[clang] [Clang] Make the result type of sizeof/pointer subtraction/size_t lit… (PR #136542)

[Aaron Ballman via cfe-commits]

AaronBallman wrote:

> > > Thank you for this!
> > > I'd like to better understand the need for the changes because I have a few concerns. One concern is about compile time performance. But also, this means downstream consumers of the AST are going to have to react because they used to be able to look for a `size_t` node directly and now they have to resolve a qualified type instead. This may be acceptable, but it seems disruptive too.
> > > Also, there should be more test coverage for the changes showing that we actually do get the types correct in all the various circumstances.
> > 
> > 
> > The current inlay hint of clangd is `auto a: unsigned long = sizeof(int);`, which is misleading. At the same time, it eliminates certain conversions that clang-tidy or other cleanup tools might warn about. The C and C++ standards state that the result type of such expressions is `size_t`/`ptrdiff_t`, so while this may disrupt some downstream assumptions about prior implementations, it aligns more closely with the standard. I believe this is worthwhile, maybe there's a faster way to implement it.
> 
> Yes, but this doesn't exactly accomplish that. In C, you'll still get the underlying integer type unless there happens to be a typedef we can find, right? So you can spot a difference between:

Actually, it might be even worse; I *think* it's valid for a user to define a typedef for `size_t` themselves so long as C standard library headers are not included, because it's not a reserved identifier in that case. I'm asking on the WG14 reflectors because it matters for a test case like:
```
typedef float size_t;
static_assert(_Generic(sizeof(int), size_t : 1, default : 0));
```
where it's unclear whether that static assertion should pass or fail.

https://github.com/llvm/llvm-project/pull/136542