On 25 May 2009, at 11:19, Niels Grewe wrote: > I did regenerate it but had to do some tweaking: Clang (build from > r72105) did refuse to compile MsgSendSmallInt.m with the commandline > from the makefile. I needed to modify it this way:
r72105 should be recent enough... > clang-cc -c -ftrapv -emit-llvm -fexceptions -I/usr/lib/gcc/x86_64- > linux-gnu/4.3.3/include MsgSendSmallInt.m -o MsgSendSmallInt.bc This is just plain wrong. -emit-llvm will generate llvm IR, not bitcode, and -c is not a valid option for clang-cc: clang-cc: Unknown command line argument '-c'. Try: 'clang-cc --help' It looks like you somehow have clang installed as clang-cc. If you are using clang (the compiler driver, rather than the front end) then you will need an entirely different set of options. > "-c" was needed because clang was trying to link the object file > (which > failed), "-emit-llvm" because the "-emit-llvm-bc" switch was not > recognized and finally it kept complaining that it couldn't find the > objc/objc-api.h header, so I manually added the include path to my > gcc-version (Is this correct? I'm assuming LanguageKit is still > using the gnu-runtime and thus referred to the headers that came with > gcc's libobjc) This should already be in your search path. If it isn't, your clang install is quite badly messed up. Does 'make test' in the clang directory report anything? > Unfortunately, all the tests still fail with SIGABRT reporting a > failed > assertion deep within the innards of llvm: > > edlc: TargetRegisterInfo.cpp:59: virtual const > llvm::TargetRegisterClass* > llvm::TargetRegisterInfo::getPhysicalRegisterRegClass(unsigned int, > llvm::MVT) const: Assertion `BestRC && "Couldn't find the register > class"' failed. > > If have attached a backtrace. I hope it's of any use. This is a failure deep in the LLVM JIT code. What version of LLVM do you have installed? David _______________________________________________ Etoile-dev mailing list Etoile-dev@gna.org https://mail.gna.org/listinfo/etoile-dev