Actions

Namespaces

From Gambit wiki

Revision as of 03:45, 12 March 2008 by Dpeschel (talk | contribs) (Corrected more grammar. Reformatted one word.)

Namespaces

Introduction

Gambit implements a namespace mechanism, which is used to separate internals from user code and also to partition the Gambit internals into several parts (e.g. the compiler parts use a different namespace than the interpreter parts).

The namespace mechanism is not described in the official documentation, but it is revealed by the examples bundled with the Gambit source code. Marc maybe didn't officially document it since he didn't want to sanction it as official mechanism that he would need to support into the indefinite future, and he maybe originally just wrote it for his own purposes (those are just speculations). In the meantime, as there is now an "official" module system in the form of SNOW!, users should make use of the latter instead of using the namespace mechanism directly (the writer of these lines also intends to continue to work on his "chjmodule" module system to further explore dynamic programming and other concepts like parametrization).

How does it work?

There is a special rule for fully qualifying an identifier, and there is a special form which can be used to (a) declare the currently active default namespace, and (b) to declare a list of given identifiers to be in a particular namespace.

Full qualification

Gambit treats the # character (#\# in Scheme syntax) in identifier symbols as a special character separating the namespace from the local name. Example:

foo#bar

is the bar identifier in the "foo#" namespace (I include the # separator character here since that's also how the namespace form works and because that allows to specify the empty namespace).

There are two special namespaces. One is the "#" namespace being used for most basic internals, and the other is the "" (empty) namespace. The empty namespace is the default namespace for user code when the Gambit interpreter or compiler is started.

Declaring an identifier as belonging to a particular namespace

The

(namespace (STRING SYMBOL...))

special form specifies that the list of the given identifiers (symbols) are to be treated as belonging to the namespace given in STRING. Example:

(namespace ("foo#" bar))

says that whenever bar is mentioned in the code that follows in the same scope or file, it is to be implicitly treated as foo#bar.

Note that the identifier namespace is part of the empty namespace. If you change the default namespace (see below), you have to use ##namespace instead.

Switching the default namespace in code

Identifiers which do not contain a # character are treated as belonging to the active namespace in the current scope (see below for what the scope is). This is the empty namespace by default. This can be changed by putting

(namespace (STRING))

before the rest of the code; e.g. using a namespace declaration without a list of identifiers. Note that once you've changed the default namespace, you cannot refer to namespace anymore, since this identifier is only defined in the empty namespace and there is no way to fully qualify identifiers as belonging to the empty namespace, so you have to use the ##namespace identifier instead (meaning the full qualification for the "#" internals namespace) which acts the same.

Scope of namespace declarations

A namespace declaration is in effect until the end of the current lambda or let construct (note that begin doesn't open a new scope), or, if specified at the top level of the file (or just inside a begin form residing in the top level) until the end of that file.

Switching the default namespace in the repl

If entered in the repl, or run by the eval procedure, the default namespace used by the repl (which is also the empty namespace by default) is changed. This then remains active until the next namespace declaration is entered or evaled.

Usage

Modularization

The namespace mechanism can be used to partition user code in a style similar to how C works, by writing a "header" file for a library which declares all the exported values to be in the namespace that has been chosen for that library. In the Gambit source code and the examples, Marc uses the convention of adding the # character to the name of the module (e.g. if he writes a module "foo.scm" he will name the header file "foo#.scm").

Access to Gambit internals

Since the namespace handling doesn't prohibit access (it's just for the sake of convenience, not for protection), Gambit is somewhat special in that all of its internal procedures can be accessed by the user; this is very useful to learn how the internals work and also to extend the system without hacking the Gambit source code. It must be noted though, that most of those procedures do not check their types, so you will quickly get a segfault when feeding them data of the wrong type. Sometimes this missing type check is also (mis)usable (or is even used explicitly by other Gambit internal procedures) to access data structures on a lower level:

(define-structure foo a b c)
(let ((v (make-foo 123 "hello" 'world)))
  ;; (mis)treat the structure as vector:
  (let ((len (##vector-length v)))
    (let lp ((i 0))
      (if (< i len)
        (begin
          (println "value at position " i ":" (##vector-ref v i))
          (lp (+ i 1)))))))

This prints

value at position 0:#<type #2 foo>
value at position 1:123
value at position 2:hello
value at position 3:world

(Note that if you access elements behind the given length, you would access unrelated memory contents or invalid memory addresses.)

Producing low-level unsafe code explicitly

It's also worthwhile knowing that the basic primitives (like memory accesses, low-level arithmetic, etc.) which are mapped by the compiler directly to C constructs (actually C macros as defined in the include/gambit.h file) are a bunch of such identifiers in the # namespace. Like (##car obj) will be translated to ___CAR(obj) which is then expanded by the C compiler to basically a pointer dereference. So those can be used in user code to let the compiler output those low-level constructs, which do not do any type checking (and are simple enough for the C compiler to enable optimizations on the C level) and thus very fast, without declaring your whole Scheme code as being unsafe by using the declare special form (which would achieve the same thing).