Re: AC_LANG_PROGRAM(C): function 'main' without prototype

[Eric Blake]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

According to Nico R. on 6/24/2008 12:14 PM:
| Hello!

Hello, and sorry for a delayed response,

| AC_LANG_PROGRAM(C) uses a ‘main’ function without a prototype. If CFLAGS
| contains ‘-Werror’ (it does in my case, because the configure script in
| the glibc package adds it), a test using AC_LANG_PROGRAM(C), for example
| AC_LINK_IFELSE, fails.
|
| Shouldn’t the line
|   main ()
| be
|   main (void)
| instead?

Per C99 5.1.2.2.1, main must be compatible with one of these two prototypes:
main(void);
main(int, char**);

But 6.7.5.3 states "An empty list in a function declarator that is part of
a definition of that function specifies that the function has no
parameters.", which is the same result as using a list of exactly void.

If that's not enough evidence that main() is compatible with main(void)
when coupled with the function definition, then you should also look at
the example in 6.5.3.4 that shows a use of main(){...} as a valid C99 program.

| Or may this cause problems with some compilers? Should
| ‘-Wstrict-prototypes’ be removed from CFLAGS instead? And should this be
| fixed in autoconf or should I report a bug against glibc?

I argue that using the construct main(){} not a bug in autoconf - we are
using a valid C99 program (and I assume C89 as well, although I don't have
C89 in front of me at the moment).  It may be worth asking the gcc folks,
to see if they are issuing a bogus warning for a valid program, or whether
they are intentionally giving a warning because the construct is of
dubious value even though it is valid.  Meanwhile, while we try to be nice
to configuring with -Werror, we do not guarantee that it will always work
(there are just too many compiler variants that warn about too many
distinct items to make any program warning free across all compilers).

Now, as to C89 vs. K&R, I think the list sentiment the last few months has
been that these days, we can rely on C89 compilers on most reasonable
platforms.  The proposal has been made to switch the default autoconf
behavior to require C89 (ie. use the tests currently run only by using
AC_PROG_CC_C89), but to add a AC_PROG_CC_K_R macro that will allow the
rarer programs that still want to support K&R compilers in their configure
script (or put another way: right now, we default to K&R but support C89
or C99; with the addition of new macro, we default to C89 but support K&R
or C99).  But no one has written a patch to do this yet.

- --
Don't work too hard, make some time for fun as well!

Eric Blake             address@hidden
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkh5OZ4ACgkQ84KuGfSFAYCw+QCeMAcKDJ3wPVDZ5VCaBLA1xWXK
pOMAoLGdETexQVPkmL7DvE8n26Nr7+r/
=Ji3w
-----END PGP SIGNATURE-----