Feature request: Vacuous/error constraint (related to 7.7 closed type families regression)

I was trying to fix one of my closed type families examples for the new syntax, and run into an unfortunate issue. Due to a kind error in my code one of my constraints was being silently discarded (see ticket: https://ghc.haskell.org/trac/ghc/ticket/8669) The main reason I resorted to this (in hindsight) ill-kinded hack is the lack of a vacuous (i.e. never holding) constraint to produce type errors. I would love to be able to explicitly force GHC into realising "this constraint can never hold", if I had the ability to pass along a String/Symbol to be reported to the user, that would be even better. Cheers, Merijn