Re: [patch] Fix PR c++/29022

[Mark Mitchell <mark at codesourcery dot com>]

Lee Millward wrote:

> cp/
> 2006-10-18  Lee Millward  <lee.millward@codesourcery.com>
>
>        PR c++/29022
>        * parser.c (cp_parser_base_clause): If any of the
>        base clauses are invalid, discard all of them.
>
> testsuite/
> 2006-10-18  Lee Millward  <lee.millward@codesourcery.com>
>
>        PR c++/29022
>        * g++.dg/inherit/virtual2.C: New test.

I don't think this is the right fix.  In particular, I would expect this 
variant:

struct A
{
  operator int();
};

struct B : virtual A;

int foo(B &b)
{
  return b;
}

to fail even with your change.  I think the problem is that we're 
remembering that B inherits from A, even when we don't have a definition 
of B.  We shouldn't do that; if B isn't actually complete, then we 
should just forget all of its base classes.  I would suggest moving the 
xref_basetypes call out of cp_parser_class_head into 
cp_parser_class_specifier, and calling it only if we find the open curly 
brace for the class.  At that point, we're committed to completing the 
class, so we would avoid the inconsistent incomplete-but-has-bases 
situation we get at present.

Would you mind trying that approach?

Thanks,

--
Mark Mitchell
CodeSourcery
mark@codesourcery.com
(650) 331-3385 x713