Add the -fsanitize-blacklist option for the sanitizers.

[JohanEngelen]

Having fun with ASan :)
Blacklisting is necessary for some druntime functions, and LLVM/Clang already provides just the right functionality.

Will add the actual needed blacklist in a later PR.

[JohanEngelen]

Ah ofcourse. Windows mangling.

[JohanEngelen]

Any concerns?

[kinke]

Looks good after a quick glance, just one thing: we shouldn't use the IR function name, as the necessity to fix up the testcase for Win32 clearly shows. ;)

[JohanEngelen]

I thought about using the "normal" function name, without the \01 added for Win32, but then I thought that because of different underscore conventions, the start of the mangled name is already different anyway so you'd have to use a * at the start already ("*_D...."). The blacklist for Clang is using mangled names, and I don't want to deviate from that.
What I can do is to shorten the StringRef by one if we detect a '\01' at the start of the string?

[JohanEngelen]

(eager to get this in btw, it's annoying me with asan work)

[JohanEngelen]

(also, I could get rid of the extra _ added for win32.......)

[kinke]

I meant just using the regular D mangle (mangleExact()), not the ugly stuff we have to add to make LLVM use that name for the object file [and without that dubious additional underscore on some platforms] in TargetABI::mangleFunctionForLLVM().

[JohanEngelen]

Yep indeed, I am already reworking things. function.cpp should just check whether the FuncDecl is in the blacklist, without concerning itself about mangling/blacklist details. I've changed the lookup to use mangleExact (whose returned string is cached).

[JohanEngelen]

(I wanted things to be as close as possible to the real mangled name in the object file, including hashing. But... I can think of other use cases where blacklisting using the non-hashed name makes more sense. So I'm happy with mangleExact too)

[kinke]

I took the liberty to address my last nit, passing front-end objects unnaturally by reference (and so wouldn't be callable from D via D object reference, which is to be mangled as pointer for C++ functions).

[JohanEngelen]

The by-reference was intentional: the interface does not allow passing null pointers.
Inability of D code to bind to C++ class references leading to bad code :(

[JohanEngelen]

I'll merge when Travis is green.

[kinke]

the interface does not allow passing null pointers.

Refs don't protect from passing nulls under the hood, the segfault (in func()) is just a FuncDeclaration *fd = nullptr; func(*fd) away.