Checking in tests that fail

[Benson Margulies]

Do you have a policy for checking in tests that demonstrate a defect?
When diagnosing a problem, one can sometimes end up building up a
significant piece of code that will (with any luck) be a regression test
later. The same person who makes the test may not make the eventual fix.
It might make sense to check these in but skip them in the default
build.