Linux: Wide character constants are interpreted as UTF16, which is wrong.

[PathogenDavid]

This line assumes that wide character constants (IE: L"Hello, World") are UTF16, but this is not the case on Linux:

https://github.com/InfectedLibraries/Biohazrd/blob/e9145829b80dfc1d146177f66ef737930276f89f/Biohazrd/Expressions/ConstantValue.cs#L55

[PathogenDavid]

I'll probably end up just querying sizeof(wchar_t), but for posterity:

• TargetInfo is what determines the platform-specific sizes of things. You can see all the defaults here: https://github.com/InfectedLibraries/llvm-project/blob/d611e27a0f17a5775907fa048a3680fa29e22022/clang/lib/Basic/TargetInfo.cpp#L92
• Here's where wchar_t is set to be unsigned short on Windows targets: https://github.com/InfectedLibraries/llvm-project/blob/d611e27a0f17a5775907fa048a3680fa29e22022/clang/lib/Basic/Targets/OSTargets.h#L737
• Here's the flow for getting the size of a builtin (with wchar_t as an example):
  • ASTContext::getTypeInfo (attempts to find info from cache)
  • ASTContext::getTypeInfoImpl
  • TargetInfo::getWCharWidth
  • TargetInfo::getTypeWidth

One interesting thing of note is that all builtins (apparently) end up being one of the five different sizes with signed/unsigned variants. Some of the sizes are completely fixed (for example, char is always 8 bits in Clang, short is always 16.) We can also use this to indirectly see which platforms are weird. (For instance, it looks like bool is 8 bits on everything except for PPC which uses 32-bits.)

[PathogenDavid]

Since the translation unit is not available in the area where we decode constants, I actually ended up modifying pathogen_ComputeConstantValue in MochiLibraries/llvm-project@d9c68a3 to use the appropriate UTF kind for wchar_t constants. I also added a bit which is set when the string was a wchar_t constant just in case we ever care.