[PATCH] D16465: [MS ABI] Prevent some expressions from evaluating to a constant

[Reid Kleckner via cfe-commits]

rnk added a comment.

Richard really is the constexpr owner so I'd like him to chime in if he can.


================
Comment at: include/clang/AST/Decl.h:1092
@@ -1091,3 +1091,3 @@
   /// succeeded, 0 otherwise.
-  APValue *evaluateValue() const;
-  APValue *evaluateValue(SmallVectorImpl<PartialDiagnosticAt> &Notes) const;
+  APValue *evaluateValue(bool RespectMicrosoftABIIfNeeded = false) const;
+  APValue *evaluateValue(SmallVectorImpl<PartialDiagnosticAt> &Notes,
----------------
Maybe this parameter would be better named `IsStaticLocalInit` or something.

================
Comment at: include/clang/Basic/DiagnosticASTKinds.td:151
@@ -150,1 +150,3 @@
   "%plural{1:byte|:bytes}1">;
+def note_constexpr_microsoft_abi_declrefexpr : Note<
+  "in the Microsoft ABI static local variables cannot contain references to "
----------------
> One solution to this would be to create a variation of evaluateValue() which activates this new behavior and only use it from CodeGenModule::EmitConstantInit(), so that the behavior of evaluateValue() in this context doesn't change. How does that sound?

Right, only having this behavior change in the awkward static local case is good.

> By the way, I just realized that CheckPotentialExpressionContainingDeclRefExpr() eats this diagnostic because of the SpeculativeLookForDeclRefExprRAII object. Should I propagate it up from CheckPotentialExpressionContainingDeclRefExpr() to make it user visible? Right now the test case above only emits "constexpr variable 'p' must be initialized by a constant expression" without any notes.

Yeah, I'd like it if we could get that note out. I'm surprised we can't use the normal recursive traversal to find the DeclRefExprs...


http://reviews.llvm.org/D16465