test with bad group still leads to version creation, except missing data

[Alex-Jordan]

If you have a test named myTest, where a problem is defined like group:mySet. But suppose there is no set named mySet. You will get an error screen about no problems being available in mySet. (And maybe you would get the same message if mySet exists but has zero problems in it.)

OK, fine. But your user will still get assigned to mySet,v1. And mySet,v1 has been partly created, but not completely. After navigating away from the above error screen, a user can try to go to version 1 of the test. Now they will get a different error message complaining about the set not have a close date. But the real issue is that lots of things about the version were not initialized. The initialization process was aborted midway without cleaning up after itself.

[drgrice1]

@glarose says:

	# sanity check: if the group set hasn't been defined or doesn't
	# actually contain problems (oops), then we can't very well assign
	# this problem to the user.  we could go on and assign all other
	# problems, but that results in a partial set.  so we die here if
	# this happens.  philosophically we're requiring that the instructor
	# set up the sets correctly or have to deal with the carnage after-
	# wards.  I'm not sure that this is the best long-term solution.
	# FIXME: this means that we may have created a set version that
	# doesn't have any problems.  this is bad.  but it's hard to see
	# where else to deal with it---fixing the problem requires checking
	# at the set version-creation level that all the problems in the
	# set are well defined.  FIXME

That comment was added to the code in 2005 with a commit message of "Gateway update: deal with missing problem groups in gateway tests slightly more gracefully."