0xffff;
recompiling and rerun.

Sorry I can't be of more help than that, but maybe something above
will help us make some progress.

.Geoff