Re: posix/glob.c: update from gnulib

[DJ Delorie <dj@redhat.com>]

Paul Eggert <eggert@cs.ucla.edu> writes:
>> Used config.h instead of libc-config.h
>
> I don't see why this change is needed. This code is inside "#ifndef 
> _LIBC" so this change should have no effect for glibc. And the change is 
> harmful for Gnulib, since for this file Gnulib relies on including 
> libc-config.h instead of plain config.h.

I was just leaving that line as it was in glibc, but as you note, it
shouldn't make a difference.  glibc does have a config.h though, and no
libc-config.h, which is what made me think this was significant.

>> The #ifdef around #define dirfd() was changed to #undef due to
>> conflicts between glibc's internal and external definitions of
>> dirfd().  This has been reported to gnulib.
>
> I updated Gnulib to reflect this change; see first attached patch. That 
> being said, I don't fully understand it. Wouldn't it be more efficient 
> for glibc glob to use glibc's internal dirfd by whatever name you prefer?

Perhaps, but doing so involved more than just using a macro, because a
lot of what the internal macro does depends on having insight into the
internals of other modules (like typedef names, for example), which
involves other includes (for those typedefs) and such.  It was far
easier to just leave it alone than to go down that (perhaps shallow ;)
rabbit hole.  I.e. it was the minimum fix.

Since it's just a dereference I don't think it will affect performance.

> Anyway, the only difference between what you proposed for glibc and 
> current Gnulib glob is the second attached patch; could you please merge 
> that into your proposal? That way, the two glob.c files can be 
> identical, which is a good thing.

Done.