Re: [BIKESHEDDING PATCH] Generated tests are now just a thin layer around other tests.

[Ralf Wildenhues]

* Stefano Lattarini wrote on Sun, Jun 06, 2010 at 10:00:23PM CEST:
> At Sunday 06 June 2010, Ralf Wildenhues wrote:
> > 
> > The patch is still not right, so I'm not pushing it.  A generated
> >  test foo-p.test now needs to depend upon foo.test,
> >  this is not reflected in the makefile.
> Right  :-(

And really I'd prefer if foo-p.test would not depend upon foo.test.

> Not however that the older makefile didn't have a much better handling
> of tests' dependencies.

Well, we came from non-parallel tests, where recheck and RECHECK_LOGS
did not exist, so all you could do to limit testings was to set TESTS
explicitly.  There, you set check_SCRIPTS/check_PROGRAMS for a global
set of dependencies and would be done with it.  The Automake package
still reflects this to some point.  And really, the check_SCRIPTS are
in a way global dependencies to all tests.

> Maybe the pending patch of mine "Improve
> declaration of dependencies in the testsuite" might help here, if
> properly extended.  For reference:
>  <http://lists.gnu.org/archive/html/automake-patches/2010-05/msg00012.html>

This looks interesting.  Lemme see ...

> > Either we rewrite defs.in so that it is idempotent and sourced
> > twice,
> Mhh... that would entail a rethinking of the whole 'required=...' stuff.
> Which wouldn't be bad in the long run, but we should think thoroughly
> before venturing down that road, to get the interface right this time.

Well, the *-p.test could set something like am_skip_defs which then
would pretty much set testsrcdir and curdir only, and then unset
am_skip_defs, so the defs sourcing of *.test is full, WDYT?

Cheers,
Ralf