I have type X indexed by kind S with several functions that work on X. For example, f converts X S1 to X S2 (it doesn't use X S1 in this simplified example, though).
{-# LANGUAGE DataKinds, GADTs, TemplateHaskell, TypeFamilies #-}
import Data.Singletons
import Data.Singletons.Sigma
import Data.Singletons.TH
singletons [d|
data S = S1 | S2 | S3 | S4
|]
data X (s :: S) = X
f :: X S1 -> X S2
f x = X
Now I'd like to define functions that may return X S2 or X S3 depending on its argument. The straightforward way would be using Either.
g1 :: Bool -> X S1 -> Either (X S2) (X S3)
g1 True x = Left X
g1 False x = Right X
But I don't want to take this approach because I need to have nested Eithers when a function returns more types.
Another approach would be using Sigma like this.
g2 :: Bool -> X S1 -> Sigma S (TyCon X)
g2 True x = SS2 :&: X
g2 False x = SS3 :&: X
But this doesn't express the idea that g2 returns only X S2 or X S3. I can express this idea by introducing a wrapper around X.
data Y (s :: S) where
Y2 :: X S2 -> Y S2
Y3 :: X S3 -> Y S3
singletons [d|
type Z s = Y s
|]
g3 :: Bool -> X S1 -> Sigma S ZSym0
g3 True x = SS2 :&: Y2 X
g3 False x = SS3 :&: Y3 X
But it's cumbersome to define these wrappers for each combination and unwrap them on caller sites. It'd be nice if I can directly restrict types using g2 approach, for example, by like applying type constraints, but I'm not sure how I can do it.
How can I directly restrict types using g2 approach?
I'm using GHC 8.8.4 with singletons-2.6.
This question looks too simplified and contrived; it would be nice to have some more concrete motivation. But here is a shot in the dark.
We can define a variant of
Sigmawith a predicate on the first component:Some code to define the predicate seems unavoidable:
But now we can just use
X: