racket/doc/srfi-std/srfi-57.html
Matthew Flatt 28a3f3f0e7 r5rs and srfi docs and bindings
svn: r9336
2008-04-16 20:52:39 +00:00

1390 lines
53 KiB
HTML

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML><HEAD><title>SRFI 57: Records</title>
</HEAD>
<BODY>
<H1>Title</H1>
Records
<H1>Author</H1>
Andr&eacute; van Tonder
<H1>Status</H1>
This SRFI is currently in ``final'' status. To see an explanation of each
status that a SRFI can hold, see <A
href="http://srfi.schemers.org/srfi-process.html">here</A>. You can access
previous messages via <A
href="http://srfi.schemers.org/srfi-57/mail-archive/maillist.html"> the
archive of the mailing list</A>.
<P>
<UL>
<LI>Received: <A HREF="http://srfi.schemers.org/cgi-bin/viewcvs.cgi/*checkout*/srfi/srfi-57/srfi-57.html?rev=1.1">2004/09/08</A></LI>
<LI>Draft: 2004/09/08 - 2005/01/08</LI>
<LI>Revised: <A HREF="http://srfi.schemers.org/cgi-bin/viewcvs.cgi/*checkout*/srfi/srfi-57/srfi-57.html?rev=1.2">2004/09/22</A></LI>
<LI>Revised: <A HREF="http://srfi.schemers.org/cgi-bin/viewcvs.cgi/*checkout*/srfi/srfi-57/srfi-57.html?rev=1.3">2004/11/08</A></LI>
<LI>Revised: <A HREF="http://srfi.schemers.org/cgi-bin/viewcvs.cgi/*checkout*/srfi/srfi-57/srfi-57.html?rev=1.6">2004/11/25</A></LI>
<LI>Revised: <A HREF="http://srfi.schemers.org/cgi-bin/viewcvs.cgi/*checkout*/srfi/srfi-57/srfi-57.html?rev=1.7">2004/12/08</A></LI>
<LI>Revised: <A HREF="http://srfi.schemers.org/cgi-bin/viewcvs.cgi/*checkout*/srfi/srfi-57/srfi-57.html?rev=1.8">2005/03/07</A></LI>
<LI>Final: 2005/03/07</LI>
</UL>
<H1>Abstract</H1>
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.
<H1>Rationale</H1>
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.
<ul>
<li>A syntax for constructing record values by field label. Such a
feature is common in the prior art [2]. Indeed, the ability to
address fields by labels instead of position is one of the main
reasons for introducing the concept of records in the first
place.
Taking advantage of this feature can ease coding and increase readability and
robustness of code.
</li><li>
A mechanism for defining and incrementally extending
polymorphic functions on families
of record types sharing a set of field labels. This automates
a common programming pattern which would otherwise be laborious to
code explicitly.
</li><li>Facilities for record composition, functional record update,
including polymorphic
update, and linear in-place update.
These facilitate functional style programming with records.
</li><li>A declaration syntax that allows shorter variations and
optional elements, to save labor and reduce namespace
pollution.
</li></ul>
<H1>Specification</H1>
<h2>Declaration</h2>
<pre> &lt;command or definition&gt;
-&gt; &lt;record type definition&gt;
-&gt; &lt;record scheme definition&gt; ; addition to 7.1.6 in R5RS
&lt;record type definition&gt; -&gt; (define-record-type &lt;type clause&gt;
&lt;constructor clause&gt;
&lt;predicate clause&gt;
&lt;field clause&gt; ...)
-&gt; (define-record-type &lt;type clause&gt;
&lt;constructor clause&gt;)
-&gt; (define-record-type &lt;type clause&gt;)
&lt;record scheme definition&gt; -&gt; (define-record-scheme &lt;scheme clause&gt;
&lt;deconstructor clause&gt;
&lt;predicate clause&gt;
&lt;field clause&gt; ...)
-&gt; (define-record-scheme &lt;scheme clause&gt;
&lt;deconstructor clause&gt;)
-&gt; (define-record-scheme &lt;scheme clause&gt;)
&lt;type clause&gt; -&gt; &lt;type name&gt;
-&gt; (&lt;type name&gt; &lt;scheme name&gt; ...)
