>
with Clojure (and Scala's functional support)There's this perception that Scala is less FP than a language like Clojure and I don't understand why, when in my (albeit limited) experience, I found the contrary to be true. This impression that's perpetuated is IMHO only folklore coming from people that only played around with these languages superficially.
Scala feels more like a functional programming language to me because:
(-) Code written in Scala tends to be more FP. One of the popular Scala frameworks, Play 2.0, is using Iteratees for dealing with responses and I/O in general, a concept used in Haskell, being pure awesomeness if you want to do Comet/Websockets. Play itself feels more reactive than Node.js, even though Node.js was built from the ground up for this. Also monadic types are used everywhere, from the standard collections to the Maybe monad (Option) to Futures/Promises. In Clojure you rarely see this, because Clojure is so dynamic that working with complicated container types is unfeasible.
(-) the Scala collections library provides more variants of immutable data-structures optimized for different things and those immutable data-structures in Scala are doing the right thing - first they play well with each other (e.g. mapping over a Map produces another Map if the function returns a tuple, but produces a Seq in case it returns a single element and there are many more examples where Scala simply knows what you mean).
Many times you pick concrete implementations like Vectors because of the provided invariants (e.g. indexing in O(1) or efficient adding/removing of elements on the right) and changing those invariants is IMHO bad design. Doing operations on a Vector like adding elements shouldn't produce a list, even if that list is lazy or whatever other reason. In Scala you know what operations are going to behave lazily because such invariants are encoded in the type. Adding/removing/transforming elements from a lazy Stream or view or other kind of lazy collection is still going to produce a lazy collection. And this is extremely useful in languages where laziness is not the default.
From the perspective of its rich collection of immutable collections, Scala is more functional than popular ML languages in general, except maybe Haskell. In Ocaml idiomatic code includes the usage of mutable collections, while in Scala the mutable versions are second-class citizens - and before you say that Clojure doesn't have mutable collections, think again - arrays are part of standard code and working with Java collections is quite comfortable and given as example in Clojure books.
(-) STM and atoms are great. I use STM in Scala a lot lately. However stuff like atoms and STM are still mutable state and come with many of the same gotchas as plain vars and pretending like this is otherwise is dangerous. Precisely for this reason it's useful to be able to implement your own atoms and your own STM mechanism that is more suitable to the problem at hand, that better encapsulates state and so on and because of the reality we live in, the internals of such components can get messy, while still keeping referential transparency for the outside world or other kinds of invariants - STM in Scala is a library (and quite nice in fact since it forces you to use Refs inside transactions with compile-time checks); Erlang-style Actors are a library too; Futures/Promises are a library too. All of them created with Scala.
(-) Scala is OOP while Clojure pretends like it isn't. I'm using "pretends", as multi-methods were introduced in CLOS, an OOP extension to Common Lisp, which made no effort to hide the fact that multi-methods are still all about OOP.
But just because OOP is very efficient (and has been used) to hide mutable state, that doesn't make OOP any less orthogonal to functional programming. Maybe tying up data with functions operating on that data leads to bad design, but besides the fact that type-classes are also part of idiomatic Scala and used everywhere, OOP is actually a quite nice and comfortable ad-hoc mechanism for polymorphism, being quite orthogonal to FP. Clojure is only able to have multi-methods and protocols because it is dynamic, but as far as FP is concerned this has drawbacks. Also in regards to multi-methods, I've seen code in the wild that dispatches on some global state instead of their arguments, since you can dispatch on anything and I'm really talking about mutable state.
IMHO I can't find evidence that Clojure is more FP than Scala. I'm more experienced with Scala than Clojure, so maybe this is why. But IMHO, Scala is more potent if you want to explore concepts of functional programming, being in my opinion the second choice after Haskell.
And I do think that this perception only has legs because Scala has a Java-like "while" loop and doesn't hide the fact that "vars" are really vars.