This example will not work within a module or in DrRacket’s definitions window, but it will work in the interactions window, for reasons that are explained by the end of Namespaces.
The eval function takes a representation of an expression or definition (as a “quoted” form or syntax object) and evaluates it:
> (eval '(+ 1 2))
The power of eval is that an expression can be constructed dynamically:
> (define (eval-formula formula) (eval `(let ([x 2] [y 3]) ,formula))) > (eval-formula '(+ x y))
> (eval-formula '(+ (* x y) y))
Of course, if we just wanted to evaluate expressions with given values for x and y, we do not need eval. A more direct approach is to use first-class functions:
> (define (apply-formula formula-proc) (formula-proc 2 3)) > (apply-formula (lambda (x y) (+ x y)))
> (apply-formula (lambda (x y) (+ (* x y) y)))
However, if expressions like (+ x y) and (+ (* x y) y) are read from a file supplied by a user, for example, then eval might be appropriate. Similarly, the REPL reads expressions that are typed by a user and uses eval to evaluate them.
Also, eval is often used directly or indirectly on whole modules. For example, a program might load a module on demand using dynamic-require, which is essentially a wrapper around eval to dynamically load the module code.
15.1.1 Local Scopes
The eval function cannot see local bindings in the context where it is called. For example, calling eval inside an unquoted let form to evaluate a formula does not make values visible for x and y:
> (define (broken-eval-formula formula) (let ([x 2] [y 3]) (eval formula))) > (broken-eval-formula '(+ x y))
cannot reference an identifier before its definition
in module: top-level
The eval function cannot see the x and y bindings precisely because it is a function, and Racket is a lexically scoped language. Imagine if eval were implemented as
(define (eval x) (eval-expanded (macro-expand x)))
then at the point when eval-expanded is called, the most recent binding of x is to the expression to evaluate, not the let binding in broken-eval-formula. Lexical scope prevents such confusing and fragile behavior, and consequently prevents eval from seeing local bindings in the context where it is called.
You might imagine that even though eval cannot see the local bindings in broken-eval-formula, there must actually be a data structure mapping x to 2 and y to 3, and you would like a way to get that data structure. In fact, no such data structure exists; the compiler is free to replace every use of x with 2 at compile time, so that the local binding of x does not exist in any concrete sense at run-time. Even when variables cannot be eliminated by constant-folding, normally the names of the variables can be eliminated, and the data structures that hold local values do not resemble a mapping from names to values.
Since eval cannot see the bindings from the context where it is called, another mechanism is needed to determine dynamically available bindings. A namespace is a first-class value that encapsulates the bindings available for dynamic evaluation.
Informally, the term namespace is sometimes used interchangeably with environment or scope. In Racket, the term namespace has the more specific, dynamic meaning given above, and it should not be confused with static lexical concepts.
Some functions, such as eval, accept an optional namespace argument. More often, the namespace used by a dynamic operation is the current namespace as determined by the current-namespace parameter.
When eval is used in a REPL, the current namespace is the one that the REPL uses for evaluating expressions. That’s why the following interaction successfully accesses x via eval:
> (define x 3) > (eval 'x)
In contrast, try the following simple module and running it directly in DrRacket or supplying the file as a command-line argument to racket:
#lang racket (eval '(cons 1 2))
This fails because the initial current namespace is empty. When you run racket in interactive mode (see Interactive Mode), the initial namespace is initialized with the exports of the racket module, but when you run a module directly, the initial namespace starts empty.
In general, it’s a bad idea to use eval with whatever namespace happens to be installed. Instead, create a namespace explicitly and install it for the call to eval:
#lang racket (define ns (make-base-namespace)) (eval '(cons 1 2) ns) ; works
The make-base-namespace function creates a namespace that is initialized with the exports of racket/base. The later section Manipulating Namespaces provides more information on creating and configuring namespaces.
15.1.3 Namespaces and Modules
As with let bindings, lexical scope means that eval cannot automatically see the definitions of a module in which it is called. Unlike let bindings, however, Racket provides a way to reflect a module into a namespace.
The module->namespace function takes a quoted module path and produces a namespace for evaluating expressions and definitions as if they appeared in the module body:
> (module m racket/base (define x 11)) > (require 'm) > (define ns (module->namespace ''m)) > (eval 'x ns)
The double quoting in ''m is because 'm is a module path that refers to an interactively declared module, and so ''m is the quoted form of the path.
The module->namespace function is mostly useful from outside a module, where the module’s full name is known. Inside a module form, however, the full name of a module may not be known, because it may depend on where the module source is located when it is eventually loaded.
From within a module, use define-namespace-anchor to declare a reflection hook on the module, and use namespace-anchor->namespace to reel in the module’s namespace:
#lang racket (define-namespace-anchor a) (define ns (namespace-anchor->namespace a)) (define x 1) (define y 2) (eval '(cons x y) ns) ; produces (1 . 2)