&lt;scheme clause&gt; -&gt; &lt;scheme name&gt;
-&gt; (&lt;scheme name&gt; &lt;parent scheme name&gt; ...)
&lt;constructor clause&gt; -&gt; (&lt;constructor name&gt; &lt;field label&gt; ...)
-&gt; &lt;constructor name&gt;
-&gt; #f
&lt;deconstructor clause&gt; -&gt; (&lt;deconstructor name&gt; &lt;field label&gt; ...)
-&gt; &lt;deconstructor name&gt;
-&gt; #f
&lt;predicate clause&gt; -&gt; &lt;predicate name&gt;
-&gt; #f
&lt;field clause&gt; -&gt; (&lt;field label&gt; &lt;accessor clause&gt; &lt;modifier clause&gt;)
-&gt; (&lt;field label&gt; &lt;accessor clause&gt;)
-&gt; (&lt;field label&gt;)
&lt;accessor clause&gt; -&gt; &lt;accessor name&gt;
-&gt; #f
&lt;modifier clause&gt; -&gt; &lt;modifier name&gt;
-&gt; #f
&lt;field label&gt; -&gt; &lt;identifier&gt;
&lt;... name&gt; -&gt; &lt;identifier&gt;
</pre>
<h3>Record types</h3>
An instance of <code>define-record-type</code> is equivalent to the following:
<ul>
<li> A list of field labels is associated with the record type <code>&lt;type name&gt;</code>,
obtained by appending from left to right the lists of field labels
of any record
type schemes (see below) appearing in the <code>&lt;type clause&gt;</code>,
followed by the list of labels in the
<code>&lt;constructor clause&gt;</code>, followed by the labels
in order of appearance in the <code>&lt;field
clause&gt;</code>s.
Duplicates are removed from the resulting list according
to the semantics of <code>delete-duplicates</code> of SRFI-1.
Labels in the constructor clause must be
distinct. Labels in the field clauses must also be distinct.
<p>
</p></li><li>For each <code>&lt;scheme name&gt;</code> in <code>&lt;type clause&gt;</code>, the record type
<code>&lt;type name&gt;</code> is said to be an <i>instance</i> of, or to
<i>conform</i> to the corresponding
record type scheme <code>&lt;scheme name&gt;</code> and to all
parent type schemes (see below) of <code>&lt;scheme name&gt;</code>.
<p></p></li><li> <code>&lt;type name&gt;</code> is bound to a macro, described below, that can be used to construct record
values by label. It may also be registered, as specified in a
future SRFI, for performing pattern matching on record values of
type <code>&lt;type name&gt;</code>.
<p>
</p></li><li> If <code>&lt;constructor clause&gt;</code> is
of the form <code>(&lt;constructor name&gt; &lt;field label&gt; ...)</code>, then
<code>&lt;constructor name&gt;</code> is bound to a procedure that takes as many arguments as
there are <code>&lt;field label&gt;</code>s following it
and returns a new <code>&lt;type name&gt;</code> record.
Fields whose labels are listed with <code>&lt;type name&gt;</code> have the corresponding
argument as their initial value. The initial values of all other fields are unspecified.
If <code>&lt;constructor clause&gt;</code> is of the form <code>&lt;constructor name&gt;</code>,
the procedure
<code>&lt;constructor name&gt;</code> takes as many arguments as there are field labels
associated with <code>&lt;type name&gt;</code>, in the default order defined above.
<p></p><p>
<code>&lt;constructor name&gt;</code> may be
registered, in a way to be described in a future SRFI, for performing a
positional pattern match of the fields <code>&lt;field label&gt; ...</code>
of record
values of type <code>&lt;type name&gt;</code> in the first case,
or of all fields
associated with <code>&lt;scheme name&gt;</code> in the default
order defined above in the second case.
