Title

Records

Author

André van Tonder

Status

This SRFI is currently in ``draft'' status. To see an explanation of each status that a SRFI can hold, see here. It will remain in draft status until 2005/01/08, or as amended. To provide input on this SRFI, please mailto:srfi-57@srfi.schemers.org. See instructions here to subscribe to the list. You can access previous messages via the archive of the mailing list.

Abstract

We describe a syntax for defining record types. A predicate, constructor, and field accessors and modifiers may be specified for each record type. We also introduce a syntax for declaring record type schemes, representing families of record types that share a set of field labels. A polymorphic predicate and polymorphic field accessors and modifiers may be specified for each record type scheme. A syntax is provided for constructing records by field label, for in-place and for functional record update, and for composing records.

Rationale

We extend the existing SRFI-9 [1] with the following features, each listed with a brief justification. Expanded rationales appear in the specification section below.

Specification

Declaration: Record types

The syntax of a record type definition is as follows:
 <command or definition>           
   -> <record type definition>           ; addition to 7.1.6 in R5RS

 <record type definition> -> (define-record-type <type clause> 
                                                 <constructor clause> 
                                                 <predicate clause>                          
                                                 <field clause> ...)  
                          -> (define-record-type <type clause> 
                                                 <constructor clause>)  
                          -> (define-record-type <type clause>)                   

 <type clause> -> <type name>                           
               -> (<type name> <type scheme name> ...)  

 <constructor clause> -> (<constructor name> <field label> ...)               
                      -> <constructor name> 
                      -> #f

 <predicate clause> -> <predicate name>                 
                    -> #f

 <field clause> -> (<field label> <accessor clause> <modifier clause>) 
                -> (<field label> <accessor clause>)
                -> (<field label>)
            
 <accessor clause> -> <accessor name>                 
                   -> #f

 <modifier clause> -> <modifier name>                 
                   -> #f             

 <field label> -> <identifier>
 <... name>    -> <identifier>

An instance of define-record-type is equivalent to the following:

define-record-type is generative: each use creates a new record type that is distinct from all existing types, including other record types and Scheme's predefined types. Record-type definitions may only occur at top-level.

Examples

A simple record:

  (define-record-type point (make-point x y) point?
    (x get-x set-x!)           
    (y get-y set-y!))                    

  (define p (make-point 1 2))
  (get-y  p)                                 ==> 2
  (set-y! p 3))                             
  (get-y  p)                                 ==> 3                             
  (point? p)                                 ==> #t  

Optional elements:

Elements may be left out if not desired, as the following examples illustrate:
  
  (define-record-type node (make-node left right)) 
  (define-record-type leaf (make-leaf value))       
In these declarations, no predicates are bound. Also note that field labels listed in the constructor do not have to be repeated in the field clause list unless we want to bind getters or setters.
              
  (define-record-type monday)               
  (define-record-type tuesday #f tuesday?)      
Here monday has no declared constructor or predicate, while tuesday has a predicate but no constructor.
  (define-record-type node make-node #f                                   
    (left  left)                        
    (right right))                     
Here the constructor make-node has the default argument order and no predicate is bound. Also note that field labels are punned.

Declaration: Record type schemes

The syntax of a record scheme definition is as follows:
 <command or definition>           
   -> <record scheme definition>           ; addition to 7.1.6 in R5RS

 <record scheme definition> -> (define-record-scheme <scheme clause> 
                                                     <deconstructor clause> 
                                                     <predicate clause>                          
                                                     <field clause> ...)  
                            -> (define-record-scheme <type clause> 
                                                     <deconstructor clause>)  
                            -> (define-record-scheme <type clause>)                   

 <scheme clause> -> <scheme name>                           
                 -> (<scheme name> <parent scheme name> ...)  

 <deconstructor clause> -> (<deconstructor name> <field label> ...)               
                        -> <deconstructor name> 
                        -> #f

 <predicate clause> -> <predicate name>                 
                    -> #f

 <field clause> -> (<field label> <accessor clause> <modifier clause>) 
                -> (<field label> <accessor clause>)
                -> (<field label>)
            
 <accessor clause> -> <accessor name>                 
                   -> #f

 <modifier clause> -> <modifier name>                 
                   -> #f             

 <field label> -> <identifier>
 <... name>    -> <identifier>

An instance of define-record-scheme is equivalent to the following:

A note on repeated fields and redefinitions

In the following example, two record type schemes define different accessors for the same field:
  (define-record-scheme foo #f #f (x foo-x))
  (define-record-scheme bar #f #f (x bar-x))
  (define-record-type (foo-bar foo bar))
Since any value fb of type foo-bar conforms to both foo and bar, both foo-x and bar-x can be applied to fb, returning the value of the x field.

