[patch] Fix second part of PR c++/27601: ICE using offsetof with member functions

[Volker Reichelt]

On  5 Jun, Mark Mitchell wrote:
> Volker Reichelt wrote:
>> While applying
>>   http://gcc.gnu.org/ml/gcc-patches/2006-06/msg00161.html
>> I noticed that the following invalid code snippet also triggers an ICE:
>> 
>>   struct bar {
>>     void foo();
>>   };
>> 
>>   int a = __builtin_offsetof(bar, foo);
>> 
>> bug.cc:5: internal compiler error: tree check: expected field_decl, have baselink in fold_offsetof_1, at c-common.c:5992
>> Please submit a full bug report, [etc.]
> 
> Can't we catch this in the C++ parser?  __builtin_offsetof is handled by
> its own parsing routine, so before we call fold_offsetof, we already
> know what EXPR is.

You can't catch it in the parser for templates:

=================================================
struct A
{
  int foo;
};

struct B
{
  void foo();
};

template<typename T> struct C
{
  C() { __builtin_offsetof(T, T::foo); }
};

C<A> a;  // OK
C<B> b;  // error (currently ICE)
=================================================

And instead of duplicating the code in parser.c and pt.c I'd rather
check it in fold_offsetof_1.

> Wouldn't calling invalid_nonstatic_memfn_p fix this?

The problem appears with static *and* non-static member functions.
I don't see how invalid_nonstatic_memfn_p helps in this situation.

Regards,
Volker