</p></li><li> <code>&lt;predicate name&gt;</code>, is bound to a predicate procedure
that returns <code>#t</code> when given a record value that has been constructed using
the macro <code>&lt;type name&gt;</code> or the procedure <code>&lt;constructor name&gt;</code>,
and <code>#f</code> for any other
value. Values on which <code>&lt;predicate name&gt;</code>, if applied, would return
<code>#t</code>, are said to be <i>of type</i> <code>&lt;type name&gt;</code>.
<p>
</p></li><li> Field labels inherited from a <code>&lt;type scheme&gt;</code> or
introduced in the <code>&lt;constructor clause&gt;</code> do not have to be
repeated in the
<code>&lt;field clause&gt;</code>s.
Where present, <code>&lt;field
clause&gt;</code>s may provide additional information on such fields, or may
declare additional fields.
<p></p><p>
Field labels may be reused as the name of accessors or modifiers (a practice known
as punning).
</p><p>
</p><ul>
<p></p><li> Each <code>&lt;accessor name&gt;</code> is bound to
a procedure that takes a
value of type <code>&lt;type name&gt;</code>,
and returns the current value of the corresponding
field. It is an error to pass an accessor a value not of type
<code>&lt;type name&gt;</code>.
<p></p></li><li> Each <code>&lt;modifier name&gt;</code> is bound to
a procedure that takes a value of type <code>&lt;type name&gt;</code>
and a value which becomes the new value of the corresponding field.
It is an error to pass a modifier a first argument that is not of type
<code>&lt;type name&gt;</code>.
The return value of <code>&lt;modifier name&gt;</code> is unspecified.
<p>
</p></li></ul></li></ul>
<p>
<code>Define-record-type</code> 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. This SRFI only
specifies the behaviour of <code>define-record-type</code> at
top-level.
</p><h3>Record type schemes</h3>
An instance of <code>define-record-scheme</code> is equivalent to the following:
<ul>
<li> A list of field labels is associated with the type scheme <code>&lt;scheme name&gt;</code>,
obtained by appending from left to right the lists of field labels
of any parent
type schemes appearing in the <code>&lt;scheme clause&gt;</code>,
followed by the list of labels in the
<code>&lt;deconstructor clause&gt;</code>, followed by the labels
in order of appearance in the <code>&lt;field clause&gt;</code>s.
Duplicates are removed from the resulting list according
to the semantics of <code>delete-duplicates</code> of SRFI-1.
Labels in the constructor clause must be
distinct. Labels in the field clauses must also be distinct.
<p>
</p></li><li>A record type scheme is called a <i>parent scheme</i> of
<code>&lt;scheme name&gt;</code> if it appears in the
<code>&lt;scheme clause&gt;</code>, or if it is a parent scheme of
one of the <code>&lt;parent scheme name&gt;</code>'s appearing in the
<code>&lt;scheme clause&gt;</code>.
The type scheme
<code>&lt;scheme name&gt;</code> is said to
<i>extend</i> its parent type schemes. It is an error to extend a type scheme
that has not yet been defined.
<p></p></li><li> <code>&lt;scheme name&gt;</code> may be bound to a macro or otherwise
registered, in a way to be
described in a future
SRFI,
for performing pattern matching on record
values conforming to <code>&lt;scheme name&gt;</code>.
<p>
</p></li><li>If <code>&lt;deconstructor clause&gt;</code> is
of the form <code>(&lt;deconstructor name&gt; &lt;field label&gt; ...)</code>, then
<code>&lt;deconstructor name&gt;</code> may be bound to a macro or otherwise
registered, in a way to be described in a future SRFI, for performing a
positional pattern match of the fields <code>&lt;field label&gt; ...</code>
on record
values conforming to <code>&lt;scheme name&gt;</code>.
If <code>&lt;deconstructor clause&gt;</code> is of the form <code>&lt;deconstructor name&gt;</code>,
the positional match will be on all fields
associated with <code>&lt;scheme name&gt;</code>, in the default order defined above.
<p></p><p>
</p></li><li> <code>&lt;predicate name&gt;</code>, is bound to a predicate procedure
that returns <code>#t</code> when given a record value of any record type conforming
to <code>&lt;scheme name&gt;</code>,
and <code>#f</code> for any other
value.