In the following example, two declarations introduce the same accessor:

  (define-record-scheme foo     #f #f (x foo-x))
  (define-record-type (bar foo) #f #f (x foo-x))
As in any define-... form, later bindings replace earlier bindings. After the second declaration is executed, foo-x will be bound to the monomorphic accessor applicable only to values of type bar, replacing its binding to the polymorphic accessor procedure introduced in the foo declaration.

Examples

Let's declare a couple of record schemes. Record schemes do not have constructors. They introduce polymorphic predicates and accessors.
  (define-record-scheme <point #f <point? 
    (x <point.x)
    (y <point.y))

  (define-record-scheme <color #f <color?
    (hue <color.hue))
We now declare concrete instances of the above schemes. Constructors may be introduced. Predicates and accessors for concrete record types, when declared, are monomorphic.
  (define-record-type (point <point) make-point point?
    (x point.x)
    (y point.y))

  (define-record-type (color <color) make-color)

  (define-record-type (color-point <color <point) 
                      (make-color-point x y hue) color-point?
    (info color-point.info))

  (define cp (make-color-point 1 2 'blue))

  (<point?          cp)            ==> #t 
  (<color?          cp)            ==> #t
  (<point.y         cp)            ==> 2
  (<color.hue       cp)            ==> blue
  (point?           cp)            ==> #f      
  (point.x          cp)            ==> error   
  (color-point?     cp)            ==> #t
  (color-point.info cp)            ==> <undefined>

Labeled record expressions

The following syntax allows one to construct a record value by labels. The result is a record value of type <type name> with each field <field label> populated with the value of the corresponding <expression>. The order of evaluation of the expressions <expression> ... is undefined. All the <field label>s have to belong to the record type <type name>. If this condition is not satisfied, an expansion time error must be signaled. The runtime efficiency of a labeled record expression is required to be at least that of the equivalent positional constructor.
   <expression> -> (<type name> (<field label> <expression>) ...)

Rationale

The traditional practice of instantiating record values with a positional constructor procedure can lead to code that is hard to read and fragile under common operations such as adding, removing, or rearranging field declarations. The ability to populate record values by labels provides a more robust and readable alternative, especially useful when a record has more than two or three fields, or if it inherits fields from a type scheme. Field labels are checked for validity and the macro may be compiled to a positional constructor at expansion time, thus eliminating a large class of potential programmer errors at no cost in efficiency.

Example

  (color-point (info 'hi) 
               (x 1) 
               (y 2))  
            
                 ==> (color-point (hue <undefined>) (x 1) (y 2) (info hi)) 

Record update

The following syntax allows different forms of record update:
   <expression> -> (record-update  <record> <scheme name> (<field label> <expression>) ...)
                -> (record-update  <record> <type name>   (<field label> <expression>) ...)
                -> (record-update! <record> <type name>   (<field label> <expression>) ...)
                -> (record-update! <record> <scheme name> (<field label> <expression>) ...)
The first alternative is used for polymorphic functional record update. The expression <record> must evaluate to a record value that conforms to <scheme name>. The result will be a new record value of the same type as the original <record>, with the given fields updated. The original record value is unaffected. All the <field label>s have to belong to the record type scheme <scheme name>. If this condition is not satisfied, an expansion time error must be signaled.

The second alternative is used for monomorphic functional record update. The expression <record> must evaluate to a record value of type <type name>. The result will be a new record value of type <type name>, with the given fields updated. The original record value is unaffected. All the <field label>s have to belong to the record type <type name>. If this condition is not satisfied, an expansion time error must be signaled.

The third and fourth alternatives are used for linear, in-place record update. The expression <record> must evaluate to a record value of type <type name> or conforming to scheme <scheme name> . The result will be the original record value with the given fields mutated in place. Note that a useful value is returned. All the <field label>s have to belong to the record type <type name> or scheme <scheme name>. If this condition is not satisfied, an expansion time error must be signaled.

In these forms, the order of evaluation of the expressions <expression> ... is undefined.

Rationale

A mechanism for functional update facilitates and encourages functional-style programming with records.

