[FEATURE] Generics

[vladceresna]

I think it is also very important thing to have. One of patterns, where it can be useful, is my previous issue about Rust-like Error handling #1209

And thanks for your work and awesome language!

[Tuhkis]

Wren is a dynamically typed language. I feel like generics would have little to no benefit in Wren. I would even go as far as to say they'd turn out to be more of a hindrance.

In your other PR you seem to have come up with some syntax to turn when into a strongly typed language (?). I feel like all that would go against the design of Wren, which, despite having a fixed object layout, is very obviously a dynamically typed language.

[vladceresna]

@Tuhkis Can you explain what is wrong with strict typing? But still, even if we are not going to do strict typing, it would be good to at least have an optionally enabled syntax that allows writing typed code, which would increase the stability of the code itself on Wren and allow checking for a bunch of different errors at the compilation stage. Right now, because of these errors, your code can just crash in the middle. Which is pretty unstable behavior. Generics, based on my experience with Rust, are very useful for good error handling and sometimes greatly improve the overall look and readability of the syntax, especially when working with nested types such as lists.

[vladceresna]

Or there is an option that we could write something like TypeWren language separately, which would be compiled into Wren. But this is just a fantasy, just a thought.

[Tuhkis]

To clear up: I think that strict typing is a great feature in the languages that use it. Wren, however, is not designed to be one of them.

Additionally the compiler and the VM are designed around Wren being dynamically typed. For example, the compiler is single-pass which makes a lot of static analysis that strict typing would enable unnecessarily difficult or outright impossible. Wren is meant to be small and simple and implementing a robust type system, in my opinion, would go against that design.

In a dynamic language I see very little use for generics since you can assign more-or-less any value to more-or-less anything and you can do this kind of introspection: if (myEquestrian is Unicorn) { ... }.

[HallofFamer]

I think you severely underestimate the level of effort required to implement generics even for a statically or optionally typed language. I've been developing my own version of optionally typed Lox(the original version was very similar to Wren's compiler design), and there was quite a bit of ground work that needed to happen even before generics may be implemented. You would need to convert the single-pass compiler to a multi-pass compiler which generates AST from parser, and walking the AST to resolve symbols and perform typechecking on simple types. You also have to figure out how to represent type information in a compiler, and how to manage the memory of type information. Finally, you are ready for generics, and the road ahead only gets harder.

In my experience, it took almost 12 months to just redesign the compiler pipeline and add an optional static type checker, and then another 6-12 months to implement generics. If you have this amount of free time and skills, why not try implementing this yourself? The wren maintainers are volunteers and I aint sure anyone of them can afford 18-24 months of their time to work on such a large-scale feature.