<p>
</p></li><li> Field labels inherited from a <code>&lt;parent type scheme&gt;</code> or
introduced in the <code>&lt;deconstructor clause&gt;</code> do not have to be
repeated in the
<code>&lt;field clause&gt;</code>s.
Where present, <code>&lt;field
clause&gt;</code>s may provide additional information on such fields, or may
declare additional fields.
<p></p><p>
Field labels may be reused as the name of accessors or modifiers (a practice known
as punning).
</p><p>
</p><ul>
<p></p><li> Each <code>&lt;accessor name&gt;</code> is bound to
a procedure that takes a
value conforming to <code>&lt;scheme name&gt;</code>,
and returns the current value of the corresponding
field. It is an error to pass an accessor a value not conforming to
<code>&lt;scheme name&gt;</code>.
<p></p></li><li> Each <code>&lt;modifier name&gt;</code> is bound to
a procedure that takes a value conforming to <code>&lt;scheme name&gt;</code>
and a value which becomes the new value of the corresponding field.
It is an error to pass a modifier a first argument that does not conform to
<code>&lt;scheme name&gt;</code>.
The return value of <code>&lt;modifier name&gt;</code> is unspecified.
<p>
</p></li></ul></li></ul>
<p></p><h3>Examples</h3>
<h4>A simple record:</h4>
<pre> (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) ==&gt; 2
(set-y! p 3))
(get-y p) ==&gt; 3
(point? p) ==&gt; #t
</pre>
<h4>Record type schemes:</h4>
Let's declare a couple of record schemes. Record schemes do not have constructors.
They introduce polymorphic predicates and accessors.
<pre> (define-record-scheme &lt;point #f &lt;point?
(x &lt;point.x)
(y &lt;point.y))
(define-record-scheme &lt;color #f &lt;color?
(hue &lt;color.hue))
</pre>
We now declare concrete instances of the above schemes.
Constructors may be introduced.
Predicates and accessors for concrete record types, when declared, are monomorphic.
<pre> (define-record-type (point &lt;point) make-point point?
(x point.x)
(y point.y))
(define-record-type (color &lt;color) make-color)
(define-record-type (color-point &lt;color &lt;point)
(make-color-point x y hue) color-point?
(info color-point.info))
(define cp (make-color-point 1 2 'blue))
(&lt;point? cp) ==&gt; #t
(&lt;color? cp) ==&gt; #t
(&lt;point.y cp) ==&gt; 2
(&lt;color.hue cp) ==&gt; blue
(point? cp) ==&gt; #f
(point.x cp) ==&gt; error
(color-point? cp) ==&gt; #t
(color-point.info cp) ==&gt; &lt;undefined&gt;
</pre>
<h4>Optional elements:</h4>
Elements may be left out if not desired, as the following examples illustrate:
<pre>
(define-record-type node (make-node left right))
(define-record-type leaf (make-leaf value))
</pre>
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.
<pre>
(define-record-type monday)
(define-record-type tuesday #f tuesday?)
</pre>
Here <code>monday</code> has no declared constructor or predicate, while <code>tuesday</code>
has a predicate but no constructor.
<pre> (define-record-type node make-node #f
(left left)
(right right))
</pre>
Here the constructor <code>make-node</code> has the default argument order and no predicate
is bound. Also note that field labels are
punned.
<p></p><h4>A note on repeated fields and redefinitions</h4>
In the following example, two record type schemes define different accessors for the same field:
<pre> (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))
</pre>
Since any value <code>fb</code> of type <code>foo-bar</code> conforms to both
<code>foo</code> and <code>bar</code>, both <code>foo-x</code> and <code>bar-x</code>
can be applied to <code>fb</code>, returning the value of the <code>x</code> field.
<p>
In the following example, two declarations introduce
the same accessor:
</p><pre> (define-record-scheme foo #f #f (x foo-x))
(define-record-type (bar foo) #f #f (x foo-x))
</pre>
As in any <code>define-...</code> form, later bindings replace earlier bindings.
After the second declaration is executed,
<code>foo-x</code> will be bound to the monomorphic accessor applicable only to values
of type <code>bar</code>, replacing its binding to the polymorphic accessor procedure
introduced in the <code>foo</code> declaration.
