Using bugzilla to track XFAILs

[Zack Weinberg]

Richard Earnshaw <rearnsha@arm.com> writes:

> I like the idea also, but we need to make sure we do it carefully.  Often 
> a test is xfailed when it isn't really a problem with the compiler but 
> with the run-time environment.  Take, for example, 
> g77.f-torture/execute/io0.f, which is xfailed on several embedded 
> platforms because there is no support for stat() in these environments.
>
> Ideally such tests should be skipped when the compiler is not at fault.

Yes, or marked UNSUPPORTED.  dg.exp doesn't appear to have a facility
for UNSUPPORTED, but it can definitely skip tests, and it oughtn't to
be hard to add UNSUPPORTED capability in gcc-dg.exp.

> Another possibility might be to use the KFAIL feature introduced into 
> dejagnu for gdb.  A kfail (known failure) could require a PR to be 
> associated with it.

I don't like KFAIL.  I think it is introducing a distinction without a
difference, and I don't want to have to upgrade dejagnu on a whole
pile of heterogeneous computers.

zw