**We are hiring a PhD student to do research on functional high performance computing, including work on Futhark itself. See here for more details.**Source file: functions.fut

# Functions

Functions are defined with `def`

, where we can also annotate both
the parameter and return types.

```
def plus2 (x: i32) : i32 =
2 x +
```

Type inference is supported, so in many cases the types can be left off, in which case the correct type will be inferred from the definition.

```
def plus2_inferred x =
2 x +
```

Functions are (almost) first-class values, with the type of a
function from type `a`

to type `b`

written as `a -> b`

. This means
we can write higher-order functions:

```
def on_i32 (f: i32 -> i32) (x: i32) =
f x
```

When passing an argument to a higher-order function, it is often
most practical to use an *anonymous function* (sometimes called a
*lambda*), like this:

`def four = on_i32 (\x -> x + 2) 2`

A shortcut notation, called *operator sections*, allows partial
application of infix operators:

`def another_four = on_i32 (+2) 2`

By enclosing an operator in parentheses we can treat it like any variable name:

`def plus = (+)`

And we can treat a variable as an infix operator by enclosing it in backticks:

`def yet_another_four = 2 `plus` 2`

Function values are restricted as follows:

Arrays of functions are not permitted.

A function cannot be returned from an

`if`

expression.A loop parameter cannot be a function.

However, it is fine to collect functions in tuples or records.

```
def fun_tuple : (i32 -> i32, i32 -> i32) =
(plus2, plus2_inferred)
```