[PATCH] D28166: Properly merge K&R functions that have attributes

[Reid Kleckner via Phabricator via cfe-commits]

rnk added inline comments.


================
Comment at: lib/Sema/SemaDecl.cpp:11958-11962
+        // The type location may be attributed; strip the attributes to get to
+        // the function type location.
+        while (auto ATL = TL.getAs<AttributedTypeLoc>()) {
+          TL = ATL.getModifiedLoc();
+        }
----------------
rsmith wrote:
> Again, I don't like having this knowledge about what kinds of type sugar can appear in a function declaration hardcoded here. Please put this somewhere more central.
> 
> A quick look finds that `FunctionDecl::getReturnTypeSourceRange()` gets this wrong in the opposite direction: it skips parens but not attributes. Maybe we should have a `TypeLoc::getAsAdjusted<T>` or similar, that walks over type sugar nodes that represent some kind of type adjustment from a type that was written as a T to another type that is still canonically a T (`ParenType`, `AttributedType`, `ElaboratedType`).
What about AdjustedType? You would see that in this case:
  template <typename T> struct A {
    T A::*method; // adjusted to __thiscall on i686-windows
  };
  template struct A<void()>;
Do you want to look through the __thiscall adjustment?


https://reviews.llvm.org/D28166