Re: [bug-gnulib] New GNULIB glob module?

[Paul Eggert]

Derek Price <derek@ximbiot.com> writes:

> We've been removing those sort of constructs from CVS as part of the
> move away from K&R support.  IIRC, I was told that typecasts to and from
> (void *) were necessary on some really old systems, but that I could be
> confident that we wouldn't encounter any C89 systems that required it. 

That's correct.

> Perhaps this is simply legacy code related to the __ptr_t stuff and
> could be removed as well?

Yes, quite possibly.  How about if we simply assume C89-or-later for
now.  If someone runs into a problem with a bogus masquerading
compiler we can fix it later.

> why does that preclude a #include <stddef.h> in glob.h?

Because in glibc #include <glob.h> isn't supposed to define irrelevant
identifiers like wchar_t (which <stddef.h> does define).

It's OK for #include <glob.h> to define these identifiers in gnulib,
though.  Partly because we can't avoid it, and partly because we're
not as persnickety anyway.

I'll review the rest of your message later (sorry, too many meetings today!).