8.6 Attaching Contracts to Values
syntax
(contract-out unprotected-submodule contract-out-item ...)
unprotected-submodule =
| #:unprotected-submodule submodule-name contract-out-item =
(struct id/super ((id contract-expr) ...) struct-option) | (rename orig-id id contract-expr) | (id contract-expr) | #:∃ poly-variables | #:exists poly-variables | #:∀ poly-variables | #:forall poly-variables poly-variables = id | (id ...) id/super = id | (id super-id) struct-option =
| #:omit-constructor
The contract-out form treats modules as units of blame. The module that defines the provided variable is expected to meet the positive (co-variant) positions of the contract. Each module that imports the provided variable must obey the negative (contra-variant) positions of the contract. Each contract-expr in a contract-out form is effectively moved to the end of the enclosing module, so a contract-expr can refer to variables that are defined later in the same module.
Only uses of the contracted variable outside the module are checked. Inside the module, no contract checking occurs.
The rename form of contract-out exports the first variable (the internal name) with the name specified by the second variable (the external name).
The struct form of contract-out provides a structure-type definition, and each field has a contract that dictates the contents of the fields. The structure-type definition must appear before the provide clause within the enclosing module. If the structure type has a parent, the second struct form (above) must be used, with the first name referring to the structure type to export and the second name referring to the parent structure type. Unlike a struct definition, however, all of the fields (and their contracts) must be listed. The contract on the fields that the sub-struct shares with its parent are only used in the contract for the sub-struct’s constructor, and the selector or mutators for the super-struct are not provided. The exported structure-type name always doubles as a constructor, even if the original structure-type name does not act as a constructor. If the #:omit-constructor option is present, the constructor is not provided.
Note that if the struct is created with serializable-struct or define-serializable-struct, contract-out does not protect struct instances that are created via deserialize. Consider using struct-guard/c instead.
The #:∃, #:exists, #:∀, and #:forall clauses define new abstract contracts. The variables are bound in the remainder of the contract-out form to new contracts that hide the values they accept and ensure that the exported functions are treated parametrically. See new-∃/c and new-∀/c for details on how the clauses hide the values.
If #:unprotected-submodule appears, the identifier that follows it is used as the name of a submodule that contract-out generates. The submodule exports all of the names in the contract-out, but without contracts.
The implementation of contract-out uses syntax-property to attach properties to the code it generates that records the syntax of the contracts in the fully expanded program. Specifically, the symbol 'provide/contract-original-contract is bound to vectors of two elements, the exported identifier and a syntax object for the expression that produces the contract controlling the export.
Changed in version 7.3.0.3 of package base: Added #:unprotected-submodule.
syntax
(recontract-out id ...)
It re-exports id, but with positive blame associated to the module containing recontract-out instead of the location of the original site of id.
This can be useful when a public module wants to export an identifier from a private module but where any contract violations should be reported in terms of the public module instead of the private one.
> (module private-implementation racket/base (require racket/contract) (define (recip x) (/ 1 x)) (define (non-zero? x) (not (= x 0))) (provide/contract [recip (-> (and/c real? non-zero?) (between/c -1 1))]))
> (module public racket/base (require racket/contract 'private-implementation) (provide (recontract-out recip))) > (require 'public) > (recip +nan.0) recip: broke its own contract
promised: (between/c -1 1)
produced: +nan.0
in: the range of
(->
(and/c real? non-zero?)
(between/c -1 1))
contract from: public
blaming: public
(assuming the contract is correct)
at: eval:3.0
Replacing the use of recontract-out with just recip would result in a contract violation blaming the private module.
syntax
(provide/contract unprotected-submodule contract-out-item ...)
syntax
(struct-guard/c contract-expr ...)
> (struct snake (weight hungry?) #:guard (struct-guard/c real? boolean?)) > (snake 1.5 "yep") snake, field 2: contract violation
expected: boolean?
given: "yep"
in: boolean?
contract from: top-level
blaming: top-level
(assuming the contract is correct)
at: eval:2.0
8.6.1 Nested Contract Boundaries
(require racket/contract/region) | package: base |
syntax
(with-contract blame-id (wc-export ...) free-var-list ... body ...+)
(with-contract blame-id results-spec free-var-list ... body ...+)
wc-export = (id contract-expr) result-spec = #:result contract-expr | #:results (contract-expr ...) free-var-list =
| #:freevar id contract-expr | #:freevars ([id contract-expr] ...)
