[tormeh]

>The best part of Haskell is that function application is just whitespace.

It's the worst part, in my opinion. Is A B C D equal to A(B(C(D))), A(B,C,D), A(B,C(D)), A(B(C),D) or A(B(C,D))? Often you need to disambiguate with parentheses, so I guess there is some default interpretation that you can assume when reading code, but in the end it just looks like an unnecessary mental translation step.

Lisp has weird syntax, but lisp has a reason to have that syntax. Haskell has a weird syntax because it wants to either be different for different's sake or to just be plain inaccessible.

[epidemian]

Consider that, in Haskell:

1. all functions take a single argument (i.e., are curried)

2. function application is left-associative

So "a b c d" is just applying the function "a b c" with the argument "d", the same as "(a b c) d". "a b c" is the application of "c" to the function "a b", which, in turn, is the application of "b" to "a". So "a b c d" is the same as "((a b) c) d".

I personally find using the space symbol for a core concept of the language (function application) very elegant, and not dissimilar to other languages that strive to have simple and consistent syntax, like Smalltalk, where "obj meth1 meth2 meth3" is the same as "((obj meth1) meth2) meth3", even though Haskell and Smalltalk might be as different as programming languages can get :)

[tormeh]

How does currying help? You still have to figure out which argument belongs to which function. GHC demands parentheses if you try to write A(B(C(D))) as A B C D, so I'm pretty sure I'm right.

[tomp]

No, you're wrong. `A(B(C(D)))` is different from `(((A B) C) D)` (which is the parenthesized version of `A B C D`).

Currying allows simple partial application; if A is a function that "takes 3 arguments" (i.e. it has type `w -> x -> y -> z`, which means that it takes an argument of type `w` and returns a function with type `x -> y -> z`, which takes an argument ...), you can say e.g.

  A_ = A B C
  E1 = A_ D1
  E2 = A_ D2

where `A_` is the partially applied function `A`.

[michaelcampbell]

> No, you're wrong.

And you and he both gave evidence to the initial assertion that "it's confusing".

[gamegoblin]

I'm not sure if "confusing" is fair -- just different than C-family languages.

The rule is simply "The first expression is a function, all following expressions are arguments to that function"

It's no different than "The first symbol is a function, all comma-separated, parentheses-bound expressions are arguments to that function"

Both are pretty easy syntactic rules, I think.

[JoeAltmaier]

Its arguably confusing, due to the lack of parentheses factoring a list of functions into one or more invocations. How is the nesting done here? Something more subtle. A lifetime of reading parenthesized functions is going to be confounded by changing the convention - which equals 'confusing'

[tormeh]

I understand currying and partial function application. That's not the problem. The problem is that function application whitespace demands an arbitrary application precedence and obscures function signature to the reader.

[tomp]

ah, that. Well, you get used to it. Also, it helps that application has the highest precedence, so it's quite easy to read expressions like `f x y + g z` (i.e. `(f x y) + (g z)`).

[gamegoblin]

You don't have to figure out which argument belongs to which function.

The syntax is "The first expression is a function, all following expressions are arguments to that function". Fullstop.

It would be unfair of me to say "C has confusing and ambiguous calling conventions, the compiler complains if I write A(B(C(D))) as A(B,C,D)"

Because you would respond "Well of course, that's just the calling conventions of the language!"

To which I say "Exactly"

[gamegoblin]

A B C D is necessarily equal to A(B,C,D) in C-family syntax due to the precedence rules, there is no ambiguity.

The reason whitespace is more elegant is because your functions can return other functions, which due to function currying and partial application is equivalent to your function simply taking more arguments.

I'll use Python as an example.

    def flip(f): return lambda a,b : f(b,a)

That is, it's a function that takes a function of two arguments, and returns a function that does the same thing, but with the argument order flipped.

To use these function, it would look like:

    flip(f)(a,b)

In Haskell, a direct translation with the lambda would be:

    flip f = \a b -> f b a

The type of which could be written as:

    flip :: (a -> b -> c) -> (b -> a -> c)

That is, a function which takes one argument, that argument is a function that takes two arguments of types a and b and yields a value of type c, and returns a function that takes two arguments of types b and a and yields a value of type c.

We could use this function like:

    (flip f) a b

But due to function precedence we could use

    flip f a b

This segues into an equivalent (and arguably better) way to write the function taking advantage of partial application and currying:

    flip :: (a -> b -> c) -> b -> a -> c
    flip f a b = f b a

Which would be used the exact same way:

    flip f a b

This illustrates the elegance of currying.

    a -> b -> c

Is equivalent to

    a -> (b -> c)

[TazeTSchnitzel]

All Haskell functions are curried, so there's no ambiguity.

[Retra]

>Lisp has weird syntax, but lisp has a reason to have that syntax.

Lisp has that syntax because it provides a uniformity to parsing the language, and that's the same thing Haskell offers .

If I write

    a = b + c

You might as well parse that as `a(=(b(+(c))))` with the appropriate semantics.

In that sense, Haskell is just trying to do something like what Lisp does without all the extra parentheses.

[tormeh]

That's... actually really neat. So there's no operator precedence?

[Retra]

Yes, there is operator precedence, but it only matters for infix notation. Or rather, function application has highest precedence, so there is no ambiguity using it.

My broader point was that the space is the most common symbol in English because the function that it performs is the most fundamental action of the language. And since Haskell is a functional language, function application is fundamental, and thus you gain a lot of terseness by using spaces to denote it.

Lisp does the same thing: `(+ a b)`, it just puts parentheses around things to make the parsing rules simpler. If you parse it in a different way, you won't need that many parentheses.