Re: C++, extern "C" and () (was Re: [C patch] void_type_node)

[Jason Merrill <jason at redhat dot com>]

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

 > I still am not convinced by the "old C headers require this" argument,
 > which is the root cause of this debate.  In other words, I'm still not
 > at all convinced we can't just define the problem out of existence by
 > removing this extension.

I don't like that idea (surprise, surprise).

 >   - There are semantic issues raised on the lists here that suggest
 >     that there is no easy way to know what do with `extern "C" void
 >     f()'; that can mean lots of things, and it's hard to
 >     simultaneously conform to the standard and be generous.

There is an easy way.  Treat it like C: a parameter list of NULL_TREE.  The
middle-end knows what to do with that for C, so it can do the same thing
for C++.

 >   - There are lots of problems with C headers in C++; this is only one of
 >     them. 

Yes, and we make an effort to work around all of them.

 >   - The major relevant example here is the X headers -- but on systems
 >     that old are they really the only headers that have these 
 >     problems?  Or other problems that prevent compilation with G++?

No, they usually aren't the only ones.  That doesn't seem like an argument
against the extension.

 >   - EDG does not support this usage, suggesting that many tools
 >     vendors don't find support for this extension necessary.

EDG does not market to end-users; many tools vendors have control of their
own headers.

 >   - There are other means for dealing with this issue, including
 >     upgrading the X headers, or fixincludes, or ...

Upgrading the headers is not a real option.  

This sort of thing is easier to deal with in the compiler, which has more
semantic information, but I would be open (if skeptical) of a fixincludes
solution.

 > I believe that this extension might have been appropriate in the past,
 > and even that there are people who still make use of it.  But they're
 > almost certainly on antiquated non-GNU systems.

How about just disabling it on NO_IMPLICIT_EXTERN_C targets?

Jason