Re: [autoconf 2.69c] test 36: autoupdating AC_OBSOLETE fails when configure reads config.site

[Zack Weinberg]

On Wed, Oct 28, 2020 at 12:45 PM Jannick <thirdedition@gmx.net> wrote:
>
> On Tue, 27 Oct 2020 13:59:21 +0100, Jannick wrote:
> > If the system makes configure read in a config.site script, then
> >
> > 36: autoupdating AC_OBSOLETE                        FAILED (tools.at:1342)
> >
> > because the comparison of config.log is not prepared to see the additional
> > line in config.log
> >
> >  configure: loading site script /etc/config.site
> >
> > Here the relevant output snip when running " make check
> > TESTSUITEFLAGS='-x -d -v 36' " on my system with config.site:
> > [...]
> > ++ ./configure
> > --- -   2020-10-27 13:42:15.389259100 +0100
> > +++ /c/autotools/autoconf/tests/testsuite.dir/at-groups/36/stdout
> > 2020-10-27 13:42:15.288410800 +0100
> > @@ -1,3 +1,4 @@
> > +configure: loading site script /etc/config.site
> >  dst=dst src=src
> >  configure: creating ./config.status
>
> Far cleaner solution: simply pass CONFIG_SITE=\"\" to the first argument of
> AT_CHECK_CONFIGURE and leave configure_options alone.

I'm 95% sure the testsuite should block external config.site settings
from being visible to *any* of the configure runs it does.  The only
reason I can think of, why it might *not* be appropriate to make that
change, is if there's some unusual environment where a bunch of
config.site settings are necessary to make the C compiler work at all,
and we could tell people using those environments to provide a wrapper
script instead.  However, we can't do this unconditionally in
AT_CHECK_CONFIGURE, because some tests set CONFIG_SITE themselves and
expect AT_CHECK_CONFIGURE to pick it up.  I'll think about how to do
it properly.

zw