Note that polymorphic record update is not reducible to the other operations we have listed and therefore has to be provided as a built-in primitive [2].

The linear version update! is provided especially for cases where the programmer knows that no other references to a value exist to produce what is, observationally, a pure-functional result. In these cases, an update operation may be replaced by update! for efficiency. See SRFI-1 for a good discussion of the rationale behind linear update procedures. Note, however, that in contrast with the linear procedures in SRFI-1, update! here is required to mutate the original record.

Examples

Monomorphic update:

  (define p (point (x 1) (y 2)))

  (record-update p point (x 7))      ==> (point (x 7) (y 2))
  p                                  ==> (point (x 1) (y 2))   - original unaffected
Polymorphic update:
  (define cp (color-point (hue 'blue) (x 1) (y 2)))
 
  (record-update cp <point (x 7))    ==> (color-point (info <undefined>) (hue blue) (x 7) (y 2))
  cp                                 ==> (color-point (info <undefined>) (hue blue) (x 1) (y 2))
In-place update:
  (record-update! cp <point (x 7)))  ==> (color-point (info <undefined>) (hue blue) (x 7) (y 2))
  cp                                 ==> (color-point (info <undefined>) (hue blue) (x 7) (y 2))

Record composition

The following syntax provides a shorthand for composing record values:
   <expression> -> (record-compose (<import name> <record>) 
                                   ...
                                   (<export-type name> (<field label> <expression>) ...)

   <import name> -> <type name>
                 -> <scheme name>
Here each expression <record> must evaluate to a record value of type <type name> or conforming to type scheme <scheme name>. The expression evaluates to a new record value of type <export-type name> whose fields are populated as follows: For each field label belonging to <import name> that is also a field label of the type <export-type name>, the corresponding field of <record> is copied into the result. This is done for all imports from left to right, dropping any repeated fields. The additional fields <field label> are then populated with the value of the corresponding <expression>, overwriting any fields with the same labels already imported. Any remaining fields are undefined. All the <field label>s have to belong to the record type <export type name>. If this condition is not satisfied, an expansion time error must be signaled.

The order of evaluation of the expressions <record> ... <expression> ... is undefined. All these expressions will be evaluated, even if their values might not be used in the result.

Rationale

Monomorphic record update is a special case of record-compose. The latter may be used to express more general updates polymorphic in the argument but monomorphic in the result type.

Calculi for composing record values, such as the above scheme, may be used, for example, as units are used in PLT Scheme, or for writing what amounts to modules and functors in the sense of ML.

Examples

Use record-compose for updates polymorphic in the argument but monomorphic in the result type:
  (define cp (make-color-point 1 2 'green))

  (record-compose (<point cp) (point (x 8)))   ==> (point (x 8) (y 2))
A more general composition example:
  (define cp (make-color-point 1 2 'green))
  (define c  (make-color 'blue))
  
  (record-compose (<point cp)                ; polymorphic import - only fields x and y of cp taken
                  (color  c)                 ; monomorphic import
                  (color-point (x 8)         ; overrides imported field
                               (info 'hi)))                 
                                      
                                         ==> (color-point (info hi) (hue blue) (x 8) (y 2))
Small module-functor example:
  
  (define-record-type monoid #f #f 
    (mult monoid.mult) 
    (one  monoid.one))

  (define-record-type abelian-group #f #f 
    (add  group.add) 
    (zero group.zero)
    (sub  group.sub))

  (define-record-type ring #f #f
    (mult ring.mult) 
    (one  ring.one)
    (add  ring.add) 
    (zero ring.zero)
    (sub  ring.sub))

  (define integer-monoid (monoid (mult *) 
                                 (one  1)))

  (define integer-group (abelian-group (add  +)
                                       (zero 0)
                                     (sub  -)))

  (define (make-ring g m)          ; simple functor a la ML
    (record-compose (monoid m)
                    (abelian-group g)
                    (ring)))

  (define integer-ring (make-ring integer-group 
                                  integer-monoid))
  
  ((ring.add integer-ring) 1 2)    ==> 3

Implementation

The reference implementation uses the macro mechanism of R5RS. It assumes an existing implementation of SRFI-9, here denoted srfi-9:define-record-type. It also has a trivial use of case-lambda from SRFI-16.

This version depends on define being treated as a binding form by syntax-rules. This is true for recent versions of portable syntax-case as used in Chez Scheme. It is also true for PLT, for Scheme48, and possibly others.

