[PATCH] D39937: [Sema] Improve diagnostics for const- and ref-qualified member functions

[Roman Lebedev via Phabricator via cfe-commits]

lebedev.ri added inline comments.


================
Comment at: test/CXX/over/over.match/over.match.funcs/p4-0x.cpp:22-24
+  void lvalue() &; // expected-note 2 {{'lvalue' declared here}}
+  void const_lvalue() const&;
+  void rvalue() &&; // expected-note {{'rvalue' declared here}}
----------------
aaron.ballman wrote:
> jtbandes wrote:
> > aaron.ballman wrote:
> > > Can you add examples that cover the other diagnostic wordings as well (volatile, restrict, combinations, etc)?
> > I've been working on this, but I actually can't trigger the `restrict` variants. Do you know whether this is something that's expected to work? The implicit object param doesn't seem to retain its restrict-ness (full disclosure, I have almost no prior experience with `restrict`...):
> > 
> > ```
> >   void c() const;
> >   void v() volatile;
> >   void r() __restrict__;
> >   void cr() const __restrict__;
> >   void cv() const volatile;
> >   void vr() volatile __restrict__;
> >   void cvr() const volatile __restrict__;
> > ```
> > ```
> > void test_diagnostics(const volatile X0 &__restrict__ cvr) {
> >   cvr.g(); // expected-error {{not marked const, volatile, or restrict}}  -- actually produces "not marked const or volatile"
> >   cvr.c(); // expected-error {{not marked volatile or restrict}}  -- actually produces "not marked volatile"
> >   cvr.v(); // expected-error {{not marked const or restrict}}  -- actually produces "not marked const"
> >   cvr.r(); // expected-error {{not marked const or volatile}}
> >   cvr.cr(); // expected-error {{not marked volatile}}
> >   cvr.cv(); // expected-error {{not marked restrict}}  -- actually produces no error
> >   cvr.vr(); // expected-error {{not marked const}}
> >   cvr.cvr();
> > }
> > ```
> Given that `restrict` is a Cism, it's entirely possible that it's not really supported as a member function qualifier. In that case, one more test for `const volatile` should be sufficient.
Note that you might want to test it regardless, to not miss if it starts to warn for some reason.


https://reviews.llvm.org/D39937