Hello tech list, clang for 32-bit powerpc has a bug that breaks va_arg(3) when the argument type is an object or block in Objective-C. This breaks GNUstep on macppc. This clang diff fixes GNUstep. Objective-C uses pointers to represent objects and blocks, so this diff tells clang's va_arg to handle things with pointer representations as pointers.
Anthony Richardby found and reported the bug, https://bugs.llvm.org/show_bug.cgi?id=47921 I posted this diff at https://reviews.llvm.org/D90329 but upstream llvm might do something else. I used a clang with this diff to rebuild src and xenocara on macppc. ok to commit? --George Index: clang/lib/CodeGen/TargetInfo.cpp =================================================================== RCS file: /cvs/src/gnu/llvm/clang/lib/CodeGen/TargetInfo.cpp,v retrieving revision 1.1.1.2 diff -u -p -r1.1.1.2 TargetInfo.cpp --- clang/lib/CodeGen/TargetInfo.cpp 9 Aug 2020 15:51:11 -0000 1.1.1.2 +++ clang/lib/CodeGen/TargetInfo.cpp 28 Oct 2020 23:43:54 -0000 @@ -4248,8 +4248,8 @@ Address PPC32_SVR4_ABIInfo::EmitVAArg(Co // }; bool isI64 = Ty->isIntegerType() && getContext().getTypeSize(Ty) == 64; - bool isInt = - Ty->isIntegerType() || Ty->isPointerType() || Ty->isAggregateType(); + bool isInt = Ty->isIntegerType() || Ty->hasPointerRepresentation() || + Ty->isAggregateType(); bool isF64 = Ty->isFloatingType() && getContext().getTypeSize(Ty) == 64; // All aggregates are passed indirectly? That doesn't seem consistent
