[PATCH] D48040: Implement constexpr __builtin_*_overflow

[Eli Friedman via Phabricator via cfe-commits]

efriedma added inline comments.


================
Comment at: lib/Sema/SemaChecking.cpp:210
+    Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg);
+    TheCall->setArg(I, Arg.get());
   }
----------------
erichkeane wrote:
> efriedma wrote:
> > Can you split this change into a separate patch?  Testcase:
> > 
> > ```
> > int a() {
> >   const int x = 3;
> >   static int z;
> >   constexpr int * y = &z;
> >   return []() { return __builtin_sub_overflow(x,x,y); }();
> > }```
> Can you clarify what you mean?  That above testcase (with added captures) seems to work currently.  What difference in behavior should I be expecting?
The testcase should type-check as-is, without adding any captures.  Reading the value of a constexpr variable, or a const variable of integer type, isn't an odr-use.  This doesn't work correctly at the moment because the lvalue-to-rvalue conversions are missing from the AST. Compare to the following:

```
int a() {
  const int x = 3;
  static int z;
  constexpr int * y = &z;
  return []() { return __builtin_sub_overflow((int)x,(int)x,(int*)y); }();
}
```


https://reviews.llvm.org/D48040