[Ffmpeg-devel] [PATCH] have cs_test check for sigsegv at smaller widths and sigill

[Ivo]

Hi,

On Friday 13 April 2007 16:40, Uoti Urpala wrote:
> On Fri, 2007-04-13 at 16:20 +0200, Michael Niedermayer wrote:
> > On Fri, Apr 13, 2007 at 02:09:37PM +0200, Ivo wrote:
> > > No, i get it. I didn't think good enough about what sigsegv actually
> > > means to the state of the program. So, is the following patch ok and
> > > have the person that runs cs_test run gdb afterwards and/or fix the
> > > bug?
> >
> > i still dont understand what sense this patch has, if you found a
> > scaler which segfaults, fix the segfault or provide a proper bugreport
> > with gdb output (though fixing the segfault is more welcome)
> > changeing cs_test.c so that it triggers the scaler to segfault after
> > the patch has no sense it just breaks cs_test.c ...
>
> I think Ivo's point is to use cs_test.c as a testing tool, and he wants
> to change it so that it has testcases which reproduce the current bug in
> the scalers (and originally wanted to use the sigsegv handler in the
> test program to print details about problems found).

Yes, that's what I wanted. I agree with Michael that cs_test should not 
break though, so here's a patch that fixes rgb24tobgr24 and rgb32tobgr32. 
The MMX code is only run now if src_size is bigger or equal to the size of 
the units the MMX code handles.

After that I would like to change the loop in cs_test so it tests for every 
width down to a single pixel if that's ok with you.

--Ivo
-------------- next part --------------
A non-text attachment was scrubbed...
Name: twosegfaults.patch
Type: text/x-diff
Size: 1032 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20070413/596b1c09/attachment.patch>