Use efficient indexed Sum from Oleg's Freer Monads paper about generics-mrsop

[arianvp]

Given this a second thought today. Possible reason why it's still slow is because a lot of the code still uses Constr instead of Constr' and we're calling fromConstr and toConstr' in a lot of places. This wil mean we're still pattern matching on N-ary sums directly (just in some other ismorphic form that also requires the compiler to type-check exhaustiveness?)

Perhaps we should rewrite match and inj to use Constr' directly; instead of Constr

edit: Oh wait, but just compiling from and to is already slow... so that can't be it :/

[serras]

There is also the package freer-simple, which seems better maintained, in case you want to add a dependency on a package.

[VictorCMiraldo]

Branch victor-efficient-NS compiles just fine with

data NS :: (k -> *) -> [k] -> * where
  NS :: (IsNat n) => Constr' xs n -> p (Lkup n xs) -> NS p xs

newtype Constr' (xs :: [k]) (n :: Nat)
= Constr' { unConstr' :: Word }  

But this does not fix the memory usage issue. ☹️