WG1 Module Proposal
Steven Ganz [steven_ganz] <email@example.com>
A module syntax is provided that is general enough to allow imports to be specified when a module is included into a context. Modules have separate namespaces, and can import and export variables and macros with fine-grained control. Restrictions (not fully specified here) can be imposed on the use of these forms in macros to allow for varying levels of static analysis.
Modules should be true abstractions over syntax. The charter indicates that a module system for WG1 should be defined to "promote extensibility." Extensibility is enhanced by the ability to reuse modules while varying their imports.
This document proposes the following forms that can occur in any context where a definition is allowed (modulo any restrictions on their use in macro definitions) and that evaluate to an unspecified value:
The following syntax is provided:<module> := (module [<module-name>] (import <import-spec> ...) (export <export-spec> ...) <expr> <expr> ...) <import-spec> := (module <include-sets>) | <identifier> <export-spec> := <identifier> | ([<module-name>] <identifier> [(implicit-export <identifier> ...)]) <include> := (include-module <include-sets> ...) <include-sets> := <module-name> | (only <include-sets> <export-identifier> ...) | (except <include-sets> <export-identifier> ...) | (prefix <include-sets> <export-identifier> ...) | (rename <include-sets> (<old-identifier> <new-identifier>) ...)
A module has an optional name that, if specified, is added to the following namespace and that can be referenced from within an include-spec. A module specifies identifiers and modules to import and identifiers to export. It has at least one body expression. The exported identifiers of a module must be defined in the module body. The imported identifiers (both explicit and via an <include-sets>) may appear free in the module body.
Modules may be included in three ways:
In either case, the import identifiers (restricted and renamed as described below) must be bound in the current scope. This implies that for a module definition at top level, imported modules should have no such import identifiers. The export identifiers are then considered bound in the following scope to their definitions in the module body.
Upon inclusion of a module, its sets of imported and exported identifiers can be modified as follows:
It is a syntax error if the same identifier is imported twice, from any combination of modules or multiple import forms.
Macros may be restricted to preclude expanding into module and/or include-module forms.
Finally, there are two forms for exporting identifiers from a module. The simple form just specifies the identifier.
The more complex form associates an identifier with a set of implicit exports.
The more complex form allows a specification of implicit exports. If a macro is bound to an identifier with implicit-exports and that identifier is exported, then the macro may expand into identifiers that are implicitly exported without it being an unbound identifier error. This allows bindings to be implicitly visible to the outside world without requiring the module or library form to explicitly export them. Rather, a macro may expand into them, instead. By default, if no implicit-export form indicates the set of implicit identifiers associated with a given syntax, and that syntax is exported, then all of the definitions and bindings visible to the macro are implicitly exported (this corresponds to the R6RS behavior). However, if an implicit-export form does exist, the set of identifiers implicitly exported for that syntax is completely and fully specified by the implicit-export form.
In that case, the export applies to the nearest module form where they are encountered. That is, if one module form encapsulates another that in turn encapsulates a third that export form, that export form applies to the module form, rather than the library form which wraps or contains the module form.
Notice, it is only necessary to define mon-let ONCE and we can retrieve two versions.
The following are not considered in this proposal, but could potentially be considered in extensions/refinements of this proposal:
Also, the precise restrictions on macros, if any, are not specified here.
The forms provided here are not unlike those of Chez's module system, but the flexibility they provide is more like units from PLT Scheme. Much of the notation for modifying identifier sets is supported by R6RS.