Hacker News Reader

Comment by Zababa | Hacker News Reader

I'm curious about how your new language would work. From my (limited) understanding, the big differences between Haskell and SML are the type system and stictness/lazyness. I think a Haskell-like type system would be necessary if you want to eschew side effects (although I may be wrong), so the big difference would be that it's strict?

View on HN

The future is effect systems, not monads. The next ML will be more similar to Koka than Haskell, and yes, it'll be strict.

Maybe it'll even be named OCaml if the efforts to bring effect typing into it are eventually merged :-).

Flix might be this next ML. It is a new programming language with a Hindley-Milner type system notable for having effect polymorphism. It has an otherwise well-rounded design inspired by OCaml (among others).

https://flix.dev/

(I am not affiliated with Flix. I started following its development recently because it looked interesting.)

Flix looks interesting, the Java interop seem to be a great feature too, but I'm wondering if they'll have traction. The JVM already has Scala, and while it's not exactly the same thing it will be "good enough" for most people. Considering Scala has usually lots of churn and already a split ecosystem, plus the upcoming transition to 3, it's going to be hard to find a place. Especially since Java itself is gaining pattern matching, records and the like.

So if I understand everything correctly effect systems are another way of encoding side effects into the type system? If so, why are they preferable to monads? That's the part I don't understand. For the strictness, I think now most people agree that a bit of lazyness is great when you need it (iterators, streams) but the vast majority of the language should be indeed strict.

> Maybe it'll even be named OCaml if the efforts to bring effect typing into it are eventually merged :-).

From what I read recently that's the plan, although they'll focus first on domains for 5.0, and merge the effects later.

Effect systems are both a runtime thing (similar to exceptions or lisp's conditions, in some sense), and a type-level thing (have a set of effects attached to an arrow, with type inference that is able to infer the effects and compose them nicely). Both are necessary for the future ML™, I think: the runtime part gives semantics to effects and composes super well (better than monad transformers, afaict: no need to `lift` everything); the typing part keeps the code honest. The effects also have to compose nicely with higher-order functions (so that `map` carries over the effects of its functional parameter).

> From what I read recently that's the plan, although they'll focus first on domains for 5.0, and merge the effects later.

Yes indeed! The timeline for typed effects is not clear to me, it might take a long time, especially with the stringent constraints of backward compatibility.

That said I'm not an expert, I'm just very hyped :-)

Thank you for the explanation! I feel like I understand this better now.

> The timeline for typed effects is not clear to me, it might take a long time, especially with the stringent constraints of backward compatibility.

From what I understand, the "long time" will be probably less than modular implicits, which are estimated at at least a few years away. There are a few informations here https://discuss.ocaml.org/t/multicore-ocaml-may-2021/7990 and here https://discuss.ocaml.org/t/multicore-ocaml-june-2021/8134, where at least we can see that the effect branch is up to date with 4.12. The performance seems really good, I can't find it anymore but they had the same performance as Go's net/http with http/af. It's a very exciting space!

monadic IO is basically a type of effect system.

the big issue is that you have to do a LOT of complex shuffling of types and values in order to deal with effects.

You can actually avoid a ton of this by thinking about purity in a less expression-level way, like how Koka does it. And, I think I'd probably try implementikng something like it (or eff) in my language.

You can always replicate laziness by making your own thunks. But this just isn't as convenient.

Thanks for the explanation.

If you took SML and removed reference cells and the parts of the standard library that involve IO, then you'd have a pure language (depending on whether you consider exceptions to be effects). It would not be useful for application programming, but would be fine for teaching functional programming.

The Hope language would be a educational SML w/o side-effects.

https://en.wikipedia.org/wiki/Hope_(programming_language)

That's true. I think it also ties into best practices such as hexagonal architecture, DDD, and the "functional core, imperative shell" idea. You design the core of your program as data and pure functions, and then call that part from parts of the program that use IO. Maybe something like when scheme is used for teaching could work, with different "language levels", where some part disallow IO and reference cells. Or maybe a different file extension that's checked by the compiler to be pure. This could be a great way to teach functional programming but also software engineering.

BTW (and I'm replying to directly so you see it), I think you might like to see this:

https://github.com/yallop/effects-bibliography

Lots of information on varieties of effect systems

Reply on news.ycombinator.com