Note [Superclasses and satisfiability]

GHC/Tc/Solver.hs:734 compiler 2 tickets

Expand superclasses before starting, because (Int ~ Bool), has
(Int ~~ Bool) as a superclass, which in turn has (Int ~N# Bool)
as a superclass, and it's the latter that is insoluble.  See
Note [The equality types story] in GHC.Builtin.Types.Prim.

If we fail to prove unsatisfiability we (arbitrarily) try just once to
find superclasses, using try_harder.  Reason: we might have a type
signature
   f :: F op (Implements push) => ..
where F is a type function.  This happened in #3972.

We could do more than once but we'd have to have /some/ limit: in the
the recursive case, we would go on forever in the common case where
the constraints /are/ satisfiable (#10592 comment:12!).

For straightforward situations without type functions the try_harder
step does nothing.

References 1

Referenced by 2