rnk added a comment.
Your code won't catch this test case:
static int x;
extern inline const bool *f() {
static const bool p = !&x;
return &p;
}
Getting this exactly right is going to be a challenge. =/
================
Comment at: include/clang/Basic/DiagnosticASTKinds.td:151
@@ -150,1 +150,3 @@
"%plural{1:byte|:bytes}1">;
+def note_constexpr_microsoft_abi_declrefexpr : Note<
+ "the constant expression cannot contain a reference to a variable as a
Microsoft "
----------------
We should add this test case and decide what to do with it:
static int x;
inline int **f() {
static constexpr int *p = true ? 0 : &x;
return &p;
}
Currently, in your patch, this diagnostic will come out. MSVC compiles this to
guarded, dynamic initialization, despite the constexpr. ;_;
David thinks we should just give the user the real deal constexpr behavior,
even though it's ABI incompatible.
================
Comment at: include/clang/Basic/DiagnosticASTKinds.td:152-153
@@ -151,1 +151,4 @@
+def note_constexpr_microsoft_abi_declrefexpr : Note<
+ "the constant expression cannot contain a reference to a variable as a
Microsoft "
+ "ABI extension">;
----------------
This isn't an extension, so we should say something else. Maybe:
"in the Microsoft ABI static local variables cannot contain references to
variables"
================
Comment at: lib/AST/ExprConstant.cpp:9008
@@ -8917,1 +9007,3 @@
+ if (Ctx.getTargetInfo().getCXXABI().isMicrosoft())
+ InitInfo.useMicrosoftABI();
----------------
This should be limited in scope to only apply to static locals. We should be
able to statically initialize globals.
http://reviews.llvm.org/D16465
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