The SRFI specification was designed with the constraint that all record expressions containing field labels be translatable into positional expressions at macro-expansion time. For example, labeled record expressions and patterns should be just as efficient as positional constructors and patterns. This is true for the reference implementation.

Only the names mentioned in the specification should be visible to the user. Other names should be hidden by a module system or naming convention.

The last section contains a few examples and (non-exhaustive) tests.

Reference implementation

;============================================================================================
; IMPLEMENTATION:
;
; Andre van Tonder, 2004.
;
;============================================================================================

(define-syntax define-record-type    
  (syntax-rules ()
    ((define-record-type . body)
     (parse-declaration #f . body))))

(define-syntax define-record-scheme    
  (syntax-rules ()
    ((define-record-scheme . body)
     (parse-declaration #t . body))))

(define-syntax parse-declaration    
  (syntax-rules ()
    ((parse-declaration is-scheme? (name super ...) constructor-clause predicate field-clause ...)
     (build-record 0 constructor-clause (super ...) (field-clause ...) name predicate is-scheme?))
    ((parse-declaration is-scheme? (name super ...) constructor-clause)
     (parse-declaration is-scheme? (name super ...) constructor-clause #f))  
    ((parse-declaration is-scheme? (name super ...))
     (parse-declaration is-scheme? (name super ...) #f #f))
    ((parse-declaration is-scheme? name . rest)
     (parse-declaration is-scheme? (name) . rest))))

(define-syntax record-update!
  (syntax-rules ()
    ((record-update! record name (label exp) ...)
     (meta
      `(let ((r record)) 
         ((meta ,(name ("setter") label)) r exp)
         ...
         r)))))

(define-syntax record-update
  (syntax-rules ()
    ((record-update record name (label exp) ...)
     (name ("is-scheme?")
           (meta                                                         
            `(let ((new ((meta ,(name ("copier"))) record)))
               (record-update! new name (label exp) ...)))
           (record-compose (name record) (name (label exp) ...))))))    
           
(define-syntax record-compose
  (syntax-rules ()
    ((record-compose (export-name (label exp) ...))
     (export-name (label exp) ...))
    ((record-compose (import-name record) ... (export-name (label exp) ...))
     (help-compose 1 (import-name record) ... (export-name (label exp) ...)))))

(define-syntax help-compose
  (syntax-rules ()
    ((help-compose 1 (import-name record) import ... (export-name (label exp) ...))
     (meta
      `(help-compose 2
                     (meta ,(intersection
                             (meta ,(export-name ("labels")))
                             (meta ,(remove-from (meta ,(import-name ("labels")))
                                                 (label ...)
                                                 if-free=))
                             if-free=))
                     (import-name record) 
                     import ...
                     (export-name (label exp) ...))))
    ((help-compose 2 (copy-label ...) (import-name record) import ... (export-name . bindings))
     (meta
      `(let ((r record))
         (record-compose import ...
           (export-name (copy-label ((meta ,(import-name ("getter") copy-label)) r))
                        ...
                        . bindings)))))))

(define-syntax build-record
  (syntax-rules ()
   ((build-record 0 (constructor . pos-labels) . rest)              ; extract positional labels from constructor clause
    (build-record 1 (constructor . pos-labels) pos-labels . rest))  ; 
   ((build-record 0 constructor . rest)                             ; 
    (build-record 1 (constructor . #f) () . rest))                  ; 
   ((build-record 1 constructor-clause (pos-label ...) (super ...)  
                    ((label . accessors) ...) . rest)
    (meta 
     `(build-record 2
                    constructor-clause
                    (meta ,(union (meta ,(super ("labels")))        ; compute union of labels from supers,
                                  ...                               ; constructor clause and field clauses
                                  (pos-label ...) 
                                  (label ...)      
                                  top:if-free=))
                    ((label . accessors) ...)
                    (meta  ,(union (meta ,(super ("supers")))       ; compute transitive union of supers
                                   ...
                                   top:if-free=))
                    . rest)))
    ((build-record 2 (constructor . pos-labels) labels . rest)      ; insert default constructor labels if not given
     (syntax-if pos-labels
                (build-record 3 (constructor . pos-labels) labels . rest)
                (build-record 3 (constructor . labels)     labels . rest)))
    ((build-record 3 constructor-clause labels ((label . accessors) ...) . rest)
     (meta 
      `(build-record 4
                     (meta ,(remove-from labels                     ; separate the labels that do not appear in a
                                         (label ...)                ; field clause for next step
                                         top:if-free=))
                     ((label . accessors) ...) 
                     constructor-clause
                     labels
                     . rest)))
    ((build-record 4
                   (undeclared-label ...)
                   (field-clause ...)
                   (constructor . pos-labels)
                   labels
                   supers
                   name
                   predicate
                   is-scheme?)
     (meta
      `(build-record 5                                              ; generate identifiers for constructor, predicate
                     is-scheme?                                     ; getters and setters as needed 
                     name
                     supers
                     supers
                     labels 
                     (meta ,(to-identifier constructor))   
                     (meta ,(add-temporaries pos-labels))           ; needed for constructor below
                     (meta ,(to-identifier predicate))
                     (meta ,(augment-field field-clause)) 
                     ... 
                     (undeclared-label (meta ,(generate-identifier))
                                       (meta ,(generate-identifier)))
                     ...)))
    ((build-record 5
                   is-scheme?
                   name
                   (super ...)
                   supers
                   (label ...)
                   constructor  
                   ((pos-label pos-temp) ...) 
                   predicate
                   (field-label getter setter)
                   ...)  
     
