This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [RFC] list of tests in makefile.
- From: "Joseph S. Myers" <joseph at codesourcery dot com>
- To: Florian Weimer <fweimer at redhat dot com>
- Cc: <libc-alpha at sourceware dot org>
- Date: Fri, 31 May 2013 16:42:44 +0000
- Subject: Re: [RFC] list of tests in makefile.
- References: <20130528134406 dot GA11989 at domone dot kolej dot mff dot cuni dot cz> <20130528173923 dot 567052C06E at topped-with-meat dot com> <51A50707 dot 1050102 at redhat dot com> <20130528193711 dot F08F02C073 at topped-with-meat dot com> <51A5F706 dot 4000305 at redhat dot com>
On Wed, 29 May 2013, Florian Weimer wrote:
> On 05/28/2013 09:37 PM, Roland McGrath wrote:
> > > Agreed. Though I think a data-driven approach might also be acceptable
> > > e.g. the list of tests is in a file e.g. nptl.tests, libc.tests etc.
> >
> > Sure. Lists in the source tree, not wildcards.
>
> And we could add a script that could generate them from "git ls-files"
> (auto-resolving conflicts).
Note that I think lists of tests should be able to include metadata rather
than just a list of test names; see
<http://sourceware.org/ml/libc-alpha/2012-09/msg00276.html>.
I don't see any advantage to moving the list out of the makefiles. And
since there are plenty of existing cases where the list of tests has
dependencies on lots of makefile conditionals, metadata support would need
to come first. Even then, it's not obvious that each makefile variable
used to condition tests should have its own property name for metadata.
nptl/Makefile, for example, has conditions on $(have-forced-unwind),
$(build-shared), $(have-z-execstack). Maybe rather than separate property
names, it's better to have "enabled-yes" and "enabled-no" properties
(enabled-no means the test is disabled, enabled-yes is ignored), so the
tests lists could include e.g.
tst-execstack:enabled-$(build-shared):enabled-$(have-z-execstack) - which
of course indicates keeping the list in the makefiles, so that the
makefile variables get properly interpreted in the tests setting.
--
Joseph S. Myers
joseph@codesourcery.com