The first with-contract form cannot appear in expression position. All names defined within the first with-contract form are visible externally, but those names listed in the wc-export list are protected with the corresponding contract. The body of the form allows definition/expression interleaving if its context does.
The second with-contract form must appear in expression position. The final body expression should return the same number of values as the number of contracts listed in the result-spec, and each returned value is contracted with its respective contract. The sequence of body forms is treated as for let.
The blame-id is used for the positive positions of contracts paired with exported ids. Contracts broken within the with-contract body will use the blame-id for their negative position.
If a free-var-list is given, then any uses of the free variables inside the body will be protected with contracts that blame the context of the with-contract form for the positive positions and the with-contract form for the negative ones.
syntax
(define/contract id contract-expr free-var-list init-value-expr)
(define/contract (head args) contract-expr free-var-list body ...+)
> (define/contract distance (>=/c 0) 43.52)
> (define/contract (furlongs->feet fr) (-> real? real?) (* 660 fr)) ; a contract violation expected here: > (furlongs->feet "not a furlong") furlongs->feet: contract violation
expected: real?
given: "not a furlong"
in: the 1st argument of
(-> real? real?)
contract from: (function furlongs->feet)
blaming: top-level
(assuming the contract is correct)
at: eval:3.0
The define/contract form treats the individual definition as a contract region. The definition itself is responsible for positive (co-variant) positions of the contract, and references to id outside of the definition must meet the negative positions of the contract. Since the contract boundary is between the definition and the surrounding context, references to id inside the define/contract form are not checked.
; an unsual predicate that prints when called
> (define (printing-int? x) (displayln "I was called") (exact-integer? x))
> (define/contract (fact n) (-> printing-int? printing-int?) (if (zero? n) 1 (* n (fact (sub1 n))))) > (fact 5) ; only prints twice, not for each recursive call
I was called
I was called
120
If a free-var-list is given, then any uses of the free variables inside the body will be protected with contracts that blame the context of the define/contract form for the positive positions and the define/contract form for the negative ones.
> (define (integer->binary-string n) (number->string n 2))
> (define/contract (numbers->strings lst) (-> (listof number?) (listof string?)) #:freevar integer->binary-string (-> exact-integer? string?) ; mistake, lst might contain inexact numbers (map integer->binary-string lst)) > (numbers->strings '(4.0 3.3 5.8)) integer->binary-string: contract violation
expected: exact-integer?
given: 4.0
in: the 1st argument of
(-> exact-integer? string?)
contract from: top-level
blaming: (function numbers->strings)
(assuming the contract is correct)
at: eval:3.0
syntax
(define-struct/contract struct-id ([field contract-expr] ...) struct-option ...)
(define-struct/contract (struct-id super-struct-id) ([field contract-expr] ...) struct-option ...)
The define-struct/contract form only allows a subset of the struct-option keywords: #:mutable, #:transparent, #:auto-value, #:omit-define-syntaxes, #:property and #:omit-define-values.
> (define-struct/contract fish ([color number?])) > (make-fish 5) #<fish>
> (make-fish #f) make-fish: contract violation
expected: number?
given: #f
in: the 1st argument of
(-> number? symbol? any)
contract from: (struct fish)
blaming: top-level
(assuming the contract is correct)
> (define-struct/contract (salmon fish) ([ocean symbol?])) > (make-salmon 5 'atlantic) #<salmon>
> (make-salmon 5 #f) make-salmon: contract violation
expected: symbol?
given: #f
in: the 2nd argument of
(-> any/c symbol? symbol? any)
contract from: (struct salmon)
blaming: top-level
(assuming the contract is correct)
> (make-salmon #f 'pacific) make-fish: contract violation
expected: number?
given: #f
in: the 1st argument of
(-> number? symbol? any)
contract from: (struct fish)
blaming: top-level
(assuming the contract is correct)
syntax
(invariant-assertion invariant-expr expr)
Unlike the specification of a contract, an invariant-assertion does not establish a boundary between two parties. Instead, it simply attaches a logical assertion to the value. Because the form uses contract machinery to check the assertion, the surrounding module is treated as the party to be blamed for any violations of the assertion.
