On 11 November 2013 21:12, Andreas Tobler <andre...@freebsd.org> wrote: > On 11.11.13 22:02, Peter Maydell wrote: >> I think I would agree with the commenter in the bug report you >> reference (http://llvm.org/bugs/show_bug.cgi?id=17882) >> that this is not a clang bug. We shouldn't be relying on the >> compiler's dead code elimination to get rid of references to >> functions that don't exist in certain configurations. This will >> always be unreliable (especially if compiling without optimization). >> Instead we should either be using ifdefs or stub functions (probably >> the latter in this case). > > I know it is a difficult business. And probably you're right, but from a > dump users point of view I do not agree. > > I'm used to gcc, which is able to compile this, and I expect other > compilers to be able to do the same. The compiler should work for me not > vice versa :)
Unfortunately C doesn't work that way. If you rely on behaviour which isn't guaranteed by the compiler authors then you have unreliable programs. I bet if you asked the gcc developers they'd say they didn't guarantee this to work either. >> If you put a stub implementation of kvm_arch_get_supported_cpuid() >> into target-i386/kvm-stub.c does this fix the compilation issue? > > Will try and let you know. I've actually reproduced this with my clang/macos system (it was masked by another build failure for which there's a patch on list that hasn't been applied yet). I'm working up a patch that I'll send shortly. thanks -- PMM