<h2>Labeled record expressions</h2>
The following syntax allows one to construct a record value by labels. The result
is a record value of type <code>&lt;type name&gt;</code> with each field
<code>&lt;field label&gt;</code> populated with the value of the corresponding
<code>&lt;expression&gt;</code>. The order of evaluation of the expressions
<code>&lt;expression&gt; ...</code> is undefined. All the
<code>&lt;field label&gt;</code>s have to belong to the record type <code>&lt;type name&gt;</code>.
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.
<pre> &lt;expression&gt; -&gt; (&lt;type name&gt; (&lt;field label&gt; &lt;expression&gt;) ...)
</pre>
</p><p>
The order of evaluation of the expressions
<code>&lt;expression&gt; ...</code> is undefined.
<h4>Rationale</h4>
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.
<h4>Example</h4>
<pre> (color-point (info 'hi)
(x 1)
(y 2))
==&gt; (color-point (hue &lt;undefined&gt;) (x 1) (y 2) (info hi))
</pre>
<h2>Record update</h2>
The following syntax allows different forms of record update:
<pre> &lt;expression&gt; -&gt; (record-update &lt;record&gt; &lt;scheme name&gt; (&lt;field label&gt; &lt;expression&gt;) ...)
-&gt; (record-update &lt;record&gt; &lt;type name&gt; (&lt;field label&gt; &lt;expression&gt;) ...)
-&gt; (record-update! &lt;record&gt; &lt;type name&gt; (&lt;field label&gt; &lt;expression&gt;) ...)
-&gt; (record-update! &lt;record&gt; &lt;scheme name&gt; (&lt;field label&gt; &lt;expression&gt;) ...)
</pre>
The first alternative is used for polymorphic functional record update. The expression
<code>&lt;record&gt;</code> must evaluate to a record value that conforms to
<code>&lt;scheme name&gt;</code>.
The result will be a new record value of the same type as
the original <code>&lt;record&gt;</code>, with the given fields updated. The original
record value is unaffected. All the
<code>&lt;field label&gt;</code>s have to belong to the record type scheme <code>&lt;scheme name&gt;</code>.
If this condition is not satisfied, an expansion time error must be signaled.
<p>
The second alternative is used for monomorphic functional record update. The expression
<code>&lt;record&gt;</code> must evaluate to a record value of type
<code>&lt;type name&gt;</code>. The result will be a new record value of type
<code>&lt;type name&gt;</code>, with the given fields updated. The original
record value is unaffected. All the
<code>&lt;field label&gt;</code>s have to belong to the record type <code>&lt;type name&gt;</code>.
If this condition is not satisfied, an expansion time error must be signaled.
</p><p>
The third and fourth alternatives are used for linear, in-place record update. The expression
<code>&lt;record&gt;</code> must evaluate to a record value of type
<code>&lt;type name&gt;</code> or conforming to scheme <code>&lt;scheme name&gt;</code> . The result will be the original record value
with the given fields
mutated in place.
Note that a useful value is returned. All the
<code>&lt;field label&gt;</code>s have to belong to the record type <code>&lt;type name&gt;</code> or scheme <code>&lt;scheme name&gt;</code>.
If this condition is not satisfied, an expansion time error must be signaled.
</p><p>
In these forms, the order of evaluation of the expressions
<code>&lt;expression&gt; ...</code> is undefined.
</p><h4>Rationale</h4>
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].
<p></p><p>The linear version
<code>update!</code> 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 <code>update</code>
operation may be replaced by <code>update!</code> 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, <code>update!</code> here is <i>required</i>
to mutate the original record.
