WIP: linux patchset

[ihnorton]

Oh good, now I can segfault in Qt! 😄. I figured it makes more sense to just open this as a wip until it works and then refine it to something mergeable.

further changes from last night:

• another missing libcxxffi in an eval'd ccall definition
• RTLD_GLOBAL for the Qt libs

in Julia I also added:

extern "C" void *cjulia_type_to_llvm(jl_value_t* jt) { return (void*) julia_type_to_llvm(jt); }

because that function is otherwise C++ mangled, so could not be found.

[ihnorton]

Ah, here we go! The segfault was due (apparently) to passing argc=0 into QApplication.

[jakebolewski]

🎆

[ihnorton]

@Keno I'm not sure if clang_compiler->getHeaderSearchOpts().UseBuiltinIncludes = 1; and the following options are doing what we expect - I had to manually defineMacro("__linux") for the Qt example to work (otherwise Qt thinks it is on an unknown platform).

[Keno]

Yeah, there's probably some other preprocessor options to set.

[Keno]

Try 69a0f28

[Keno]

What was the problem here? I have just pushed a change which calls Triple::normalize before passing it in. Does that fix it?

[ihnorton]

Yes. I missed normalize, good to know.

[Keno]

I did too. I even submitted a patch to clang to fix all the triple settings, when it was pointed out to me that I had missed normalize.

[ihnorton]

It almost seems like the default behavior should be normalize and the current behavior should be called getUnhelpfulTargetTriple.

[ihnorton]

Rebased, not much left :)

(I do still have to re-link Julia against LLVM3.5-svn to avoid missing symbols)

[Keno]

Yes, I do to. But now the Makefile picks up USE_LLVM_SHLIB from Make.user, which makes it exceedingly convenient ;)

[ihnorton]

Ah, cool!