(C++) patch to protected member access

[Jason Merrill]

>>>>> Mark Mitchell <mark@codesourcery.com> writes:

>>>>> "Jason" == Jason Merrill <jason@cygnus.com> writes:
 Jason> Note that my patch also fixes

 Jason> struct A { protected: int i; };

 Jason> struct B: public A { };

 Jason> struct C: public A { void f () { B::i = 0; } };

 Jason> Here, i is accessible when named in B because it's
 Jason> accessible when named in A.  Before my patch, g++ was
 Jason> treating B specially in protected_accessible_p, which is
 Jason> not appropriate.

 >>> I assume you meant `C : public B'?

 Jason> Nope, I meant what I wrote.  It's an odd thing to do, but I
 Jason> believe well-formed.

 > I'll believe you.  But, it seems to me that, conceptually, this ought
 > to fall under the "you can't use protected things unless you're part
 > of the object yourself" rule.  That code can't say:

 >   B b;
 >   b.i = 3;

 > because `i' is protected in `B' and `b' is not a `C' object.  So, it
 > seems odd to say that `B::i' is legal.

It's legal because B::i means this->B::i, and *this is a 'C' object.  The
explicit scope only affects name lookup, which finds A::i, which is
accessed through this.

It used to be that 'B::i' meant something like '((B*) this)->i', but the
committe has abandoned that.  I don't really agree, but there it is.

Jason