[FFmpeg-devel] [PATCH] mpeg12dec: don't assert on unknown chroma format

[Hendrik Leppkes]

On Wed, Sep 30, 2015 at 6:03 PM, Paul B Mahol <onemda at gmail.com> wrote:
> Dana 30. 9. 2015. 17:19 osoba "Hendrik Leppkes" <h.leppkes at gmail.com>
> napisala je:
>>
>> On Wed, Sep 30, 2015 at 4:35 PM, Paul B Mahol <onemda at gmail.com> wrote:
>> > On 9/30/15, Hendrik Leppkes <h.leppkes at gmail.com> wrote:
>> >> On Wed, Sep 30, 2015 at 2:51 PM, Michael Niedermayer <michaelni at gmx.at>
>> >> wrote:
>> >>> On Wed, Sep 30, 2015 at 02:39:21PM +0200, Hendrik Leppkes wrote:
>> >>>> On Wed, Sep 30, 2015 at 2:33 PM, Michael Niedermayer <
> michaelni at gmx.at>
>> >>>> wrote:
>> >>>> > On Wed, Sep 30, 2015 at 02:29:43PM +0200, Hendrik Leppkes wrote:
>> >>>> >> On Wed, Sep 30, 2015 at 2:10 PM, Michael Niedermayer
>> >>>> >> <michaelni at gmx.at> wrote:
>> >>>> >> > On Wed, Sep 30, 2015 at 12:42:59PM +0200, Hendrik Leppkes wrote:
>> >>>> >> >> The chroma format can be still unset in postinit when a badly
> cut
>> >>>> >> >> stream
>> >>>> >> >> starts with a slice instead of a sequence header. This is a
> common
>> >>>> >> >> occurance when feeding avcodec from a Live TV stream.
>> >>>> >> >> ---
>> >>>> >> >>  libavcodec/mpeg12dec.c | 1 -
>> >>>> >> >>  1 file changed, 1 deletion(-)
>> >>>> >> >>
>> >>>> >> >> diff --git a/libavcodec/mpeg12dec.c b/libavcodec/mpeg12dec.c
>> >>>> >> >> index 5d916d1..b3c2c45 100644
>> >>>> >> >> --- a/libavcodec/mpeg12dec.c
>> >>>> >> >> +++ b/libavcodec/mpeg12dec.c
>> >>>> >> >> @@ -1389,7 +1389,6 @@ static int
>> >>>> >> >> mpeg_decode_postinit(AVCodecContext *avctx)
>> >>>> >> >>              case 1: avctx->chroma_sample_location =
>> >>>> >> >> AVCHROMA_LOC_LEFT; break;
>> >>>> >> >>              case 2:
>> >>>> >> >>              case 3: avctx->chroma_sample_location =
>> >>>> >> >> AVCHROMA_LOC_TOPLEFT; break;
>> >>>> >> >> -            default: av_assert0(0);
>> >>>> >> >>              }
>> >>>> >> >>          } // MPEG-2
>> >>>> >> >
>> >>>> >> > This assert double checks that the context init which uses
>> >>>> >> > width/height/chroma format is done after the chroma format and
> w/h
>> >>>> >> > has
>> >>>> >> > been read from the headers
>> >>>> >> > if this is reached without the headers being read then the code
> is
>> >>>> >> > buggy and removing the assert will not fix that bug
>> >>>> >> >
>> >>>> >> > also if there is no sequence header how is width/height set ?
>> >>>> >> > theres a check for width/height before mpeg_decode_postinit is
>> >>>> >> > called
>> >>>> >> >
>> >>>> >>
>> >>>> >> The dimensions are set by the caller in the avctx before opening
> the
>> >>>> >> codec, which apparently inits the mpeg context as well.
>> >>>> >>
>> >>>> >> Feel free to fix it differently, error out or something, but I can
>> >>>> >> trip an assert with input data, which should never happen.
>> >>>> >
>> >>>> > how can this be reproduced ?
>> >>>> >
>> >>>>
>> >>>> I can only trigger it with API usage, not through the CLI tools, so I
>> >>>> cannot produce a test case.
>> >>>
>> >>> is this with a unmodified up to date ffmpeg ?
>> >>> i fixed this assert failing a while ago
>> >>> (b54e03c9dc2a05324c08b503bfe7535c49c0f281)
>> >>>
>> >>>
>> >>
>> >> Its up to date, but slightly modified - but nothing pertinent to the
>> >> mpeg2 decoder as far as I know.
>> >> I'll just add this to the list of modifications then. I can clearly
>> >> trigger this assert in a release build (which is hilarious on its own,
>> >> asserts are a debugging tool), which crashes the application, so thats
>> >> no good.
>> >
>> > WTF, can you just give sample/link something?
>>
>> Like I said earlier in the thread, it happens with API usage in a Live
>> TV scenario. I cannot produce a sample of that.
>
> Not even backtrace?
>

postinit is only called from one place, the trace isn't very interesting.

mpeg_decode_postinit
decode_chunks
mpeg_decode_frame
avcodec_decode_video2