+Note [Printing KindedTyVars]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+Trac #3830 reminded me that we should really only print the kind
+signature on a KindedTyVar if the kind signature was put there by the
+programmer. During kind inference GHC now adds a PostTcKind to UserTyVars,
+rather than converting to KindedTyVars as before.
+
+(As it happens, the message in #3830 comes out a different way now,
+and the problem doesn't show up; but having the flag on a KindedTyVar
+seems like the Right Thing anyway.)
+