Whereas after spending even a day with it I found myself wanting to run screaming back to Haskell.
No typeclasses, no monads, more primitive pattern matching, no guards, use of massive nested elseif considered idiomatic ...
Simple tasks wind up feeling tedious because so many of the standard tools and patterns of any other ML language are simple missing completely. The result feels like trying to work with one of those random toy Lisps people make in a weekend and throw up on Github.
It frustrates me, because in theory I do like the concept of the Elm architecture, I love ML-style type systems, and I work with reactive programming every day in ClojureScript. But the language it's attached to is so incomplete that it quickly just becomes a pain.
That's exactly the dividing line I see in those who love Elm and those who don't. People coming from Ruby / JS backgrounds appreciate a modern type system and it isn't so different from languages they are used to that it is hard to get going. People already used to modern functional programming miss the abstractions they are used to that Elm doesn't provide. I find this really interesting. As a community we don't give enough attention to the sociological aspects of programming. You can see similar things in the adoption of languages like Go and Elixir.
(Sorry for my poor english but i need some clarification) Do you put Go, Elixir and Elm is the same bag, i.e langages a step further JS in term of abstraction, i.e langages easy to understand when you come from plain JS? And then, you put Purescript and Haskell in another bag, i.e langages with advanced abstractions absent from "casual" langages, i.e langages requiring a lot of effort to learn? Did I summarize your idea correctly?
I think understood the Elm/Haskell/Purescript, i.e:
Javascript dev sees Elm and might think "Wow,a type-system and it is awesome, didn't expect that"
Haskell dev sees Elm and might think "I am missing so many features with these types" and consider something else, probably pure-script, that has similar toolbelt.
a-saleh is correct. Regarding Go and Elixir, I think Ruby / Python seem them and think "wow, I can program in a familiar style and I get performance and concurrency!" Haskell etc. people look at them and think "this smells."
Look no further! Try Purescript. It's basically Haskell (albeit strictly evaluated) and has everything you listed, plus a record system I really wish Haskell would adopt.
I'm currently using it in production with purescript-thermite, a thin wrapper over React, and it's worked out great.
Elm is great for people coming from JS, but not so much for people coming from Haskell.
That was my first thought, too: Elm is quite similar to OCaml, so why not using OCaml in the backend? With all that server-side stuff around MirageOS, you can not only build a server app, but a full server VM within OCaml.
On the other hand, OCaml also targets JavaScript coding in the browser, via js_of_ocaml or the newer bucklescript.
I would be very interested in a comparison of client-side development with OCaml+js_of_ocaml (or OCaml+bucklescript) versus Elm.
From what I understand Elm is planned to target Node eventually. However the language is pretty young right now and the main work is going into making client side Elm the best it can be. The language author doesn't want to branch out too far.
It makes my heart sing that a different VM than JVM or .Net is gaining more traction, and as a static types fan it makes me happy that a statically typed language may be getting on Beam.
It will definitely be a very different compile target. But Elm really doesn't take any inspiration from Javascript, so I don't think it will change much about the fundamentals of the language. Of course they'll need to add capabilities for concurrency, OTP, etc.
I think the compile target is basically not a big problem, and there's been explicit work to make the Browser just one Platform. Also, see the `Process` module in the core docs for a hint of the future, regardless of where it lands (this would also work well for WebWorkers etc, and is handled right now by a naive scheduler). Concern over the compile target should be super low - if you've built a language of sorts, you should have a feel for how it's not that big of a deal to tweak the target.
You can get a "kind of Elm" experience on the backend with Flow and JavaScript / NodeJs. The type system of Flow is not as strict as Elm, but you have the main parts (including disjoint unions) and can use existing JavaScript code. With React and Redux you also get a similar framework for the frontend.
I have been thinking about toying with JVM's Graal compiler (JIT and customizeable deopts for the price of an interpreter). But with a language that has few corner cases. I was thinking maybe regexes, or brainfuck, or QBasic.
I'd just looove to try and put Elm on the JVM, but I just can't at the moment.
Elixir and Erlang have a type checker as well. Not Hindley–Milner but one called "success typing" and is checked by a tool called Dialyzer. It is rather good actually:
But is still optional to compile. The more precise and better the type annotations, the more helpful it is. If it can deduce that some inconsistency or type error occurring it will let user know. If it is not sure, it won't say anything.
Another point is because of isolated process heaps and extra fault tolerance, it is possible to get a high degree of assurance from a system built in Erlang or Elixir even with dynamic typing. And there are certainly many examples of that.
Dialyzer is indeed nice, but as you say, it's completely optional, it's not Hindley-Milner, and it relies entirely on thorough, correct type annotations from the developer.
Elixir also follows the Ruby philosophy of hiding as much behind the syntax as possible, whereas Elm tends to be very explicit. So I stand by my point that Elixir is a very different language in practice from Elm.
> Elixir also follows the Ruby philosophy of hiding as much behind the syntax as possible, whereas Elm tends to be very explicit.
This is not true. While Elixir may not make things as explicit as Elm (I don't know Elm well enough to assert or refute such a statement), the author of Elixir has stated in various occasions that Elixir prefers explicit to implicit.
For instance, in Elixir, unless one explicitly defines a String.Chars protocol for a data structure, string interpolation of that data structure will not compile. In contrast, pretty much everything can be implicitly interpolated in Ruby.
Another example is function calls. In Elixir, if f is not a named function but a variable bound to an anonymous function, you cannot call it by writing something like f(x). Instead, you must put a dot after f and write f.(x) to state clearly that you are calling an anonymous function.
I use Erlang and I like it a bit more than Elixir in general exactly because it is more explicit. It involves slightly more syntax but there is less hidden magic stuff happening behind the scene.
You are right about the type system but still, Elixir is a lighter, friendlier functional programming language compared to haskel or erlang. Not having previous FP experience, I'm having a good time learning Elixir + Elm.
Could you clarify? I don't immediately see the relevance of this. Why does it matter if the compilation target is or is not also strongly typed, and how does that imply anything about the experience of using a high-level dynamically typed language vs. a high-level statically typed language?
I understood the (some number of grands-)parent comment to be saying "Elixir seems a good candidate [for a compilation target]." If that was not your interpretation, then it is certainly non-sequiter. Re-reading, I think I misread something in a surrounding comment, though I'm not sure what.
I said "Elixir seems a good candidate" for backend development, not as a compilation target. The OP asked for suggestions regarding light, friendly FP languages for backend dev.
No typeclasses, no monads, more primitive pattern matching, no guards, use of massive nested elseif considered idiomatic ...
Simple tasks wind up feeling tedious because so many of the standard tools and patterns of any other ML language are simple missing completely. The result feels like trying to work with one of those random toy Lisps people make in a weekend and throw up on Github.
It frustrates me, because in theory I do like the concept of the Elm architecture, I love ML-style type systems, and I work with reactive programming every day in ClojureScript. But the language it's attached to is so incomplete that it quickly just becomes a pain.