http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45841
--- Comment #8 from Hans-Peter Nilsson <hp at gcc dot gnu.org> 2010-09-30 11:46:47 UTC --- (In reply to comment #4) > Or do you guys happen to have a setup I can ssh into? :v) Nominally, you *could* repeat my findings with a cross-compiler setup as described by <http://gcc.gnu.org/simtest-howto.html> but using that to investigate an execution failure might be a bit too challenging for a newcomer. In my autotester using that kind of setup, this failure is a regression. There are certainly other failures (for different reasons; bugs/shortcomings in newlib, bugs in the test-cases etc.), but this one is a *regression*. I guess from the earlier comments, your patch is rather exposing another issue, not the direct cause. The trace I mentioned (I'll try to get to it this week) will show me at least the sequence of basic libc or system operations were used (seek, read, etc.) and I hope then you, by knowing the code around the patch, can help me by telling whether a specific call is sane or bogus, so we can pinpoint the failing code. No, the trace can't be used with gdb, but only because that's a feature that's not yet been implemented. ;-)