[PATCH] D13704: [Fix] Allow implicit conversions of the address of overloadable functions in C + docs update

[Richard Smith via cfe-commits]

rsmith accepted this revision.
This revision is now accepted and ready to land.

================
Comment at: lib/Sema/SemaOverload.cpp:10419
@@ -10418,3 +10429,1 @@
-                                 ResultTy) ||
-          (!S.getLangOpts().CPlusPlus && TargetType->isVoidPointerType())) {
         Matches.push_back(std::make_pair(
----------------
george.burgess.iv wrote:
> rsmith wrote:
> > Why is the `void*` check removed from this case? Note that clang and GCC intentionally treat these two cases differently today:
> > 
> >     int f();
> >     void *p = f; // ok (warning under -pedantic)
> >     int *q = f; // warning: incompatible pointer types
> > 
> > (That is: the first is a silent-by-default extension and the second is a warn-by-default extension.)
> Because this is overload resolution logic, so we shouldn't care about what warnings we will emit :)
> 
> This is how we act prior to applying this patch:
> 
> ```
> void f(int) __attribute__((overloadable));
> void f(double) __attribute__((overloadable, enable_if(0, "")));
> 
> void *fp = f; // OK. This is C and the target is void*.
> void (*fp)(void) = f; // Error. This is C, but the target isn't void*.
> ```
> 
> I'm simply removing the "the target must be a `void*`" restriction; the user should still get warnings in the latter case (the tests changed in test/Sema/pass-object-size.c make sure of this).
OK, this seems fine so long as we somewhere choose exact matches over inexact ones:

    void f(int) __attribute__((overloadable));
    void f(int, int) __attribute__((overloadable));
    void g(void (*)(int));
    void h() { g(f); } // should pick f(int)


http://reviews.llvm.org/D13704