15.3 Scripting Evaluation and Using load

Historically, Lisp implementations did not offer module systems. Instead, large programs were built by essentially scripting the REPL to evaluate program fragments in a particular order. While REPL scripting turns out to be a bad way to structure programs and libraries, it is still sometimes a useful capability.

Describing a program via load interacts especially badly with macro-defined language extensions [Flatt02].

The load function runs a REPL script by reading S-expressions from a file, one by one, and passing them to eval. If a file "place.rkts" contains

(define city "Salt Lake City")
(define state "Utah")
(printf "~a, ~a\n" city state)

then it can be loaded in a REPL:

> (load "place.rkts")

Salt Lake City, Utah

> city

"Salt Lake City"

Since load uses eval, however, a module like the following generally will not work—for the same reasons described in Namespaces:

#lang racket
(define there "Utopia")
(load "here.rkts")

The current namespace for evaluating the content of "here.rkts" is likely to be empty; in any case, you cannot get there from "here.rkts". Also, any definitions in "here.rkts" will not become visible for use within the module; after all, the load happens dynamically, while references to identifiers within the module are resolved lexically, and therefore statically.

Unlike eval, load does not accept a namespace argument. To supply a namespace to load, set the current-namespace parameter. The following example evaluates the expressions in "here.rkts" using the bindings of the racket/base module:

#lang racket
(parameterize ([current-namespace (make-base-namespace)])
  (load "here.rkts"))

You can even use namespace-anchor->namespace to make the bindings of the enclosing module accessible for dynamic evaluation. In the following example, when "here.rkts" is loaded, it can refer to there as well as the bindings of racket:

#lang racket
(define there "Utopia")
(define-namespace-anchor a)
(parameterize ([current-namespace (namespace-anchor->namespace a)])
  (load "here.rkts"))

Still, if "here.rkts" defines any identifiers, the definitions cannot be directly (i.e., statically) referenced by in the enclosing module.

The racket/load module language is different from racket or racket/base. A module using racket/load treats all of its content as dynamic, passing each form in the module body to eval (using a namespace that is initialized with racket). As a result, uses of eval and load in the module body see the same dynamic namespace as immediate body forms. For example, if "here.rkts" contains

(define here "Morporkia")
(define (go!) (set! here there))

then running

#lang racket/load
(define there "Utopia")
(load "here.rkts")
(printf "~a\n" here)

prints “Utopia”.

Drawbacks of using racket/load include reduced error checking, tool support, and performance. For example, with the program

#lang racket/load
(define good 5)
(printf "running\n")

DrRacket’s Check Syntax tool cannot tell that the second good is a reference to the first, and the unbound reference to bad is reported only at run time instead of rejected syntactically.