[PATCH] D11789: Modify DeclaratorChuck::getFunction to be passed an Exception Specification SourceRange

[Richard Smith]

rsmith added inline comments.

================
Comment at: lib/Sema/SemaDecl.cpp:7447-7450
@@ -7446,1 +7446,6 @@
 
+      if (const FunctionProtoType *FPT = R->getAs<FunctionProtoType>()) {
+        if (FPT->hasExceptionSpec()) {
+          auto LocBeg = D.getFunctionTypeInfo().getExceptionSpecLocBeg();
+          auto LocEnd = D.getFunctionTypeInfo().getExceptionSpecLocEnd();
+          Diag(LocBeg, diag::err_function_concept_exception_spec)
----------------
nwilson wrote:
> rsmith wrote:
> > This will assert if there isn't a `FunctionTypeInfo` for the declaration, which can theoretically happen if it's declared via an (ill-formed today) `typedef`. (It also might not provide a source range if the exception specification is implicit, for instance because the function template is a destructor or deallocation function, but passing an empty SourceRange to the FixItHint should just result it in being ignored.)
> Hmm, I'm not sure if we'd run into that case because I don't believe we can have a `concept` specified as a typedef (check is yet to be added) and a check exists for being in a non-namespace scope. Do you think a check should still be added verifying that the FunctionTypeInfo exists?
Yes, you should check that a `FunctionTypeInfo` exists (`isFunctionDeclarator`) rather than assuming that it does. The testcase would look something like this:

    typedef int Fn() noexcept;
    template<typename T> concept Fn C;

I think we'll discard the `noexcept` early at the moment, but that's not something you should be subtly relying on here (especially given that http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2015/n4533.html is looming on the horizon). Something like:

  SourceRange Range;
  if (D.isFunctionDeclarator())
    Range = D.getFunctionTypeInfo().getExceptionSpecRange();

... should handle this safely.


http://reviews.llvm.org/D11789