Having -> denote a return variable has the benefit that you the syntax "a -> b" is non-ambiguously a function type.

For example, a function that accepts a function as an argument can be written:

    fn foo(f: int -> string)

Or if the argument name isn't needed in this context (e.g. think interface declarations):

  fn foo(int -> string)

If you see ":" as meaning "has type", then using it for a return type isn't necessarily more consistent, because it's not the function that has a type.

In functional programming (the origin of this syntax),

  add(a: int, b: int): int

can be read as a function 'add' when applied to the parameter tuple (a: int, b: int) has type int, I believe Ocaml, SML, actually support this syntax, whereas '->' (arrow) is a function type constructor in functional programming with 'A -> B' denoting a function from A to B. The type of add by itself in a language like Haskell would be:

add :: (int, int) -> int

Not quite, in ML languages, a function declared like this:

  let sum a b = a + b

...has the type:

  float -> float -> float

Same in Haskell.

These languages "emulate" multiple arguments through partial function application. OCaml does not use tuples for arguments.

You can certaintly have tupled function arguments in SML/Ocaml, in fact that seems to have been the more common style there, whereas Haskell uses curried functions more commonly.

As far as the original sugestion, using '->' (arrow) to denote a return type (like Rust) is nonsense in a functional language and kind of butchering the functional inherited syntax IMHO.

I've never heard of OCaml tuples being more common in arguments. If anything, my impression is they're discouraged (in OCaml, not SML) as a replacement for curried arguments. E.g. see [1].

> is nonsense in a functional language

SML/OCaml and Haskell all use ->, so I'm not sure what you mean by this?

[1] https://stackoverflow.com/questions/10666913/why-prefer-curr...

It is quite common in SML/ML code I've encountered, probably not as common in modern OCaml. BTW, in OCaml

  let add ((a:int), (b:int)) : int = a + b

Works just fine, whereas:

  let add ((a:int), (b:int)) -> int = a + b

is nonsense, because '->' is a type constructer in OCaml (a function from types to types) not part of an ad-hoc syntax for function declarations (like Rust).

Right, I'm not sure with OCaml is inconsistent there. -> is also used in pattern matching and in the syntax "fun x -> x + 1", perhaps it's some historical thing to avoid ambiguity?

The above usage of -> is a syntax error in OCaml, -> can be used in the left hand side of a let expression, but only as part of a type expression. As you say the other usage is used to construct terms (clauses) that are used in pattern matching, of course Haskell does something similar.

The Rust syntax seems to emulate functional programming, but inconsistently since -> is not used purely as a type constructor or term constructor something like a mix of both in the defintion of functions, also it doesn't seem to support currying.

Some reasonable, but I see a new inconsistency here. Shouldn't

    fn foo(fn (int) -> string)

be more consistent?

It's not inconsistent, just terser. You could say that "fn" declares a function by name; the type of foo is int -> str.

In Haskell, this is written:

  foo :: int -> str

Of course, Haskell is based on lambda calculus, so multiple arguments are just a generalization of partial function application:

  foo :: int -> int -> str

(A function taking two int arguments and returning a string, which is indistinguishable from a function that takes a single int and returns a function that takes a single int and returns a string.)

Ah, looks both have their own rationality. ":" is ok for single-value returns. AS3 also uses ":".

The Haskell language looks interesting. Maybe I should spend some time to study it.

Thats precisely the reason its written like that in Mun