Continuing my effort to port clang to powerpc-darwin, I've been building and testing libc++, in anticipation of the upcoming C++11 requirement for the stage-1 host compiler.
I've built libc++ two ways: against the existing system libsupc++.a (old, shipped with apple's gcc-4.0) and against a built libc++abi.dylib.
These builds include the recent FDE/EH-frame patches from Iain which fixed most exception-related failures and libc++abi test failures, of which only two remain: http://llvm.org/bugs/show_bug.cgi?id=17505.
I've also faked 64b atomics on PPC32 using std::lock_guard<mutex>, which is probably horribly-slow, but better-than-nothing, but lets the atomic tests pass.
And here are my results, 150k each (or append .bz2 for compressed):
libsupc++ is rather old and missing the dependent EH ABI that is in libc++abi, and it is missing and implementation of exception_ptr, which probably accounts for the majority of test differences.
Is there any way to provide the missing features on top of the existing supc++ ABI, or will that only result in incompatibility?
Continuing down the TODO list is triaging the test failures common to both libc++ builds. Any insights on these would be greatly appreciated.