[cfe-dev] Missing lvalue-to-rvalue cast for first arg of ->* ?

[John McCall]

On Jun 30, 2011, at 12:17 PM, Enea Zaffanella wrote:
> Il 30/06/2011 19:16, John McCall ha scritto:
>> On Jun 30, 2011, at 1:55 AM, Enea Zaffanella wrote:
>>> When dumping the AST for the following C++ program:
>>> 
>>> struct B { int mem; };
>>> 
>>> void foo(B* pb, int B::* pmem) {
>>> pb->mem;
>>> pb->*pmem;
>>> }
>>> 
>>> it can be seen that the lvalue pointer pd, when used as the first
>>> argument of MemberExpr pb->mem, is subject to an LValueToRValue implicit
>>> cast.
>>> 
>>> This is not the case when pb is used as the first argument in the
>>> member-pointer operator ->*.
>>> 
>>> Similarly, lvalue member-pointers (such as pmem) occurring as the second
>>> argument of ->* and .* operators are not subject to LValueToRValue
>>> implicit casts.
>>> 
>>> Is this difference meant (and why)?
>> 
>> No, it isn't.  Thanks for pointing it out;  fixed in r134170.
>> 
>> John.
>> 
> 
> Thank you for the very quick fix!
> 
> I have another question, regarding the AST structure of
> CXXPseudoDestructorExpr nodes.
> 
> ==============
> template <class T> class Node {
>  void bar() { this->~Node<T>(); }
> };
> 
> template <class T>
> void foo(Node<T>* n) { n->~Node<T>(); }
> ==============
> 
> In method bar(), the "destructed type" (i.e., Node<T> after the tilde)
> is a TemplateSpecializationTypeLoc (TSTLoc, for short), which is fine.
> 
> In function foo(), parameter n is declared to be a TSTLoc,
> but the destructed type is a DependentTemplateSpecializationTypeLoc
> (DTSTLoc).
> 
> I think I have read some comments in the sources telling that this is
> quite a dark corner of the standard ... anyway what is surprising to me
> is the fact that in foo() we have a DTSTLoc node with *NO* nested name
> specifier. I was under the impression that a DTSTLoc should always have
> a non-empty NNS.

It's possible to have a DTST without an NNS;  this is one case,
and another would be something like:

  foor->template A<T>::bah();

Basically, we use a DTST whenever we can't do the lookup for the
template name yet, but we nonetheless know it has to be a template
and has to be a type.

John.