[Llvm-bgq-discuss] clang/bgclang

[Biddiscombe, John A.]

Thomas

>Sounds like some debug info would help isolate the problem library, i'd suggest enabling the dynamic loader's trace facility.  Set this environment variable:
LD_DEBUG=files,libs
<

Interesting. Prints out all the information regarding the libraries that are loaded, but the info is essentially the same as ldd shows me, no surprises are obvious, nothing that I didn’t expect to see gets pulled in.

>
or
LD_DEBUG=all
<

Gosh, even more info. Looks as though all the symbol names are resolved as one might expect - and then it segfaults. there are no clues near the segfault that any symbol gets found twice or anything that might corrupt the offset table.

I’m going to try  to build against the gcc-8.8.2 toolchain today, (though I know it won’t make any difference since the 4.8.2 compiled binaries segfault in the same place). what troubles me is that the binaries compiled with bgclang++11 did not segfault - so something must be different, just don’t know what.

JB


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alcf.anl.gov/pipermail/llvm-bgq-discuss/attachments/20140507/e57ffbbf/attachment.html>