PR automake/450 not yet fixed (Was: FYI: don't explicitly trace...)

[Peter Breitenlohner]

On Tue, 1 Mar 2005, Alexandre Duret-Lutz wrote:

> I'm installing this on HEAD and branch-1-9.
> It turned out to be easier than I dreaded.
>
> 2005-03-01  Alexandre Duret-Lutz  <address@hidden>
>
>         Fix for PR automake/450:
>         * aclocal.in (trace_used_macros): Do not explicitly trace files
>         included by configure.ac.
>         * tests/acloca19.test: New file.
>         * tests/Makefile.am (TESTS): Add acloca19.test.
>         Report from Peter Breitenlohner.
>
> Index: aclocal.in
> ===================================================================
> RCS file: /cvs/automake/automake/aclocal.in,v
> retrieving revision 1.104.2.4
> diff -u -r1.104.2.4 aclocal.in
> --- aclocal.in  5 Jan 2005 19:04:21 -0000       1.104.2.4
> +++ aclocal.in  1 Mar 2005 22:06:20 -0000
> @@ -418,6 +418,8 @@
> {
>   my %files = map { $map{$_} => 1 } keys %macro_seen;
>   $files{'acinclude.m4'} = 1 if -f 'acinclude.m4';
> +  # File included by $configure_ac are redundant.
> +  $files{$configure_ac} = 1;
>   %files = strip_redundant_includes %files;
>   # configure.ac is implicitly included.
>   delete $files{$configure_ac};
> Index: tests/Makefile.am
> ............

Hi,

I have applied the above patch to our installed version of aclocal
(from automake-1.9.5), and this certainly solves part of the problem,
but unfortunately not all of it.

Aclocal doesn't choke anymore on "case1" of the tests I sent you recently;
the generated aclocal.m4 contains, however, a line
        m4_include([common.ac])
and thus a subsequent autoconf (2.59) now chokes - with the same error
message as before.

I have simulated running your new test case "acloca19.test" without any such
problems; thus acloca19 misses part of the problems of case1.

Changing the test as follows:

cat >configure.in <<'END'
AC_INIT([acloca19], [1.0])
m4_include([aconfig.ac])
FOO
AC_OUTPUT
END

cat >aconfig.ac <<'END'
AM_INIT_AUTOMAKE
AC_DEFUN([FOO], [# bar
])
sinclude([bconfig.ac])
END

demonstrates this.

regards
Peter Breitenlohner <address@hidden>