Created attachment 81788 [details] [review]
Possible fix

Prevents segfault by providing a "better" default for intra_dc_precision in 2 ways:
* as a safe default in any case
* by including mpeg1/2video in the family of mpeg derivatives, thereby getting a default from the set of parameters configured for these
[btw; the defaults for these parameters pretty much correspond to those of mencoder]

more ffmpeg mysteries...

The backtrace is the following:

Program received signal SIGSEGV, Segmentation fault.

+ Trace 115166

Looking deeper I can't see the relationship between setting proper intra_dc_precision and the fact that s->y_dc_scale_table has bogus value (here 0x1).

Also, you say setting a default of 0 for intra_dc_precision is what mencoder does. According to the mencoder man page, the default is 8 (ffmpeg default value):

      dc
              intra DC precision in bits (default: 8).  If you specify vcodec=mpeg2video this value can be 8, 9, 10 or 11.

I'm commiting only the first part of the patch, which states mpeg1video and mpeg2video are part of the mpeg family since that makes sense and gets rid of the segfault.

If you can come with proof that the second part of the patch is really needed, please open a new bug with the affected encoders.

2007-03-02  Edward Hervey  <edward@fluendo.com>

	* ext/ffmpeg/gstffmpegcfg.c: (gst_ffmpeg_cfg_init):
	Add mpeg1video and mpeg2video to the list of mpeg-derivatives for
	configuration settings.
	Renamed the mpeg4 category to mpeg.
	Fixes #403739

Well, it's fixed, so this is mainly "for the record" ;-)

(In reply to comment #2)
> Looking deeper I can't see the relationship between setting proper
> intra_dc_precision and the fact that s->y_dc_scale_table has bogus value (here
> 0x1).

I am admittedly not particularly knowledgeable in lower levels of encoding,
but at least both dc coefficient (precision) and q scale affect quality;
and it is imaginable (?) that y__dc_scale__table (setup) is somehow affected
by the precision/bits of dc coefficient (in ffmpeg then at least apparently ;-))

(In reply to comment #3)
> Also, you say setting a default of 0 for intra_dc_precision is what mencoder
> does. According to the mencoder man page, the default is 8 (ffmpeg default
> value):
> 
>       dc
>               intra DC precision in bits (default: 8).  If you specify
> vcodec=mpeg2video this value can be 8, 9, 10 or 11.

From avcodec.h:
    /**
     * precision of the intra dc coefficient - 8.
     * - encoding: set by user
     * - decoding: unused
     */
     int intra_dc_precision;

So, due to the - 8 shift, the user level default of 8 does correspond to setting intra_dc_precision = 0 (ah, ffmpeg ..., and we now also know that ffmpeg does not really set it to a default of 0 aka 8 (?), or it would not segfault)

> I'm commiting only the first part of the patch, which states mpeg1video and
> mpeg2video are part of the mpeg family since that makes sense and gets rid of
> the segfault.
Yes, it will, because the first part is sort of a superset of the second part.
By including it in the MPEG-family, intra_dc_precision is then one of the many parameters that gets a default (of 0 aka 8) in this case.

> If you can come with proof that the second part of the patch is really needed,
> please open a new bug with the affected encoders.
Either part of the patch by itself would fix it.
The second part is/was there for additional safety;
if ever one would feel like "playing" with the categories so that intra_dc_precision does not get a safe default anymore, there would be segfault (danger) again.  Hence, a hard-coded default = mencoder was provided.
As such, it is not needed for MPEGx in (the now) present setting, and have no proof forthcoming for other encoders, but it is likely a safe(r) default given mencoder's practice.