[tikhonj]

Several things jump out immediately:

  - Haskell has exceptionally thorough type inference (almost all type declarations are optional).
  - anything impure (e.g. involving an I/O) is typed as such
  - the types are easier to read and tend to give a good idea of what the function in question is doing
  - much more complicated types are readable
  - idioms like Monads lead to a lot of code that is much more abstract and generic than anything I've seen in Java. There are tons of library functions that can work on lists, or IO things, or parsers, or random values...

Some things that I found particularly cool in Haskell: - functions (and even constants!) can be polymorphic on their return type. The function read, for example, always takes in a String but returns whatever type you happen to need. - You can easily express things like a "list of list": [[a]]. I don't even know how to begin writing a function like join in Java. (On lists, join takes a list of lists of something and flattens it.)

Overall, I've found Haskell's type system to be much more useful, and much less of a burden, than Java's.

That said, I'm just a college student with limited experience with either language. I've used Java more than Haskell but I've been using the latter more recently; I've never used C#.

[skybrian]

Maybe it's better with more experience or when you're reading your own code, but I find that reading a Haskell program is often like reading a math paper - you can't skim it; you have to slow way down and study it, one line at a time, and look up the definition of each term as you go and think about what it means, and try out simple examples. And sometimes you don't have the background to understand the underlying concepts, so you have to study a prerequisite first.

So I think to like Haskell, you have to truly believe that more abstraction is better. It's very much a mathematician's language.

The thing is, when writing code, I don't want to write a math paper. When I'm done, I want it to be as smooth and easy to read and as obvious as an article in the New Yorker.

[jerf]

"When I'm done, I want it to be as smooth and easy to read and as obvious as an article in the New Yorker."

Your metaphor is apt; if your code is going to be that smooth, you need to learn Haskell to the same extent you know English for the New Yorker. The Internet suggests that the New Yorker is written for a "10th grade reading level", but of course the average American has between an 8th or 9th grade level.

It does get easier with practice to read the language, and what's left after that is whatever core difficult the code being expressed has regardless of the target language, the essential difficulty. Haskell can hardly be blamed for bringing that to the foreground where other languages stuff it in the background.

[skybrian]

If I learn Haskell well and start writing code that's idiomatic for that audience, I'm limiting my audience to the few people who also know Haskell (and the particular Haskell idioms I'm using). It's like writing in Esperanto instead of English. My ideal is more along the lines of the Python examples that Peter Norvig writes, which can be understood by pretty much any programmer, even if they're not all that fluent in Python.

[nightski]

It has nothing to do with being a mathematician. There are common patterns as any language. Those patterns often end up as higher order functions. But at some point, it becomes very natural to read.

[Dn_Ab]

The higher order monadic join concept cannot be implemented in java but a simpler function in terms of flattening a list of list is easy enough. In a functional language a simple way would be to fold over (++) append.

so in java a simple way would be to initialize an l = arrayList and then do a foreach on each list in the list of the lists and foreach element in each list add it to the arrayList l. Not as elegant, much more how instead of what but 4 noisy lines max.

C# has list.selectMany( _ => _ ) for that.

[eru]

> In a functional language a simple way would be to fold over (++) append.

Though you probably wouldn't want to do use that approach for performance reasons.