This is the mail archive of the
binutils@sources.redhat.com
mailing list for the binutils project.
Re: gas testsuite approch
- From: Hans-Peter Nilsson <hp at bitrange dot com>
- To: Alan Modra <amodra at bigpond dot net dot au>
- Cc: "Svein E. Seldal" <Svein dot Seldal at solidas dot com>, binutils <binutils at sources dot redhat dot com>
- Date: Mon, 16 Sep 2002 22:16:35 -0400 (EDT)
- Subject: Re: gas testsuite approch
On Tue, 17 Sep 2002, Alan Modra wrote:
> On Mon, Sep 16, 2002 at 04:35:41PM -0400, Hans-Peter Nilsson wrote:
> > On Mon, 16 Sep 2002, Svein E. Seldal wrote:
> > > But when I look at the existing tests, some target excludes the test
> > > entirely, while others uses XFAIL. Take the c30 target as an example. It
> > > does not xfail it - it simply skips the test. Which method should I use?
> >
> > Wrong choice of words on my part, but the C30 example still
> > fits. Skipping is better than xfail in this case: an xfail is
> > supposed to be a known failure and signalling a failure would be
> > wrong here.
>
> Hmm, I prefer xfail, simply because an XPASS result alerts you to
> something unusual. In this case, an XPASS would indicate a failure!
If you mean adding xfailing test-cases for targets that are not
designed to pass the test, for the odd case that there'd be a
bug that makes the test pass, I can't agree that'd make sense;
that's not what xfail is for. (Testing for error is supposed to
be done in other ways.)
Of course I agree that getting an xpass for an xfailing test you
didn't intend to fix should lead to further investigation. In
some case the cause for the xpass might be a general rewrite
that actually fixed the bug. ;-)
brgds, H-P