</p><h4>Examples</h4>
Monomorphic update:
<pre> (define p (point (x 1) (y 2)))
(record-update p point (x 7)) ==&gt; (point (x 7) (y 2))
p ==&gt; (point (x 1) (y 2)) - original unaffected
</pre>
Polymorphic update:
<pre> (define cp (color-point (hue 'blue) (x 1) (y 2)))
(record-update cp &lt;point (x 7)) ==&gt; (color-point (info &lt;undefined&gt;) (hue blue) (x 7) (y 2))
cp ==&gt; (color-point (info &lt;undefined&gt;) (hue blue) (x 1) (y 2))
</pre>
In-place update:
<pre> (record-update! cp &lt;point (x 7))) ==&gt; (color-point (info &lt;undefined&gt;) (hue blue) (x 7) (y 2))
cp ==&gt; (color-point (info &lt;undefined&gt;) (hue blue) (x 7) (y 2))
</pre>
<h2>Record composition</h2>
The following syntax provides a shorthand for composing record values:
<pre> &lt;expression&gt; -&gt; (record-compose (&lt;import name&gt; &lt;record&gt;)
...
(&lt;export-type name&gt; (&lt;field label&gt; &lt;expression&gt;) ...))
&lt;import name&gt; -&gt; &lt;type name&gt;
-&gt; &lt;scheme name&gt;
</pre>
Here each expression <code>&lt;record&gt;</code> must evaluate to a record value of type
<code>&lt;type name&gt;</code> or conforming to type scheme <code>&lt;scheme name&gt;</code>. The expression
evaluates to a new record value of type <code>&lt;export-type name&gt;</code>
whose fields are
populated as follows: For each field label belonging to <code>&lt;import name&gt;</code>
that is also a field label of the type
&lt;export-type name&gt;, the corresponding field of <code>&lt;record&gt;</code>
is copied into the result. This is done for all imports from left to
right, dropping any repeated fields. The additional fields <code>&lt;field label&gt;</code>
are then populated with the value of the
corresponding <code>&lt;expression&gt;</code>, overwriting
any fields with the same labels already imported. Any remaining fields are undefined.
All the
<code>&lt;field label&gt;</code>s have to belong to the record type <code>&lt;export type name&gt;</code>.
If this condition is not satisfied, an expansion time error must be signaled.
<p>
The order of evaluation of the expressions <code>&lt;record&gt; ...</code> and
<code>&lt;expression&gt; ...</code> is undefined. All the
expressions <code>&lt;record&gt; ...</code> will be evaluated, even
if their values might not be used in
the result.
</p><h4>Rationale</h4>
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.<p>
Monomorphic record update is a special case of <code>record-compose</code>. The latter
may be used to express more general updates polymorphic in the
argument but monomorphic in the result type.
</p><p>
</p><h4>Examples</h4>
Use <code>record-compose</code> for updates polymorphic in the argument but
monomorphic in the result type:
<pre> (define cp (make-color-point 1 2 'green))
(record-compose (&lt;point cp) (point (x 8))) ==&gt; (point (x 8) (y 2))
</pre>
A more general composition example:
<pre> (define cp (make-color-point 1 2 'green))
(define c (make-color 'blue))
(record-compose (&lt;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)))
==&gt; (color-point (info hi) (hue blue) (x 8) (y 2))
</pre>
Small module-functor example:
<pre>
(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) ==&gt; 3
</pre>
<H1>Implementation</H1>
<p>
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 contains a trivial use of
case-lambda from SRFI-16.
</p><p>
The reference implementation, though relatively portable as a set of
<code>syntax-rules</code> macros, is slow. For practical
implementations, it is recommended that a procedural macro system be
used. Such implementations are provided separately in the discussion
archives. Unless otherwise stated by the author(s), they are covered
by the same copyright agreement as this document.
</p><p>
This version depends on <code>define</code> being treated as a binding
form by <code>syntax-rules</code>. 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. It also assumes
that the implementation of SRFI-9 binds the type name passed to it, which is a
hygienically introduced internal identifier,
using <code>define</code>.
</p><p>
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.
</p><p>
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.
</p><p>
The last section contains a few examples and (non-exhaustive) tests.
</p>
<h2>Reference implementation </h2>
<pre>;============================================================================================
; 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 ('&lt;undefined&gt;))))))
((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)))))
</pre>
<h2>Tests and examples</h2>
<pre>;============================================================================================
; 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) ;==&gt; #t
(point.y p) ;==&gt; 2
(point.y-set! p 7)
(point.y p) ;==&gt; 7
; Simple record schemes.
