On this page:
3.1 Type Annotation and Binding Forms
3.1.1 Annotating Definitions
3.1.2 Annotating Local Binding
3.1.3 Annotating Functions
3.1.4 Annotating Single Variables
3.1.5 Annotating Expressions
3.2 Type Inference
3.2.1 When do you need type annotations?
3.3 New Type Names
8.14.900

3 Specifying Types🔗ℹ

The previous section introduced the basics of the Typed Racket type system. In this section, we will see several new features of the language, allowing types to be specified and used.

3.1 Type Annotation and Binding Forms🔗ℹ

In general, variables in Typed Racket must be annotated with their type. A later subsection (When do you need type annotations?) introduces a heuristic which more precisely details when type annotations are needed.

3.1.1 Annotating Definitions🔗ℹ

We have already seen the : type annotation form. This is useful for definitions, at both the top level of a module

(: x Number)
(define x 7)

and in an internal definition

(let ()
  (: x Number)
  (define x 7)
  (add1 x))

In addition to the : form, almost all binding forms from racket are replaced with counterparts which allow the specification of types. Typed Racket’s define form allows the definition of variables in both top-level and internal contexts.

(define x : Number 7)
(define (id [z : Number]) : Number z)

Here, x has the type Number, and id has the type (-> Number Number). In the body of id, z has the type Number.

3.1.2 Annotating Local Binding🔗ℹ
(let ([x : Number 7])
  (add1 x))

The let form is exactly like let from racket, but type annotations may be provided for each variable bound. Here, x is given the type Number. The let* and letrec are similar. Annotations are optional with let and variants.

(let-values ([([x : Number] [y : String]) (values 7 "hello")])
  (+ x (string-length y)))

The let*-values and letrec-values forms are similar.

3.1.3 Annotating Functions🔗ℹ

Function expressions also bind variables, which can be annotated with types. This function expects two arguments, a Number and a String:

(lambda ([x : Number] [y : String]) (+ x 5))

This function accepts at least one String, followed by arbitrarily many Numbers. In the body, y is a list of Numbers.

(lambda ([x : String] . [y : Number *]) (apply + y))

This function has the type (-> String Number * Number). To specify the return type, add a type annotation after the arguments:

(lambda ([x : String] . [y : Number *]) : (U Number String) (apply + y))

Functions defined by cases may also be annotated:

(case-lambda [() 0]
             [([x : Number]) x])

This function has the type (case-> (-> Number) (-> Number Number)). To specify the return type, either annotate the entire function or use the expression annotation form (ann) inside each case.

3.1.4 Annotating Single Variables🔗ℹ

When a single variable binding needs annotation, the annotation can be applied to a single variable using a reader extension:

(let ([#{x : Number} 7]) (add1 x))

This is equivalent to the earlier use of let. This is mostly useful for binding forms which do not have counterparts provided by Typed Racket, such as match:

(: assert-symbols! ((Listof Any) -> (Listof Symbol)))
(define (assert-symbols! lst)
  (match lst
    [(list (? symbol? #{s : (Listof Symbol)}) ...) s]
    [_ (error "expected only symbols, given" lst)]))
3.1.5 Annotating Expressions🔗ℹ

It is also possible to provide an expected type for a particular expression.

(ann (+ 7 1) Number)

This ensures that the expression, here (+ 7 1), has the desired type, here Number. Otherwise, the type checker signals an error. For example:

> (ann "not a number" Number)

eval:2:0: Type Checker: type mismatch

  expected: Number

  given: String

  in: Number

3.2 Type Inference🔗ℹ

In many cases, type annotations can be avoided where Typed Racket can infer them. For example, the types of all local bindings using let and let* can be inferred.

(let ([x 7]) (add1 x))

In this example, x has the type Exact-Positive-Integer.

Similarly, top-level constant definitions do not require annotation:

(define y "foo")

In this examples, y has the type String.

Finally, the parameter types for loops are inferred from their initial values.

(let loop ([x 0] [y (list 1 2 3)])
  (if (null? y) x (loop (+ x (car y)) (cdr y))))

Here x has the inferred type Integer, and y has the inferred type (Listof Integer). The loop variable has type (-> Integer (Listof Integer) Integer).

3.2.1 When do you need type annotations?🔗ℹ

The last several subsections explained several ways to add type annotations and explained that type inference allows some annotations to be left out. Since annotations can often be omitted, it is helpful to know the situations in which they are actually required.

The following four rules of thumb will usually suffice to determine if a type annotation is necessary.

An expression or definition needs a type annotation if it:

Here are examples that correspond to each of the cases above:

Example 1:
(: fn (-> String Symbol))
(define (fn str) ...)
Example 2:
(: fn (-> String Symbol))
(define fn (lambda (str) ...))
Example 3:

(map (lambda ([n : Integer]) (add1 n)) '(1 2 3))

Example 4:
(: maybe-animal (Option String))
(define maybe-animal #f)
(set! maybe-animal "Odontodactylus scyllarus")

In all four cases, if the type annotation is omitted then the inferred type will often be too conservative (e.g., Any) and the code may not type-check.

3.3 New Type Names🔗ℹ

Any type can be given a name with define-type.

(define-type NN (-> Number Number))

Anywhere the name NN is used, it is expanded to (-> Number Number). Type names may be recursive or even mutually recursive.