[FFmpeg-devel] [PATCH] AAC decoder channel layout support

[Reimar Döffinger]

On Fri, Oct 16, 2009 at 12:15:47PM -0400, Jason Millard wrote:
> >
> > So the demuxer is borked? It's the demuxer's job to pass full frames
> > to the decoder.
> >
> >
> I think it may be. I finally had a chance to do more testing.
> 
> I removed my 0xfff bypass patch and curl'd some streams until I finally ran
> into the situation.
> 
> This stream does not begin with a sync FFF and will not play.
> 
> FFplay version SVN-r20209-snapshot, Copyright (c) 2003-2009 Fabrice Bellard,
> et al.
>   built on Oct 16 2009 12:00:33 with gcc 4.0.1 (Apple Inc. build 5493)
>   configuration:
>   libavutil     50. 3. 0 / 50. 3. 0
>   libavcodec    52.36. 0 / 52.36. 0
>   libavformat   52.39. 0 / 52.39. 0
>   libavdevice   52. 2. 0 / 52. 2. 0
>   libswscale     0. 7. 1 /  0. 7. 1
> [aac @ 0x1039e00]max_analyze_duration reached
> Input #0, aac, from '/Users/jsm174/Desktop/streams/stream':
>   Duration: 00:00:15.38, bitrate: 61 kb/s
>     Stream #0.0: Audio: aac, 22050 Hz, 2 channels, s16, 61 kb/s
> [aac @ 0x103ae00]aac_decode_frame() - enter - show_bits: 028B
> [aac @ 0x103ae00]00000000  28 b2 bf e9 5d 7f f4 b5 cf ea de df e9 89 a1 df
> (...]...........
> [aac @ 0x103ae00]00000010  af fe d5 47 2f ef d7 94 c3 bd dd f3
> ...G/.......
> [aac @ 0x103ae00]channel element 1.4 is not allocated
> [aac @ 0x103ae00]aac_decode_frame() - enter - show_bits: 0FFF

I _think_ I have seen this issue before, demuxer passes and
incomplete/invalid first frame on to the decoder (which is fine so far),
but the decoder parses out some invalid stream properties out of that
broken frame and then sticks to them even if a valid frame comes up
later.
Which means that the first frame passed to the decoder must be "100%"
valid or it will decode nothing at all.
(note that in the details my description probably isn't right, but for
what I can tell that seemed to be the issue in a general sense as far as
I could figure out).