Coercion improvements

[zth]

Listing some ideas for improvements to coercion/subtyping.

• [ ] A variant can be coerced to a polyvariant if the polyvariant is closed and the runtime representation matches
• [ ] Investigate dicts - what can be coerced?
• [ ] Handle more coercion + unboxed variants. For example. "hello" :> JSON.t and 1. :> JSON.t works, but not 1 :> JSON.t, true :> JSON.t or dict{"test": "hello"} :> JSON.t

[nojaf]

A variant can be coerced to a polyvariant if the polyvariant is closed and the runtime representation matches

Not sure if this example falls under that, but it would be great if we could do the following:

let f = (v:s) => ()

type t = | Foo | Bar | Meh

let _ = f(Foo) // without writing `Foo :> string`

[zth]

I don't think that example is complete...? What is s?

[nojaf]

Oh, sorry s for string.

[zth]

Ah right. So you essentially mean implicit coercion. No, that's something we've intentionally decided against. We want any type coercion to be explicit. There are many reasons:

1. You might not actually want a type coercion in various contexts, since types have meaning depending on context
2. It needs to be clear from the code where coercion is happening
3. Coercion can come with some perf loss in the type checking if the types are complex, so doing it should be explicit

[nojaf]

Got it, thanks!