; Record schemes don't have constructors.
; The predicates and accessors are polymorphic.
(define-record-scheme &lt;point #f &lt;point?
(x &lt;point.x)
(y &lt;point.y))
(define-record-scheme &lt;color #f &lt;color?
(hue &lt;color.hue))
; Concrete instances of the above schemes.
; Constructors may be declared.
; Predicates and accessors, when provided, are monomorphic.
(define-record-type (point &lt;point) make-point point?
(x point.x)
(y point.y))
(define-record-type (color &lt;color) make-color)
(define-record-type (color-point &lt;color &lt;point) (make-color-point x y hue) color-point?
(extra color-point.extra))
(define cp (make-color-point 1 2 'blue))
(&lt;point? cp) ;==&gt; #t
(&lt;color? cp) ;==&gt; #t
(color-point? cp) ;==&gt; #t
;(point.x cp) ;==&gt; error
(&lt;point.y cp) ;==&gt; 2
(&lt;color.hue cp) ;==&gt; blue
(color-point.extra cp) ;==&gt; &lt;undefined&gt;
; 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))) ;==&gt; (point (x 7) (y 2))
(show p) ;==&gt; (point (x 1) (y 2)) - original unaffected
; Polymorphic functional update:
(show
(record-update cp &lt;point (x 7))) ;==&gt; (color-point (extra &lt;undefined&gt;) (hue blue) (x 7) (y 2))
(show cp) ;==&gt; (color-point (extra &lt;undefined&gt;) (hue blue) (x 1) (y 2))
; In-place update:
(show
(record-update! cp &lt;point (x 7))) ;==&gt; color-point (extra &lt;undefined&gt;) (hue blue) (x 7) (y 2))
(show cp) ;==&gt; color-point (extra &lt;undefined&gt;) (hue blue) (x 7) (y 2))
; Use record-compose for updates polymorphic in argument but monomorphic in result type:
(show
(record-compose (&lt;point cp) (point (x 8)))) ;==&gt; (point (x 8) (y 2))
(show cp) ;==&gt; (color-point (extra &lt;undefined&gt;) (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 (&lt;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))))
;==&gt; (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) ;==&gt; 3
; Example of tree data type
(define-record-scheme &lt;tree #f &lt;tree?)
(define-record-type (node &lt;tree) make-node node?
(lhs node.lhs)
(rhs node.rhs))
(define-record-type (leaf &lt;tree) make-leaf leaf?
(val leaf.val))
(define (tree-&gt;list t)
(cond
((leaf? t) (leaf.val t))
((node? t) (cons (tree-&gt;list (node.lhs t))
(tree-&gt;list (node.rhs t))))))
(define t
(make-node (make-node (make-leaf 1)
(make-leaf 2))
(make-leaf 3)))
(&lt;tree? t) ;==&gt; #t
(tree-&gt;list t) ;==&gt; ((1 . 2) . 3)
</pre>
<h2>References</h2>
<pre>[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
</pre>
<H1>Copyright</H1>
<p>Copyright (C) Andr&eacute; van Tonder (2004). All Rights Reserved.</p>
<p>
Permission is hereby granted, free of charge, to any person obtaining
a copy of this software and associated documentation files (the
"Software"), to deal in the Software without restriction, including
without limitation the rights to use, copy, modify, merge, publish,
distribute, sublicense, and/or sell copies of the Software, and to
permit persons to whom the Software is furnished to do so, subject to
the following conditions:
</p>
<p>
The above copyright notice and this permission notice shall be
included in all copies or substantial portions of the Software.
</p>
<p>
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
</p>
<HR>
<ADDRESS>Author: <A href="mailto:andre@het.brown.edu">Andr&eacute; van
Tonder</A></ADDRESS>
<ADDRESS>Editor: <A href="mailto:srfi-editors@srfi.schemers.org">David Van Horn</A></ADDRESS>
<!-- Created: Tue Mar 16 19:01:34 EST 2004 --><!-- hhmts start -->
Last modified: Wed Dec 8 15:31:34 EST 2004
<!-- hhmts end --></BODY></HTML>