This means, for example, that the assertion is checked on recursive calls, when an invariant is used on the right-hand side of a definition:
> (define furlongss->feets (invariant-assertion (-> (listof real?) (listof real?)) (λ (l) (cond [(empty? l) empty] [else (if (= 327 (car l)) (furlongss->feets (list "wha?")) (cons (furlongs->feet (first l)) (furlongss->feets (rest l))))])))) > (furlongss->feets (list 1 2 3)) '(660 1320 1980)
> (furlongss->feets (list 1 327 3)) furlongss->feets: assertion violation
expected: real?
given: "wha?"
in: an element of
the 1st argument of
(-> (listof real?) (listof real?))
contract from: invariant-assertion
at: eval:5.0
Added in version 6.0.1.11 of package base.
syntax
8.6.2 Low-level Contract Boundaries
syntax
(define-module-boundary-contract id orig-id contract-expr pos-blame-party source-loc name-for-blame context-limit)
pos-blame-party =
| #:pos-source pos-source-expr source-loc =
| #:srcloc srcloc-expr name-for-blame =
| #:name-for-blame | blame-id context-limit =
| #:context-limit limit-expr
The identifier id is defined as a macro transformer that consults the context of its use to determine the name for negative blame assignment (using the entire module where a reference appears as the negative party).
The positive party defaults to the module containing the use of define-module-boundary-contract, but can be specified explicitly via the #:pos-source keyword.
The source location used in the blame error messages for the location of the place where the contract was put on the value defaults to the source location of the use of define-module-boundary-contract, but can be specified via the #:srcloc argument, in which case it can be any of the things that the third argument to datum->syntax can be.
The name used in the error messages will be orig-id, unless #:name-for-blame is supplied, in which case the identifier following it is used as the name in the error messages.
If #:context-limit is supplied, it behaves the same as it does when supplied to contract.
> (module server racket/base (require racket/contract/base) (define (f x) #f) (define-module-boundary-contract g f (-> integer? integer?)) (provide g))
> (module client racket/base (require 'server) (define (clients-fault) (g #f)) (define (servers-fault) (g 1)) (provide servers-fault clients-fault)) > (require 'client) > (clients-fault) g: contract violation
expected: integer?
given: #f
in: the 1st argument of
(-> integer? integer?)
contract from: 'server
blaming: client
(assuming the contract is correct)
at: eval:2.0
> (servers-fault) g: broke its own contract
promised: integer?
produced: #f
in: the range of
(-> integer? integer?)
contract from: 'server
blaming: (quote server)
(assuming the contract is correct)
at: eval:2.0
Changed in version 6.7.0.4 of package base: Added the #:name-for-blame argument.
Changed in version 6.90.0.29: Added the #:context-limit argument.
syntax
(contract contract-expr to-protect-expr positive-blame-expr negative-blame-expr)
(contract contract-expr to-protect-expr positive-blame-expr negative-blame-expr #:context-limit limit-expr)
(contract contract-expr to-protect-expr positive-blame-expr negative-blame-expr value-name-expr source-location-expr)
The contract expression adds the contract specified by contract-expr to the value produced by to-protect-expr. The result of a contract expression is the result of the to-protect-expr expression, but with the contract specified by contract-expr enforced on to-protect-expr.
The values of positive-blame-expr and negative-blame-expr indicate how to assign blame for positive and negative positions of the contract specified by contract-expr. They may be any value, and are formatted as by display for purposes of contract violation error messages.
> (contract integer? #f 'pos 'neg 'timothy #f) timothy: broke its own contract
promised: integer?
produced: #f
in: integer?
contract from: pos
blaming: pos
(assuming the contract is correct)
> (contract integer? #f 'pos 'neg #f #f) broke its own contract
promised: integer?
produced: #f
in: integer?
contract from: pos
blaming: pos
(assuming the contract is correct)
If specified, source-location-expr indicates the source location reported by contract violations. The expression must produce a srcloc structure, syntax object, #f, or a list or vector in the format accepted by the third argument to datum->syntax.
If #:context-limit is supplied, the following expression must evaluate to either #f or a natural number. If the expression evaluates to an natural number, the number of layers of context information is limited to at most that many. For example, if the number is 0, no context information is recorded and the error messages do not contain the section that starts with in:.