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 ModuleSystems in WG2's repo for R7RS-large.


2010-03-05 01:12:57
3fixed typoshistory

As the charter states, "the purpose of this work is to facilitate sharing of Scheme code." To this effect, a module system is explicitly given as a requirement in the charter. We need to consider all the different types of module systems supported by R5RS and R6RS implementations and come up with a proposal that can gain widespread use.

Module systems serve the purpose of encapsulating code and managing namespaces. Although LAMBDA lets us manage local identifiers, it does't allow us to encapsulate macros, and it wants for a friendlier interface for module-like uses. Some of the existing general strategies for implementations include:

Many implementations provide a static module syntax that cannot be extended. They just provide the basics needed for importing and exporting (optionally renamed) identifiers from other modules. The module form itself, and any auxiliary forms such as `import' and `export,' may or may not be expandable as syntax.

A natural approach to creating modules is to build them on top of, and allow them to be composed with, macros. It is relatively easy to implement this for variables only (not importing or exporting syntax), and one sample implementation can be found in lexmod. Allowing importing and exporting syntax requires non-portable extensions. Chez Scheme uses this approach.

An alternative approach to extending a module system is to use a DSL for the module syntax that can extend itself. Scheme48 and Chibi take this approach, supplying a DSL that is essentially Scheme itself. This allows equivalent extensibility to SyntacticModules while keeping the module extensions clearly separate from the language they describe, and making it easier to statically analyze modules.

First-class environments that can be passed as a second argument to `eval' (or the equivalent) can be used to implement modules directly. If macros are not first-class, the environments would need some interface for passing back and forth macro bindings. The disadvantage of this approach is that many implementations do not support first-class environments.

Whatever implementation strategy is used, a syntax for the standard module form must be chosen. Modulo keyword names, options include:

(module <declarations ...) <body> ...

Seen in most other programming languages, the `module' form is just a single declaration, and the rest of the file contains the actual code. This has the distinct disadvantage that it can't be implemented easily in many strategies, or on top of existing module systems.

(module <declarations> ... <body> ...)

Used in the R6RS proposal and many Scheme implementations, this is simple but opens questions of whether declarations may be expanded from macros, and if so makes any static analysis of the module impossible without expanding the body. Sometimes disliked because it requires indenting the body of the module.

(module (<declarations> ...) <body> ...)

This avoids the issues above, simply by delimiting the declarations with a pair of parentheses so they are all known in advance. If they are static (can't be expanded from macros), then a simple rule of allowing and ignoring any unknown declaration keyword allows for easy forward-compatibility and implementation-specific extensions.

(module <declarations> ...)

The module form only allows declarations - any code needs to be specified with declarations such as (include <file>) or (body <code> ...). This is the syntax used in Scheme48 and Chibi-Scheme. It is equivalent in expressiveness to the delimited wrapper approach, trading an additional level of indentation for no extra parens around the declarations.

(module <declarations> ... ---- <body> ...)

where `----' is some arbitrary symbol chosen to act as a delimiter between the declarations and the body of the library. Otherwise the same as above, the shortcut syntax allows the body to have only one level of indentation instead of two.

One frequent debate with respect to syntax is whether to keep the module declaration and source in one file or to split them across separate files. However, whichever module system is used this largely boils down to user preference. The syntaxes which include an implicit body suggest a single file, but so long as an (include <file>) or similar form is provided, any such system can move the body to a separate file. From the other side, syntaxes with no <body> suggest an `include' is required, but all such systems provide a way to inline the body in the module declaration (begin' in Scheme48 and body' in Chibi). So this preference shouldn't affect your choice of syntax.

Existing Proposals: