Question about subtyping

Hello, I’m studying Grace and curious about the way it handles type inference with subtyping.

So consider this case which TYPES ok:

; cabal run -- grace interpret --annotate - <<< '[{answer: null}, {answer: 42}]'
[ { "answer": null }, { "answer": 42 } ] : List { answer: Optional Natural }

and this which FAILS:

; cabal run -- grace interpret --annotate - <<< '[{answer: 42}, {answer: null}]'
Not a subtype
...

So the depending on the order of elements in the list it either succeeds or fails which can be then fixed (suggested by the README) by adding a type annotation with an existential List ?:

; cabal run -- grace interpret --annotate - <<< '[{answer: 42}, {answer: null}] : List ?'
[ { "answer": 42 }, { "answer": null } ] : List ?

I think this can be a bit confusing to users (that in those two similar cases they either required or not to add type annotations). Do you think it’s worth fixing and/or possible to fix by synthesising List ? type for List constructor?

[Gabriella439]

I don’t think higher rank polymorphism is important for a configuration language. The main reason I went with the bidirectional type-checking algorithm is because out of all the algorithms I tried it seemed the easiest to extend with new features

[andreypopp]

I don’t think this is fixable short of overhauling the type checker to use something other than bidirectional type inference.

Just to clarify — the “Local Type Inference” paper uses bidirectional type checking (in fact I’ve seen many cite it as the first paper describing it, though they themselves say that this is folklore).

But nevertheless, you’re right, the change is probably not so trivial and I’m not sure how well it can be combined with polymorphic subtyping which Grace has now. So this issue can probably be closed.

By the way — do you think higher rank polymorphism is important for a configuration language (I don’t have a good intuition where it can be used)?