Re: autoconf 2.5, 2.13, and Vim

[Russ Allbery]

Akim Demaille <address@hidden> writes:
>>>>>> "Paul" == Paul Eggert <address@hidden> writes:

> Paul> One compatibility point, though.  Currently, autoconf assumes
> Paul> that <sys/types.h> exists, but this assumption is valid only for
> Paul> POSIX-related items, as <sys/types.h> is required only by POSIX,
> Paul> not by the C Standard.

> That's the first time ever someone suggest to check for it!

Yeah, I'm a touch skeptical about this... while what Paul says is
certainly true, I've yet to see *any* system that didn't have
<sys/types.h>.  Maybe my experience isn't broad enough, but it's included
without guards in pretty much every major C package I've seen.

I think this may be a bit beyond the practical limits of autoconf's
portability; after all, its output relies on the existence of a Bourne
shell, so it already can't run on *anything*.  Although I suppose it can
theoretically cross-compile to anything....

> So, are you actually saying that *any* #include <foo.h> should be
> checked?  Instead of having acheaders list plenty of headers, shouldn't
> it just list all the headers that have specific macros, and then,
> autoscan would ask for checking all the others?

Hm.  One problem with autoscan checking for too much stuff is that in many
cases there's really no point in doing so unless the package is prepared
to deal with that header going missing.  Sure, I could check for
<sys/types.h>... but what if it doesn't exist?  Then what?

Part of me says I should argue in favor of not worrying about this until
someone actually asks for this feature because they need it in a real
program.  (Of course, Paul may well be doing that and I'm just not
understanding.)

-- 
Russ Allbery (address@hidden)             <http://www.eyrie.org/~eagle/>