This site is a static rendering of the Trac instance that was used by R7RS-WG1 for its work on R7RS-small (PDF), which was ratified in 2013. For more information, see Home. For a version of this page that may be more recent, see ImmutableParametersCowan in WG2's repo for R7RS-large.

Immutable­Parameters­Cowan

cowan
2010-07-21 20:43:55
2Incorporates much of SRFI 39 and ParametersSnellPym to make it stand-alonehistory
source

Parameters

This proposal defines parameter objects, the procedure make-parameter to create parameter objects and the parameterize special form to dynamically bind parameter objects. In the dynamic environment, each parameter object is bound to the value of the parameter. When a procedure is called the called procedure inherits the dynamic environment from the caller. The parameterize special form allows the binding of a parameter object to be changed for the dynamic extent of its body.

The proposal is an alternative to ParametersSnellPym, providing immutable parameter objects. It is a subset of SRFI 39, with the difference being that passing a non-zero number of arguments to a parameter object has undefined effect rather than assigning a new value to the parameter object. The effect of SRFI 39 mutable parameters can be emulated by binding a mutable object such as a pair to the parameter.

Rationale

The dynamic environment is the structure which allows the system to find the value returned by the R5RS procedures current-input-port and current-output-port. The R5RS procedures with-input-from-file and with-output-to-file extend the dynamic environment to produce a new dynamic environment which is in effect for the dynamic extent of the call to the thunk passed as their last argument. These procedures are essentially special purpose dynamic binding operations on hidden dynamic variables (one for current-input-port and one for current-output-port). The purpose of this proposal is to generalize this dynamic binding mechanism (which exists in all R5RS compliant systems) to allow the user to introduce new dynamic variables and dynamically bind them.

General dynamic binding mechanisms exist in several implementations of Scheme under various names, including "fluid" variables and parameter objects. The parameter objects specified in this proposal are compatible with the semantics of most, but not all, implementations of Scheme that currently support SRFI-39.

Background

Parameters are certainly useful:

  1. They can be labor-savers, passing "configuration information" down through complex call stacks without needing explicit parameter passing through functions that do nothing with the parameter other than pass it to all child procedures called, until a lower layer actually uses it.
  1. They allow isolation of concerns; as per the previous point, the intermediate procedures do not need to know what configuration information is sent to which child procedures, which is useful when the child procedures are provided by external libraries, or are arbitrary closures which might invoke arbitrary external libraries.
  1. They allow dynamically scoped state, for tasks like exception handling and thread-local storage.
  1. They generalize a mechanism that must be present within the implementation to support current-input-port and current-output-port, making it portably available to library authors.

Parameters need to be specified as part of WG1 Scheme rather than implemented by portable libraries, as a portable library has no way of knowing if the implementation provides threads; and if it does, it has no portable way of implementing dynamically scoped state in a thread-safe manner.

Specification

The dynamic environment is composed of two parts: the local dynamic environment and the global dynamic environment. The global dynamic environment is used to look up parameter objects that can't be found in the local dynamic environment. When parameter objects are created, their initial value is put into the global dynamic environment. The local dynamic environment is only extended by the parameterize form.

Parameter objects are created with the make-parameter procedure. The global dynamic environment is updated to associate the newly created parameter object to a value, which can be retrieved by invoking the parameter object as a procedure with no arguments. The parameterize special form, when given parameter objects and corresponding values, binds for the dynamic extent of its body the each parameter object to a new value. The parameterize special form behaves analogously to let when binding more than one parameter object (that is, the order of evaluation is unspecified and the new bindings are only visible in the dynamic extent of the parameterize special form). However, while let requires the user to specify identifiers to be bound, the parameter objects bound by parameterize can be specified by arbitrary Scheme expressions.

This poroposal also specifies that current-input-port and current-output-port are implemented as parameter objects created with make-parameter.

Procedures and syntax

(make-parameter init [converter]) ;procedure

Returns a new parameter object which is bound in the global dynamic environment to a cell containing the value returned by the call (converter init). If converter is not specified the identity function is used instead. The conversion procedure can be used to coerce init to a suitable type or to signal an error if its type is inappropriate.

Invoking the returned parameter object with no arguments returns its current value. The effect of invoking a parameter object with arguments is not defined by this proposal.

(define radix (make-parameter 10)) (define write-shared (make-parameter #f (lambda (x) (if (boolean? x) x (error "only booleans are accepted by write-shared"))))) (radix) ==> 10 (radix 2) (radix) ==> 2 (write-shared 0) gives an error (define prompt (make-parameter 123 (lambda (x) (if (string? x) x (with-output-to-string (lambda () (write x))))))) (prompt) ==> "123" (prompt ">") (prompt) ==> ">"

(parameterize ((expr1 expr2) ...) ''body'') ;syntax

The expressions expr1 and expr2 are evaluated in an unspecified order. The value of the expr1 expressions must be parameter objects. For each expr1 expression, the local dynamic environment is extended in an unspecified order with a binding of the parameter object expr1 to the result of the call (converter val), where val is the value of expr2 and converteris the conversion procedure of the parameter object. The resulting dynamic environment is then used for the evaluation of body, which refers to the R5RS grammar nonterminal of that name. The result(s) of the parameterize form are the result(s) of body.

(radix) ==> 2 (parameterize ((radix 16)) (radix)) ==> 16 (radix) ==> 2 (define (f n) (number->string n (radix))) (f 10) ==> "1010" (parameterize ((radix 8)) (f 10)) ==> "12" (parameterize ((radix 8) (prompt (f 10))) (prompt)) ==> "1010"

Interaction with threads

SRFI 39 does not specify how parameter objects interact with threads. ParametersSnellPym provides mutable parameters shared between threads, which threads may mutate to share information. I take the view that using parameter objects to share information between threads is an improper and unnecessary conflation of two distinct facilities, dynamic variables and thread-shared data. Instead, some mechanism for exchanging data between threads, such as mailboxes, should be provided as a part of standardizing threads, probably in WG2.

Therefore, this proposal specifies that when a parent thread creates a new child thread, the parameter objects in the child thread have initial values that are the same (in the sense of eqv?) as their current (not initial) values in the parent thread. Because parameter objects are immutable, it does not matter whether the underlying representation uses copying or sharing.

Most implementations that provides SRFI 39 can trivially provide this facility. The exceptions are implementations like Scheme48 which neither copy nor share parameter objects with child threads, but provide freshly initialized objects instead.

Thanks

Thanks to Taylor Campbell, Alaric Snell-Pym, and Marc Feeley, who as usual are not responsible for what I did to their ideas and/or prose.