GCC Cauldron: Notes from the C++ ABI BOF

[Jason Merrill]

On 01/22/2013 07:42 PM, Cary Coutant wrote:
> I believe we required an explicit attribute on the forward declaration
> in such a case.

The question is, what do we want to do for a user type that, say, has a 
std::string field.  Rejecting the program would be non-conforming, but 
otherwise we're likely to get different mangling in different 
translation units.

Templates also present a problem, as we can't propagate the tags until 
we instantiate, which might not be required when just passing pointers 
around.

As a result of these issues, I decided that the best we can do is to 
warn if a type is missing a tag that one of its subtypes has.

Jason