To be clear about my stance, I agree with your conclusions. I am a proponent of imperative programming and I don’t believe it is at odds with functional programming. For evidence of this, see my answer to Jose Valim’s challenge which uses (an interface isomorphic to) mutable state. In fact I think effect systems like effectful are the culmination of the quip that “Haskell is world’s finest imperative programming language”.
I do, however, disagree with your reasoning. I don’t see your small example program as clear, I don’t believe in forcing programming language constructs to appear in a specific order just because people read in a specific order, nor because some problems are best solved when constructs appear in a specific order. I believe allowing pure code to appear in any order, and effectful code to appear in the order necessary to achieve the desired result. One example of this distinction is between Haskell’s (pure) let expressions and monadic <- binds.
It is not only about determining an order between sub-expressions. It is also about determining the constraints on the memory in which data is actually stored. As I mentioned, recovering the efficiency of an in-place update in general is a heroic task with no strict guarantee of success.
That’s a fair point. Haskell does allow that, of course, with the State monad, or STRef/IORef for better performance, or unboxed mutable arrays for the best performance. As I said before, I admit that that is clunky but I haven’t seen anything I prefer yet (though I haven’t looked at Hylo).
At this point I am content to agree to disagree 