Title

SRFI 17: Generalized set!

Author

Per Bothner

Status

This SRFI is currently in ``final'' status. To see an explanation of each status that a SRFI can hold, see here. You can access the discussion via the archive of the mailing list.

Abstract

This is a proposal to allow procedure calls that evaluate to the "value of a location" to be used to set the value of the location, when used as the first operand of set!. For example:
(set! (car x) (car y))
becomes equivalent to
(set-car! x (car y))

Many programming languages have the concept of an lvalue. that is an "expression" that "evaluates" to a location, and which can appear on the left-hand-side of an assignment. Common Lisp has a related concept of "generalized variables" which can be used in setf and some other special forms. However, the Common Lisp concept is based on the idea of compile-time recognition of special "location-producing" functions; this does not seem to be in the "spirit of Scheme".

This SRFI proposes an extension of set! so that it provides similar functionality as Common Lisp's setf, except that the updater is associated with a procedure value, rather than a name.

Rationale

There is ample precedent for general "lvalues" on the left-hand side of an assignment. This includes most statically typed languages, and many dynamically typed languages (including APL and Common Lisp). That suggests this is a natural idiom for people. One reason may be that there are fewer procedure names to learn. Another is that it becomes visually clearer which expression is the new value, and which are parameters. Also, the visual consistency between an expression evaluated for its value and one evaluated to yield a location seems natural to people.

For most languages, the set of lvalue-producing operators is limited (typically array indexing and field selection). Some languages have general lvalues as first class values. For example Algol 68 has expressions that have reference type. However, this is made convenient by using automatic dereferencing coercions, which would not work for a dynamically typed language like Scheme. ML goes further: All mutable variables are first-class "cells", and accessing the contents of a cell requires an explicit operator. This is also not compatible with Scheme. Instead we need to stick to the model where using a variable in most contexts means using its value, but referring to a variable in certain lvalue contexts (lhs of assignment) refers to its actual location. Sticking to this model for general "lvalue expressions" in set! means that "evaluation" must be done differently from normal evaluation when in an "lvalue context". That is what this proposal does.

This is a controversial issue. This srfi does not wish to imply that all Scheme implementations should support this feature; it only requests that implementations that do implement generalized set! should be compatible with this srfi.

Specification

The special form set! is extended so the first operand can be a procedure application, and not just a variable. The procedure is typically one that extracts a component from some data structure. Informally, when the procedure is called in the first operand of set!, it causes the corresponding component to be replaced by the second operand. For example,
(set (vector-ref x i) v)
would be equivalent to:
(vector-set! x i v)

Each procedure that may be used as the first operand to set! must have a corresponding "setter" procedure. The builtin procedure setter takes a procedure and returns the corresponding setter procedure.

We define:

(set! (proc arg ...) value)
as:
((setter proc) arg ... value)

Note: This definition matches the existing Scheme convention for setter procedures, where the new value is given last. For example we can define (setter car) to be set-car!. An alternative definition would be:

((setter proc) value arg ...) ;; Not the actual definition.
This definition would work better when you consider procedures that take a variable number of arguments. This is because it is straight-forward to add one extra initial fixed argument, but if you add an extra fixed argument to the end of an argument list that has a "rest" parameter, then things get more messy. However, consistency with the existing new-value-last convention seems more valuable.

Standard setters

The following standard procedures have pre-defined setters:
(set! (car x) v) == (set-car! x v)
(set! (cdr x) v) == (set-cdr! x v)
(set! (caar x) v) == (set-car! (car x) v)
(set! (cadr x) v) == (set-car! (cdr x) v)
....
(set! (caXXr x) v) == (set-car! (cXXr x) v)
(set! (cdXXr x) v) == (set-cdr! (cXXr x) v)
(set! (string-ref x i) v) == (string-set! x i v)
(set! (vector-ref x i) v) == (vector-set! x i v)

Setting setters; properties

A setter procedure is a special case of the concept of procedures having associated properties. Other properties might include the procedures's name or usage documentation. As a hypothetical example (i.e. not part of this SRFI), we can use the Common Lisp documentation function, where for example:
(documentation sqrt)
returns the "documentation string" (if defined) for sqrt. Such properties should also be settable using generalized set!. For example:
(set! (documentation sqrt) "Calculates the square root.")

This SRFI does not propose a general "procedure properties" feature, but it should be compatible with it. It does specify the special case for the setter property. This is defined such that:

(set! (setter proc) setter)
sets the setter procedure associated with proc to setter. For example, we can assume
(set! (setter car) set-car!)
has been executed by the Scheme prologue.

Efficiency Issues

If (set! (foo ..) ...) is to be the preferred idiom, we want to make ((setter foo) ...) as efficient as (set-foo! ...). This is only possible when the compiler knows both what function the symbol foo is bound to, and the setter associated with that function. Scheme (as opposed to Common Lisp) does not say anything about a compiler or what it can inline, so we cannot say much here. A compiler that does whole-program analysis can safely inline calls using a variable bound unchangably to a known procedure; it can make the same deduction if the procedure's setter is set. If separate compilation is supported, then a compiler cannot safely make such deductions for either plain calls or setter calls, without extra information, such as a module system, compiler switches, or other non-standard declarations. Thus my belief is that this srfi does not inherently make efficient compilation more difficult. However, the next section does define a way to inherently tie a setter to a procedure, which does reduce the problem of inlining generalized set! to the standard inlining problem.

getter-with-setter

The function getter-with-setter can be used to define a procedure with associated properties. Specifically:

(getter-with-setter getter setter)
This evaluates to a new anonymous procedure which when applied invokes getter, and whose setter is setter. It is an error for a program to subsequently try to modify the setter of the resulting compound.

For example, we could define:

(define car (getter-with-setter %primitive-car %primitive-set-car!))
(define set-car! %primitive-set-car!)
The advantage of this approach that whenever a compiler can inline car, it can also inline (setter car).

Implementation

Here is a sample implementation for Twobit.

Here is a sample definition of getter-with-setter:

(define (getter-with-setter get set)
  (let ((proc (lambda args (apply get args))))
    (set! (setter proc) set)
    proc))

Copyright

Copyright (C) Per Bothner (1999, 2000). All Rights Reserved.

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:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

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.


Editor: Mike Sperber
Last modified: Mon Jul 24 12:00:06 MST 2000