     (begin
       (syntax-if is-scheme?
                  
                  (begin
                    (define-generic (predicate x) (lambda (x) #f))
                    (define-generic (getter x))
                    ...
                    (define-generic (setter x v))
                    ...
                    (define-generic (copy x)))
                  
                  (begin
                    (srfi-9:define-record-type internal-name
                                               (maker field-label ...)
                                               predicate
                                               (field-label getter setter) ...)  
       
                    (define constructor 
                      (lambda (pos-temp ...)
                        (populate 1 maker (field-label ...) (pos-label pos-temp) ...)))
       
                    (extend-predicates supers predicate)
                    (extend-accessors supers field-label predicate getter setter)
                    ...
       
                    (define (copy x)
                      (maker (getter x) ...))
                    (extend-copiers supers copy predicate)
   
                    (define-method (show (r predicate))
                      (list 'name
                            (list 'field-label (getter r)) 
                            ...))))    
       
       (define-syntax name
         (syntax-rules (field-label ...)
           ((name ("is-scheme?") sk fk)     (syntax-if is-scheme? sk fk))
           ((name ("predicate") k)          (syntax-apply k predicate))
           ((name ("supers") k)             (syntax-apply k (super ... name)))  
           ((name ("labels") k)             (syntax-apply k (label ...)))
           ((name ("pos-labels") k)         (syntax-apply k (pos-label ...)))
           ((name ("getter") field-label k) (syntax-apply k getter))   
           ...
           ((name ("getter") other k)       (syntax-apply k #f))
           ((name ("setter") field-label k) (syntax-apply k setter))  
           ...
           ((name ("setter") other k)       (syntax-apply k #f))
           ((name ("copier") k)             (syntax-apply k copy))
           ((name . bindings)               (populate 1 maker (field-label ...) . bindings))))))))


(define-syntax to-identifier
  (syntax-rules ()
    ((to-identifier #f k) (syntax-apply k generated-identifier))
    ((to-identifier id k) (syntax-apply k id))))

(define-syntax augment-field 
  (syntax-rules ()
    ((augment-field (label) k)               (syntax-apply k (label generated-getter generated-setter)))
    ((augment-field (label getter) k)        (meta `(label (meta ,(to-identifier getter)) generated-setter) k))
    ((augment-field (label getter setter) k) (meta `(label (meta ,(to-identifier getter)) 
                                                           (meta ,(to-identifier setter))) k))))

(define-syntax extend-predicates
  (syntax-rules ()
    ((extend-predicates (super ...) predicate)
     (begin
       (meta
        `(define-method (meta ,(super ("predicate")))
                        (predicate)
                        (x)
                        any?))   
       ...))))

(define-syntax extend-copiers
  (syntax-rules ()
    ((extend-copiers (super ...) copy predicate)
     (begin
       (meta
        `(define-method (meta ,(super ("copier")))
                        (predicate)
                        (x)
                        copy))    
       ...))))

(define-syntax extend-accessors
  (syntax-rules ()
    ((extend-accessors (super ...) label predicate selector modifier)
     (meta
      `(begin 
         (syntax-if (meta ,(super ("getter") label))
                    (define-method (meta ,(super ("getter") label))
                                   (predicate)
                                   (x)
                                   selector)
                    (begin))
         ...
         (syntax-if (meta ,(super ("setter") label))
                    (define-method (meta ,(super ("setter") label))
                                   (predicate any?)
                                   (x v)
                                   modifier)
                    (begin))
         ...)))))

