Fails Generating VHDL When Typeclass Has Many Instances

[exclusive-and]

The compiler will fail to generate an output port on noinlined functions when a typeclass has many specialized instances.
It's suspected that the compiler skips generating an output port if the output of a function is one of many instances. Without the output port, the Netlist generator thinks the function is being applied to too many I/O components and throws an UnderApplied error.

An example is given in the attached file (this will fail to generate, and dump a long Core message).
clash-fix.zip

[christiaanb]

What's causing the error is a bug in Clash' implementation of type family reduction, specifically, it doesn't handle nested type families. So given this even smaller reproducer:

module T1721 where

import Clash.Prelude

type family G a
type instance G () = Bool

type family F a

data T a = T a
data TP a = TP Bool

type instance F (T Bool) = TP Bool

f :: F (T (G ()))
  -> F (T (G ()))
f = id
{-# NOINLINE f #-}

topEntity :: F (T (G ()))
          -> F (T (G ()))
topEntity = f

Clash doesn't think there's any matching instance for F (T (G ())) because G () isn't reduced to Bool before doing the instance lookup for F on T (G ()). In turn, this makes Clash not finding a bit-encoding for F (T (G ())).