Re: [GNU Autoconf 2.67] testsuite: 199 246 failed

[Ralf Wildenhues]

Hi Eric,

* Eric Blake wrote on Wed, Aug 04, 2010 at 09:44:09PM CEST:
> On 08/04/2010 01:08 PM, Ralf Wildenhues wrote:
> > -# We cannot assume aclocal won't warn (aclocal-1.9 does not understand -W*
> > -# options), so check autoconf only.
> > -AT_CHECK([autoconf --force -Wno-syntax], 0, [ignore], [stderr])
> > +AT_CHECK([autoreconf -v --force -Wno-syntax], 0, [ignore], [stderr])
> 
> Doesn't autoreconf still want to run aclocal under the hood, even if you
> manually touched aclocal.m4?  The rest of the patch made sense, but I'm
> not sure about the logic of this hunk.

No, it shouldn't ever run aclocal in this setting:

  # Nevertheless, if aclocal.m4 exists and is not made by aclocal,
  # don't run aclocal.

  if (-f 'aclocal.m4')
    {
      my $aclocal_m4 = new Autom4te::XFile 'aclocal.m4';
      $_ = $aclocal_m4->getline;
      $uses_aclocal = 0
        unless defined ($_) && /generated.*by aclocal/;
    }


The key to running autoreconf however is that it might still do some
tracing internally, in addition to running plain autoconf, which is why
I think it is the right thing to exercise in this test.


We are still sorely lacking in coverage of the interplays between
autoreconf and all the tools it calls; one thing I'm having trouble
figuring out a good middle ground for is whether aclocal should pass
warning flags to autom4te it calls under the hood: a -Werror breaks a
few hundred Automake testsuite tests ... but enough of stuff off topic
for this thread.

Thanks for the quick review,
Ralf