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

[Ralf Wildenhues]

* Stefano Lattarini wrote on Sun, May 23, 2010 at 03:01:20PM CEST:
> At Sunday 23 May 2010, Ralf Wildenhues wrote:
> > Got lucky and intuition saved me from regressing branch-1.11 with
> >  this one.  I'm applying the fix below on top of this patch and
> >  merging to master.  The nontrivial point to note is that in the
> >  wrapper script, srcdir needs to be expanded at testsuite run time
> >  only, because it may differ from the srcdir value in the Automake
> >  developer tree.
> Good catch.  Note that my patch on the sepration of instspc.test has 
> this same problem too.

OK, good to know.  I haven't looked at the instspc patches yet.

> However, I think that a better way to resolve this problem would be to 
> make @abs_srcdir@ substituted in defs.in (dropping the computation of 
> $srcdir), and make defs.in robust w.r.t. multiple inclusion; then the 
> generated test files could do simply:
> 
> #!/bin/sh
> parallel_tests=yes
> . ./defs.in
> . $testsrcdir/INPUT.test

Hmm, yes, maybe.  This needs quoting for spaces in abs_srcdir.
Also, doesn't it remove the automatic 'set -x' when we detect
scripts run manually?

> > diff --git a/tests/Makefile.am b/tests/Makefile.am
> > index cff34c5..bbeb2cb 100644
> > --- a/tests/Makefile.am
> > +++ b/tests/Makefile.am
> > @@ -34,7 +34,11 @@ $(parallel_tests): Makefile.am
> >     { echo '#!/bin/sh'; \
> >       echo '# DO NOT EDIT! GENERATED AUTOMATICALLY!'; \
> >       echo 'parallel_tests=yes'; \
> > -     echo ". '$(srcdir)/$$input'"; \
> > +     echo "if test -f ./$$input; then"; \
> > +     echo "  . ./$$input"; \
> > +     echo 'else'; \
> > +     echo "  . \$$srcdir/$$input"; \
> ... this prevent the test script from working when run from the 
> command line in a VPATH build -- with an error like:
>   ./check-p.test: line 7: /check.test: No such file or directory

Yeah, that's bad.  One alternative would be to generate *-p.test in the
source tree right away; after all, 'make dist' will have them end up
there in an extracted tarball anyway.

Cheers,
Ralf