(define-syntax populate
  (syntax-rules ()
    ((populate 1 maker labels . bindings)
     (meta 
      `(populate 2 maker
                   (meta ,(order labels bindings ('<undefined>))))))
    ((populate 2 maker ((label exp) ...))
     (maker exp ...))))

(define-syntax order
  (syntax-rules ()
    ((order (label ...) ((label* . binding) ...) default k)
     (meta
      `(if-empty? (meta ,(remove-from (label* ...) 
                                      (label ...) 
                                      if-free=))
                  (order "emit" (label ...) ((label* . binding) ...) default k)
                  (syntax-error "Illegal labels in" ((label* . binding) ...)
                                "Legal labels are" (label ...)))))
    ((order "emit" (label ...) bindings default k)
     (meta 
      `((label . (meta ,(syntax-lookup label 
                                       bindings 
                                       if-free= 
                                       default)))
        ...)
      k))))


;============================================================================================
; Simple generic functions:

(define-syntax define-generic
  (syntax-rules ()
    ((define-generic (name arg ...))
     (define-generic (name arg ...)
       (lambda (arg ...) (error "Inapplicable method:" 'name
                                "Arguments:" (show arg) ... ))))
    ((define-generic (name arg ...) proc)
     (define name (make-generic (arg ...) proc)))))  
  
(define-syntax define-method
  (syntax-rules ()
    ((define-method (generic (arg pred?) ...) . body)
     (define-method generic (pred? ...) (arg ...) (lambda (arg ...) . body))) 
    ((define-method generic (pred? ...) (arg ...) procedure)
     (let ((next ((generic) 'get-proc))
           (proc procedure))
       (((generic) 'set-proc)
        (lambda (arg ...)
          (if (and (pred? arg) ...)
              (proc arg ...)
              (next arg ...))))))))

(define-syntax make-generic
  (syntax-rules ()
    ((make-generic (arg arg+ ...) default-proc)
     (let ((proc default-proc))
       (case-lambda
         ((arg arg+ ...)
          (proc arg arg+ ...))
         (()
          (lambda (msg)
            (case msg
              ((get-proc) proc)
              ((set-proc) (lambda (new)
                            (set! proc new)))))))))))

(define-generic (show x) 
  (lambda (x) (x)))

(define (any? x) #t)


;============================================================================================
; Syntax utilities:

(define-syntax syntax-error
  (syntax-rules ()))

(define-syntax syntax-apply
  (syntax-rules ()
    ((syntax-apply (f . args) exp ...) 
     (f exp ... . args))))

(define-syntax syntax-cons
  (syntax-rules ()
    ((syntax-cons x rest k) 
     (syntax-apply k (x . rest)))))

(define-syntax syntax-cons-after
  (syntax-rules ()
    ((syntax-cons-after rest x k)
     (syntax-apply k (x . rest)))))

(define-syntax if-empty?
  (syntax-rules ()
    ((if-empty? () sk fk)      sk)
    ((if-empty? (h . t) sk fk) fk)))

(define-syntax add-temporaries   
  (syntax-rules () 
    ((add-temporaries lst k)                (add-temporaries lst () k))
    ((add-temporaries () lst-temps k)       (syntax-apply k lst-temps))
    ((add-temporaries (h . t) (done ...) k) (add-temporaries t (done ... (h temp)) k))))

(define-syntax if-free=
  (syntax-rules ()
    ((if-free= x y kt kf)
      (let-syntax
          ((test (syntax-rules (x)
                   ((test x kt* kf*) kt*)
                   ((test z kt* kf*) kf*))))
        (test y kt kf)))))

(define-syntax top:if-free=
  (syntax-rules ()
    ((top:if-free= x y kt kf)
     (begin
       (define-syntax if-free=:test
         (syntax-rules (x)
           ((if-free=:test x kt* kf*) kt*)
           ((if-free=:test z kt* kf*) kf*)))
       (if-free=:test y kt kf)))))

(define-syntax meta
  (syntax-rules (meta quasiquote unquote)
    ((meta `(meta ,(function argument ...)) k)
     (meta `(argument ...) (syntax-apply-to function k)))
    ((meta `(a . b) k)
     (meta `a (descend-right b k)))
    ((meta `whatever k) (syntax-apply k whatever))
    ((meta `arg)
     (meta `arg (syntax-id)))))

