Declaration engine

[emilyaherbert]

rendered

prototype

[sezna]

If the function has to be typed when we insert it into the declaration, doesn't that mean we maintain the status quo of having to type check (and perform inference on) everything before beginning this process?

[sezna]

Sooo excited for work on this to begin 🚀

[otrho]

I don't think you've addressed out of order declarations and dependencies here? Is that another thing the decl engine was supposed to handle?

Allowing out of order declarations and to remove the declaration sorting we have now and especially to allow recursive data structures and function calls is highly desirable.

[mitchmindtree]

Nice one Emily!

I don't have much to add on top of the others' comments, though I think your insight about our current lack of clarity on how to perform type inference in a consistent manner being a core issue is a really key insight and makes a lot of sense! I can see the proposed DeclarationEngine going a long way to help with this.

The described approach reminds me a little of the Hindley-Milner type system and its approach to inference, though maybe a little less formal and jargon-y 🥲 Considering it was the solution adopted by ML, Haskell, Rust and I'm sure many other languages, perhaps it's worth taking a closer look and seeing if there's anything we can learn from it as "Prior Art"? I'd imagine it might have some key design insights particularly when it comes to handling trait (typeclass) constraints and polymorphic types (Vec<T>, (T, U), etc).

[sezna]

The described approach reminds me a little of the Hindley-Milner type system and its approach to inference,

I actually based our current type engine on Hindley-Milner when I wrote it, hence the function names like unify etc. We are already using the HM algorithms to perform unification and inference.

[emilyaherbert]

There hasn't been much activity on this RFC page, largely because I have been working on this prototype: https://github.com/emilyaherbert/declaration-engine-and-collection-context-demo/tree/master/de. The prototype should 1) help explain the idea and 2) acts as a guiding light for our implementation. Trait constraints for generic functions are working in the prototype.

Also, I've basically reworked the entire RFC, so I'd love a second round of feedback. In the meantime, I will hack on a experimental branch in the Sway repo so that we can get things moving. Don't let this scare you away from iterating on the design though!

[emilyaherbert]

ping @FuelLabs/sway-compiler would we be able to merge this?

[sezna]

🚀