eandrews added inline comments.
================ Comment at: clang/unittests/Interpreter/InterpreterTest.cpp:237-240 + std::string MangledName = MangleName(TmpltSpec); + typedef int (*TemplateSpecFn)(void *); + auto fn = (TemplateSpecFn)cantFail(Interp->getSymbolAddress(MangledName)); + EXPECT_EQ(42, fn(NewA)); ---------------- v.g.vassilev wrote: > erichkeane wrote: > > aaron.ballman wrote: > > > This test is broken for some of our internal build bots at Intel. I think > > > something suspicious is going on here, but I'm not certain of the intent > > > behind the test, so I'm not certain the best way to fix it. The behavior > > > of the test is that on an x86 Windows machine, sometimes this particular > > > test fails: > > > ``` > > > [ RUN ] IncrementalProcessing.InstantiateTemplate^M > > > unknown file: error: SEH exception with code 0x3221225477 thrown in the > > > test body.^M > > > [ FAILED ] IncrementalProcessing.InstantiateTemplate (35 ms)^M > > > ``` > > > but it's not a consistent failure (seems to happen about one out of every > > > three runs). > > > > > > `callme` is a templated member function of `B` but here we're trying to > > > call it like it's a free function. That's... not good. We could either > > > make `callme` a `static` member function so that it can be called in this > > > manner, or we could try to come up with a magic incantation to call it as > > > a PMF. > > > > > > Can you investigate, @v.g.vassilev? If it is going to take considerable > > > time to resolve, it might be worth reverting temporarily. Thanks! > > To be perhaps succinct, the problem is calling Pointer-to-member-function > > as a free-function. The cast on line 239 is completely invalid, and can't > > be done in C++ code itself without UB. > > > > We believe this shows up on x86 in particular since there is ALSO a > > calling-convention mismatch here, but this cast to TemplateSpecFn seems > > completely invalid to me. > Thanks for the ping, @aaron.ballman. Does making the function static fixes > it? If it does, please go ahead and commit the fix otherwise I can do it in > 12 hours. > > If it does not fix it, could you share the bot logs or at least the cmake > configure line and the platform? Committed here - https://reviews.llvm.org/rG3ad0c6b75ea5 CHANGES SINCE LAST ACTION https://reviews.llvm.org/D112663/new/ https://reviews.llvm.org/D112663 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