(define-syntax syntax-apply-to
  (syntax-rules ()
    ((syntax-apply-to (argument ...) function k)
     (function argument ... k))))

(define-syntax descend-right
  (syntax-rules ()
    ((descend-right evaled b k)
     (meta `b (syntax-cons-after evaled k)))))

(define-syntax syntax-id
  (syntax-rules ()
    ((syntax-id arg) arg))) 

(define-syntax remove-duplicates
  (syntax-rules ()
    ((remove-duplicates lst compare? k)
     (remove-duplicates lst () compare? k))
    ((remove-duplicates () done compare? k)
     (syntax-apply k done))
    ((remove-duplicates (h . t) (d ...) compare? k)
     (if-member? h (d ...) compare? 
                 (remove-duplicates t (d ...) compare? k)
                 (remove-duplicates t (d ... h) compare? k)))))

(define-syntax syntax-filter
  (syntax-rules ()
    ((syntax-filter () (if-p? arg ...) k)
     (syntax-apply k ()))
    ((syntax-filter (h . t) (if-p? arg ...) k)
     (if-p? h arg ...
            (syntax-filter t (if-p? arg ...) (syntax-cons-after h k))
            (syntax-filter t (if-p? arg ...) k)))))

(define-syntax if-member?
  (syntax-rules ()
    ((if-member? x () compare? sk fk) 
     fk)
    ((if-member? x (h . t) compare? sk fk)
     (compare? x h
               sk
               (if-member? x t compare? sk fk)))))

(define-syntax union
  (syntax-rules ()
    ((union (x ...) ... compare? k)
     (remove-duplicates (x ... ...) compare? k))))

(define-syntax intersection
  (syntax-rules ()
    ((intersection list1 list2 compare? k)
     (syntax-filter list1 (if-member? list2 compare?) k))))

(define-syntax remove-from
  (syntax-rules ()
    ((remove-from list1 list2 compare? k)
     (syntax-filter list1 (if-not-member? list2 compare?) k))))

(define-syntax if-not-member?
  (syntax-rules ()
    ((if-not-member? x list compare? sk fk)
     (if-member? x list compare? fk sk))))

(define-syntax generate-identifier
  (syntax-rules ()
    ((generate-identifier k) (syntax-apply k generated-identifier))))

(define-syntax syntax-if
  (syntax-rules ()
    ((syntax-if #f sk fk)    fk)
    ((syntax-if other sk fk) sk)))

(define-syntax syntax-lookup
  (syntax-rules ()
    ((syntax-lookup label () compare fail k)
     (syntax-apply k fail))
    ((syntax-lookup label ((label* . value) . bindings) compare fail k)
     (compare label label*
              (syntax-apply k value)
              (syntax-lookup label bindings compare fail k)))))

Tests and examples

;============================================================================================
; Examples:

; A simple record declaration:

(define-record-type point (make-point x y) point?
  (x point.x point.x-set!)
  (y point.y point.y-set!))

(define p (make-point 1 2))

(point? p)             ;==> #t
(point.y p)            ;==> 2
(point.y-set! p 7)
(point.y p)            ;==> 7

; Simple record schemes.
; Record schemes don't have constructors.
; The predicates and accessors are polymorphic.

(define-record-scheme <point #f <point? 
  (x <point.x)
  (y <point.y))

(define-record-scheme <color #f <color?
  (hue <color.hue))

; Concrete instances of the above schemes.
; Constructors may be declared.
; Predicates and accessors, when provided, are monomorphic.  

(define-record-type (point <point) make-point point?
  (x point.x)
  (y point.y))

(define-record-type (color <color) make-color)

(define-record-type (color-point <color <point) (make-color-point x y hue) color-point?
  (extra color-point.extra))

(define cp (make-color-point 1 2 'blue))

(<point? cp)            ;==> #t         
(<color? cp)            ;==> #t
(color-point? cp)       ;==> #t
;(point.x cp)           ;==> error 
(<point.y cp)           ;==> 2
(<color.hue cp)         ;==> blue
(color-point.extra cp)  ;==> <undefined>

; Constructing records by field labels:

(define p (point (x 1) 
                 (y 2)))
(define cp (color-point (hue 'blue) 
                        (x 1) 
                        (y 2)))

; Monomorphic functional update:

