selftest on win32

[Martin Pool]

On Mon, Nov 10, 2008 at 3:59 PM, Mark Hammond <mhammond at skippinet.com.au> wrote:
>> I think it would be a small step forward to turn these into expected
>> failures, pointing at the bug number describing the problem.  (There
>> may already be one.)
>
> Agreed - but it's not clear to me how to make this actually happen - I doubt
> we want to touch each of the tests individually (although there are 55
> failures I'm guessing some of them are due to the 'test multiplication'
> process...) and there doesn't seem an obvious single place to otherwise
> handle this.
>
> Maybe the test suite itself could grow a list of exceptions it automatically
> converts to expected failures?  win32 could have the lock contention
> exception listed, and although this introduces a risk that *new* lock
> contention failures would go unreported, it would still be an improvement
> IMO.
>
> Is this desirable?  Anything I've missed or other alternatives?

One other thing we mentioned in this area is to add a list of tests
expected to fail with particular errors, or maybe globs that can match
against them.  That might be reasonable if all the Windows failures
are in a particular module, but I suspect they're not, or they're in
something too broad to be meaningful.

If we want to treat the error as known globally maybe it'd be better
to also match against the string form of the exception so we could at
least say "LockContention........./dirstate" is expected.

-- 
Martin <http://launchpad.net/~mbp/>