arktype•2w ago

Type instantiation is excessively deep and possibly infinite

const A = type("bigint").narrow(() => true).pipe((v) => v.toString());
type A = typeof A.in.infer;

const A = type("bigint").narrow(() => true).pipe((v) => v.toString());
type A = typeof A.in.infer;

Produces error:

Type instantiation is excessively deep and possibly infinite.

Type instantiation is excessively deep and possibly infinite.

28 Replies

ssalbdivad•2w ago

Do you know if this is a regression/when it worked? This should be an easy one to fix at least. If it is only infinite when one side of the type is extracted, it's a bug in distill

SynthLuvr•2w ago

I'm pretty sure it worked in the past but I don't know when

ssalbdivad•2w ago

Yeah I made some changes recently to distill I'll fix this right away though Hmm there's really a whole set of narrowed inference that will fail if you infer out of a morph first which forces constrain<X, Y> to evaluate This is why I originally wanted constrain to not be a subtype of the original type but at some point it seemed like I could extract what I needed so I went with it. Once you have an intersection like string & {[constrained]: constraints} it is not possible to get back to string without enumerating all the possible builtins I want to check for which kinda sucks

PIat•2w ago

What would an alternative way be?

ssalbdivad•2w ago

Just have the constrained types be a tuple or something like [base, "constrain", constrained] But then they couldn't be used directly Not that they ever really are at the moment Morph types already work that way- they can't be used directly to represent the actual data

PIat•2w ago

So right now it's more of future-proofing?

ssalbdivad•2w ago

Well for now the branded types are always assignable to the base type I will keep it that way, I found some nonsense I can do to work around it Since obviously that is ideal But there's not really a builtin API for actually using the branded output types by default yet Maybe would be a global config at some point so you can choose if you want branded types But having the information encoded in the type opens up a lot of other possibilities also. The original reason I added it was to support pattern matching

PIat•2w ago

What do you mean by nonsense and how come it's ideal if it's nonsense?

ssalbdivad•2w ago

Because the external inference is ideal, but internally it requires a lot of hoop-jumping

PIat•2w ago

I see! You're jumping all the hoops so that we have ideal inference 🤗

ssalbdivad•2w ago

Well that is essentially the entire premise of arktype 😛

PIat•2w ago

And man is the inference ideal

ssalbdivad•2w ago

Well I'm fixing some cases where it was not now haha