(show
 (record-update p point (x 7)))     ;==> (point (x 7) (y 2))
(show p)                            ;==> (point (x 1) (y 2))   - original unaffected

; Polymorphic functional update:

(show 
 (record-update cp <point (x 7)))   ;==> (color-point (extra <undefined>) (hue blue) (x 7) (y 2))
(show cp)                           ;==> (color-point (extra <undefined>) (hue blue) (x 1) (y 2))

; In-place update:

(show 
 (record-update! cp <point (x 7)))  ;==> color-point (extra <undefined>) (hue blue) (x 7) (y 2))
(show cp)                           ;==> color-point (extra <undefined>) (hue blue) (x 7) (y 2))
 
; Use record-compose for updates polymorphic in argument but monomorphic in result type:

(show
 (record-compose (<point cp) (point (x 8))))  ;==> (point (x 8) (y 2))
(show cp)                                     ;==> (color-point (extra <undefined>) (hue blue) (x 7) (y 2))

; More general record composition example:

(define cp (make-color-point 1 2 'green))
(define c  (make-color 'blue))
 
(show 
 (record-compose (<point cp)                 ; polymorphic import - only fields x and y of cp taken
                 (color c)                   ; monomorphic import
                 (color-point (x 8)          ; override imported field
                              (extra 'hi))))                 
                                      
                                         ;==> (color-point (extra hi) (hue blue) (x 8) (y 2))

; Small module-functor example:
  
(define-record-type monoid #f #f 
  (mult monoid.mult) 
  (one  monoid.one))

(define-record-type abelian-group #f #f 
  (add  group.add) 
  (zero group.zero)
  (sub  group.sub))

(define-record-type ring #f #f
  (mult ring.mult) 
  (one  ring.one)
  (add  ring.add) 
  (zero ring.zero)
  (sub  ring.sub))

(define integer-monoid (monoid (mult *) 
                               (one  1)))

(define integer-group (abelian-group (add  +)
                                     (zero 0)
                                     (sub  -)))

(define (make-ring g m)          ; simple "functor"
  (record-compose (monoid m)
                  (abelian-group g)
                  (ring)))

(define integer-ring (make-ring integer-group 
                                integer-monoid))
  
((ring.add integer-ring) 1 2)    ;==> 3

; Example of tree data type

(define-record-scheme <tree #f <tree?) 

(define-record-type (node <tree) make-node node?
  (lhs node.lhs)
  (rhs node.rhs))

(define-record-type (leaf <tree) make-leaf leaf?
  (val leaf.val))

(define (tree->list t)
  (cond
    ((leaf? t) (leaf.val t))
    ((node? t) (cons (tree->list (node.lhs t))
                     (tree->list (node.rhs t))))))

(define t 
  (make-node (make-node (make-leaf 1)
                        (make-leaf 2))
             (make-leaf 3)))

(<tree? t)         ;==> #t
(tree->list t)     ;==> ((1 . 2) . 3)

References

[1] Richard Kelsey, Defining Record Types, SRFI-9: http://srfi.schemers.org/srfi-9/srfi-9.html

[2] See e.g.
    Benjamin C. Pierce, Types and Programming Languages, MIT Press 2002, and references therein.
    Mitchell Wand, Type inference for record concatenation and multiple inheritance, 
                   Information and Computation, v.93 n.1, p.1-15, July 1991
    John Reppy, Jon Riecke, Simple objects for Standard ML,
                Proceedings of the ACM SIGPLAN '96 Conference on Programming Language Design and Implementation


Copyright

Copyright (C) Andre van Tonder (2004). All Rights Reserved.

This document and translations of it may be copied and furnished to others, and derivative works that comment on or otherwise explain it or assist in its implementation may be prepared, copied, published and distributed, in whole or in part, without restriction of any kind, provided that the above copyright notice and this paragraph are included on all such copies and derivative works. However, this document itself may not be modified in any way, such as by removing the copyright notice or references to the Scheme Request For Implementation process or editors, except as needed for the purpose of developing SRFIs in which case the procedures for copyrights defined in the SRFI process must be followed, or as required to translate it into languages other than English.

The limited permissions granted above are perpetual and will not be revoked by the authors or their successors or assigns.

This document and the information contained herein is provided on an "AS IS" basis and THE AUTHOR AND THE SRFI EDITORS DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.


Author: André van Tonder
Editor: David Van Horn
Last modified: Fri Nov 26 10:45:58 EST 2004