[Checkers] r1953 regression

[Matt Papi]

One of the changes in r1953 broke a NonNull test and 2 IGJ tests. I got 
all of the tests to pass again and committed a change, but there's no 
test for whatever r1953 fixed so I can't be sure. Mahmood, could you 
verify that r1956 was okay?

Friendly reminder: be sure to run "ant all-tests" before committing. I 
don't mean to nag, but it's really much easier to verify that my 
outstanding changes don't break things if there aren't already test 
failures.

- Matt