We thought that Jim Apple's (rather complicated) test
'gadt/termination' was failing because of the move from "left/right"
in coercions to "Nth"; see the comments in Trac #1999.

However it turns that the test was failing because of an outright bug
in the 7.2 type checker, fixed by Changeset
4ed2634b.  (This bug meant that type
synonyms in given constraints weren't being expanded properly.
See test typecheck/should_compile/GivenTypeSynonym.)

So now this bug is fixed, Jim's program compiles again!  (We were
too quick to assume the problem was with the change to FC.)