to XFAIL or not to XFAIL

[Chris G. Demetriou]

dj@delorie.com (DJ Delorie) writes:
> > We have UNSUPPORTED and UNTESTED for `impossible to fix on this
> > target' and `impossible to test on this target' respectively.
> 
> The dejagnu documentation disagrees.  It lists UNSUPPORTED as
> impossible (or meaningless) to test on this target.  UNTESTED is given
> by a placeholder for a missing or incomplete test.
>
> However, the docs agree that XFAIL means "bug, but expected", and
> indicate that these would signal XPASS's when the bug is finally
> fixed.
> 
> Summary: http://www.delorie.com/gnu/docs/dejagnu/dejagnu_9.html

Based on this, i'd expect that tests which need to target_compile
things but cannot should report "UNSUPPORTED" rather than "UNTESTED".
(since "A test depends on a conditionally available feature [ability to
compile for target] that does not exist...")

I'd expect that e.g. the readelf tests for -s and -S (see my latest
patch, in http://sources.redhat.com/ml/binutils/2000-10/msg00065.html )
should in fact report "UNTESTED" rather than attempt the test (and
report XFAIL).  (since "A test case is not yet complete" -- in this
case, the existing test case doesn't address the additional sections
that those targets generate.)


Does that seem on target?  (If so, well, several of the existing
binutils/gas/ld tests are ... somewhat broken w.r.t. these
definitions.  8-)


chris