The biggest problem of very functional code on the web is simply debugging it. So much control flow is buried in compositions that there isn’t much to grab on in terms of break points or variable inspection. Pure languages exacerbate this problem by making state more difficult to dig out and pin down at run-time.
I don't see why that's a problem with FP and not just the lack of tooling in most FP languages. Debugging in F# via Visual Studio is a breeze, and OCaml's time travelling debugger looks pretty good.
do
x <- newIORef 0
writeIORef x 2
print =<< readIORef x
[0] https://en.wikipedia.org/wiki/Software_transactional_memory
The notion of monad comes from category theory [...] It first arose in the area of homological algebra, but later was recognised (due to the work of Kleisli and of Eilenberg and Moore) to have much wider applications. Its importance emerged slowly: in early days, it was not even given a proper name, but called simply a "standard construction" or a "triple" [...] Eugenio Moggi proposed that monads provide a useful structuring tool for denotational semantics [...] He showed how lambda calculus could be given call-by-value and call-by-name semantics in an arbitrary monad, and how monads could encapsulate a wide variety of programming language features such as state, exception handling, and continuations. Independent of Moggi, but at about the same time, Michael Spivey proposed that monads provide a useful structuring tool for exception handling in pure functional languages, and demonstrated this thesis with an elegant program for term rewriting [...]. He showed how monads could treat exceptions [...] and non-deterministic choice [...] in a common framework.
[1] P. Wadler, The essence of functional programming.
Scala and ML-variants like Ocaml, F#, Reason.js are perfectly good imperative languages. They don't force you to go purely functional. I always recommend that my students (whose first language is Java) start by treating Scala as a "nicer Java", and then slowly explore Scala's more powerful features. Indeed, I'd argue that Scala and the aforementioned ML-variants are (slightly) better general-purpose imperative languages than the Javas, Pythons, PHPs and Javascripts of this world, because they are more consistent, they have types/type-inference, and their statefulness fully supports higher-order state and powerful modularisation without restriction, not to mention pattern matching.
Being able to use functional idioms without compromise if and when desired (including, but not restricted to, pure functional programming) is another benefit of Scala and ML-variants.