Microsoft X64 Mangling

[Charles Davis]

On May 2, 2013, at 3:19 PM, Warren Hunt wrote:

> I've put together a patch that fixes all of the issues I found with Microsoft name mangling in 64-bit mode.  Mostly it involves adding an 'E' note to all pointers that are 64-bit (including 'this' pointers).  This is done in MicrosoftMangle.cpp.  I've updated 3 of the mangling test files with X64 tests, which provides some reasonable amount of coverage.
> 
> Please review.
> 
> Thanks,
> -Warren
> 
> (Note: This is my first patch.)
Welcome to Clang development!

> @@ -1277,6 +1278,9 @@ void MicrosoftCXXNameMangler::mangleFunctionClass(const FunctionDecl *FD) {
>          else
>            Out << 'Q';
>      }
> +  if (getASTContext().getTargetInfo().getPointerWidth(0) == 64 &&
> +      !MD->isStatic())
> +    Out << 'E';
>    } else
>      Out << 'Y';
>  }
The indentation on this looks wrong. (You can't tell from here, but it's definitely wrong in the patch itself.) Also, I think we should handle this where we mangle in the 'this' qualifiers--the 'E' is because the 'this' pointer is Extended (__ptr64 instead of __ptr32). (The relevant line is 1172.)

> @@ -1390,6 +1394,8 @@ void MicrosoftCXXNameMangler::mangleDecayedArrayType(const ArrayType *T,
>      manglePointerQualifiers(T->getElementType().getQualifiers());
>    } else {
>      Out << 'Q';
> +    if (getASTContext().getTargetInfo().getPointerWidth(0) == 64)
> +      Out << 'E';
>    }
>    mangleType(T->getElementType(), SourceRange());
>  }
Why only here? What does VC do in the global case?

There's a lot of code duplication (i.e. checks against the pointer size sprinkled throughout the code). Most of those can be collapsed into mangleQualifiers().

Someone else wrote a patch a while back. You can find it here:

http://llvm-reviews.chandlerc.com/D101

I didn't finish reviewing it because real life got in the way. :\. You might find some ideas in there.

Chip