Disclaimer
GADTs & DataKinds are unexplored territory for me, so some of the limitations and capabilities of them are unknown to me.
The Question
So I'm writing an AST for a JavaScript code emitter, and I've identified one edge case between expressions and that is that they can either be a reference or not. So I've used GADTS and datakinds to type this aspect of JavaScript expression semantics. The ast looks something like this.
Subset of the expression AST
-- at the moment I'm just using a bool to identify if the expression
-- behaves as an reference, but I'll probably change it due to the fact
-- a bool is pretty vague
data JSExp :: Bool -> * where
JSNumber :: Double -> JSExp False
JSBool :: Bool -> JSExp False
JSReference :: Text -> JSExp True
JSProperty :: JSExp a -> Text -> JSExp True
JSAssign :: JSExp True -> JSExp b -> JSExp b
This looks all fine and dandy, because an assignment expression requires the first expression to be a reference like a property expression ("test".shadyProperty
) or a reference/identifier.
The problem
Now I want to add an array literal expression, in JavaScript it shouldn't matter what is in this list so lists like this are legal
[a, 1, true, a.b]
In my AST however this is not legal because there are multiple types in the list
data JSExp :: Bool -> * where
-- ...
JSArray :: [JSExp ???] -> JSExp False
let aref = JSReference "a"
in JSArray [aref, JSNumber 2, JSBool True, JSProp aref "b"]
What goes in the place of ???
for the type? A similar problem occurs when I want to build out a constructor for JSObject's and JSFunctionCall's, as these are also expressions.
Idris's soultion
In Idris ???
would look something like this.
data JSExp : Bool -> Type where
JSArray : List (JSExp _) -> JSExp False
JSNumber : Float -> JSExp False
JSBool : Bool -> JSExp False
-- ...
Potenial soultions
Wrapping type
One soultion that isn't like the Idris one, would to have a wrapper type like this
data JSExpWrap = Refs (JSExp True) | NoRef (JSExp False)
This would make the api of my library gross and it's not what I'm looking for.
In Summary
I'm looking for the equivalent that's found in Idris, and an explanation of the implications of the solution. If there's no equivalent, a solution of the next best thing is what I'm looking for.