Hacker News new | ask | show | jobs
by DerpDerpDerp 4352 days ago
I do work in Haskell.

At the end of the day, you still write some stateful, imperative code.

It's just that your utility/work functions are pure, which means that you have predictable side-effects: you've stuck them all in one (imperative) place, or perhaps a couple places, but you're guaranteed to know where they are because the type system enforces notating which code has the ability to cause side effects. (Makes it easy to grep on type signatures, for example.)

Most of my workload is dominated by high level concerns: correctness, large-scale data flow, coordination of state, and security.

The purity of functions means that I explain all of the constraints/solutions of those problems in pieces, which I then compose with a clear data flow between them (since they can't have shared, mutable state), and only at the end, lift them to operate on the state of the system (in a stateful, imperative block).

This way of programming gives me much more confidence in replacing pieces of the code without worrying about the side effects, because the side effects are predictably related to that code.

At the end of the day, that predictability of side effects makes the majority of my work MUCH easier, even if it adds complexity to some low level data manipulation.

I easily write 10x the high level code I do low level code, so a 20% savings on that code, even if I double the complexity of low level code, is a net savings on my time/effort.

tl;dr: Functional programming is a productivity booster if most of your work is of a high level nature, even if it makes low level manipulations harder to do. Interfacing with low level imperative code can be (and is) regularly done, to get the best of both worlds.

1 comments

This is certainly an intriguing answer, and makes me want to try Haskell sometime, but it also makes me wonder if I can't get the same benefits, with a little effort and discipline, in a "semi-functional" language such as something in the ML family or even (if I can live without static typing) Lisp.

In fact, I've always made a point, when working in a semi-functional language, of writing as much code functionally as possible. My experience is that it has the same benefits as you describe. So I don't know how much more there is to be gotten by actually switching to Haskell. My guess is, most of the additional benefit would come from the type system. But maybe I would be motivated to write even more code functionally.

All being a "functional language" means is that it's easy to write with a particular paradigm in the language.

You certainly can write similar code in ML, Java, or C - or whatever language strikes your fancy. It just helps when the type system is lined up with that goal, such as forcing you to denote where side-effects occur, and ensuring only functions that expect side-effect based code can operate on them. (For what it's worth, I've used both ML and Scheme to write similar style code as part of a programming languages class; and similarly, the C code I write relies on the functional ideas as much as possible.)

I just like the combination of Haskell's syntax and type system, so I prefer it to something like C (where functions aren't exactly a first class datatype, and the type system is somewhat weaker).

I just started to work with Haskell seriously, and after some trouble I discovered exactly the same development approach as DerpDerpDerp. However since I am not yet experienced I wish I had an "observer" to print debug messages (I know that there is a debugger in EclipseFP but I prefer Emacs for performance reasons). With observer I mean a compiler pragma that lets me print Haskell values without violating the functional nature of the code.

By the way, you don't need to live without static typing in Lisp. Shen provides a powerful type-safe layer on top of Lisp.

http://www.shenlanguage.org/learn-shen/types/types_functions...

If you want to be pure, you can add a strict WriterT to your monad stack, or something similar.

If you don't mind temporary impurity, Debug.Trace provides some helper functions, but they call unsafePerformIO, which can do unexpected things.

Sounds like you're looking for the Debug.Trace module
Thanks, I'll take a look at it.
Many of the same effects can be obtained using even non-functional languages in a disciplined way. However, many of the best strategies and patterns for constructing pure, stateless software are enshrined in the Haskell standard libraries.

Once you learn the techniques and programming styles of Haskell, you can certainly apply those ideas elsewhere, but they lose a lot of their power when you (or another developer on the same code base) start "cheating".