[llvm-testresults] buildbot failure in smooshlab on llvm-gcc-powerpc-darwin9

[Jakob Stoklund Olesen]

On Sep 14, 2010, at 8:55 AM, Gabor Greif wrote:

> Hi Daniel,
> 
> this is a strange thing, no other target has a problem.
> 
> I can imagine two things:
> 
> - endianness (PPC is big-endian)
> - miscompilation by old GCC (which one is the bootstrap compiler on
> that machine?)
> 
> Can you send me a backtrace? Thanks!

Hi Gabor,

The tester is failing 2007-03-26-regscavengerassert.ll in both stage1 and stage2, so it is unlikely to be a miscompilation since the system gcc and the stage1 llvm-gcc agree that it should crash.

The two failures were segfaults in different places, so it looks like you introduced some memory corruption. It is a little tedious to get stack traces from this very slow machine, and they are not likely to help you. The memory corruption happens before the segfault.

I don't know why this would crash only on ppc, but it does seem to be consistent.

/jakob