If we don't seem to have an ANSI compiler, then omit a warning as soon
as that is discovered, because it is likely to break later configure
tests.  This doesn't seem to catch the particular HP-UX compiler I was
after, which is non-ANSI but only emits a warning on this configure
test.  Nevertheless probably better to have it in.

your platform seems to support ipv6, but actually it breaks.  This
seems to be the case for "powerpc-apple-darwin1.4".

the '#'.  Off to the Implant Office with you!

about.

sticky/setuid/setgid bits the same way as GNU ls.

reproducible results.

can be reused in tls.

Build getaddr/nameinfo into lib/.

Split code to generate "rwx-----" strings into lib/permstring.c so it
can be reused in tls.

platforms that don't have them.

--copy-links.  The readlink_stat() does need to be done in the normal case
before checking the exclude patterns because it needs to know whether or
not a file is a directory in order to properly handle a trailing slash
in an exclude pattern.  This fix makes make_file() go ahead and call
readlink_stat() but then if the latter returns an ENOENT and copy_links is
on then it will only print an error if the path is not excluded.

opening an inbound socket, we might get several address results,
e.g. for the machine's ipv4 and ipv6 name.

If binding a wildcard, then any one of them should do.  If an address
was specified but it's insufficiently specific then that's not our
fault.

However, some of the advertized addresses may not work because e.g. we
don't have IPv6 support in the kernel.  In that case go on and try all
addresses until one succeeds.