Date: Thu, 07 Mar 2013 20:28:12 +0100 From: Dimitry Andric <dim@FreeBSD.org> To: Tijl Coosemans <tijl@coosemans.org>, toolchain@freebsd.org Subject: Re: c89 broken on head? Message-ID: <5138EA4C.1060001@FreeBSD.org> In-Reply-To: <5138CD6B.2050309@coosemans.org> References: <5138CD6B.2050309@coosemans.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 2013-03-07 18:24, Tijl Coosemans wrote:
> Whatever the command line arguments, running c89 almost always results in
> the following output. Anyone else seeing this?
>
> c89: illegal option -- 1
> usage: c89 [-cEgOs] [-D name[=value]] ... [-I directory] ... [-L directory] ...
> [-o outfile] [-U name] ... operand ...
>
> where operand is one or more of file.c, file.o, file.a
> or -llibrary
Does anybody ever actually use this tool, really? :-)
In any case, what happens is that /usr/bin/c89 builds up an argv[]
array, prepending the flags "-std=iso9899:199409" and "-pedantic" to the
other arguments, but it sets argv[0] to "/usr/bin/c89" too.
If /usr/bin/cc is gcc, this causes no problems, since gcc always runs
/usr/libexec/cc1 for its first stage compilation process. It basically
ignores the value of argv[0].
When /usr/bin/cc is clang, however, it uses argv[0] to run its first
stage compilation, with -cc1 as the first argument. So this will run
/usr/bin/c89 yet again, and that complains about the unrecognized '1'
option.
It can be solved very easily, by letting c89.c set argv[0] to
/usr/bin/cc instead, similar to c99.c, as with this diff:
Index: usr.bin/c89/c89.c
===================================================================
--- usr.bin/c89/c89.c (revision 247448)
+++ usr.bin/c89/c89.c (working copy)
@@ -72,7 +72,7 @@ main(int argc, char **argv)
Argv.a = malloc((argc + 1 + N_ARGS_PREPENDED) * sizeof *Argv.a);
if (Argv.a == NULL)
err(1, "malloc");
- Argv.a[Argc++] = argv[0];
+ Argv.a[Argc++] = CC;
for (j = 0; j < N_ARGS_PREPENDED; ++j)
Argv.a[Argc++] = args_prepended[j];
while ((i = getopt(argc, argv, "cD:EgI:l:L:o:OsU:")) != -1) {
> Also, I seem to remember a discussion about making -std=gnu89 the default
> for clang when run as "cc", but nothing seems to have changed. Could this
> be picked up again, because there are in fact subtle semantic differences
> between gnu89 inline and c99 inline that old code may rely on.
Why on earth would you want gnu89 still as the default in 2013? I would
rather have it default to C11, but the support for this isn't complete
yet... :-)
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?5138EA4C.1060001>