14.4 Module Names and Loading🔗ℹ

14.4.1 Resolving Module Names🔗ℹ

The syntax/modresolve library provides additional operations for resolving and manipulating module names.

The name of a declared module is represented by a resolved module path, which encapsulates either a symbol or a complete filesystem path (see Paths). A symbol normally refers to a predefined module or module declared through reflective evaluation (e.g., eval). A filesystem path normally refers to a module declaration that was loaded on demand via require or other forms.

A module path is a datum that matches the grammar for module-path for require. A module path is relative to another module.

A resolved module path is interned. That is, if two resolved module path values encapsulate paths that are equal?, then the resolved module path values are eq?.

The default module name resolver uses collection-file-path to convert lib and symbolic-shorthand module paths to filesystem paths. The collection-file-path function, in turn, uses the current-library-collection-links and current-library-collection-paths parameters.

For the second case, the standard module name resolver keeps a table per module registry containing loaded module name. If a resolved module path is not in the table, and #f is not provided as the fourth argument to the module name resolver, then the name is put into the table and the corresponding file is loaded with a variant of load/use-compiled that passes the expected module name to the compiled-load handler.

While loading a file, the default module name resolver sets the current-module-declare-name parameter to the resolved module name (while the compiled-load handler sets current-module-declare-source). Also, the default module name resolver records in a private continuation mark the module being loaded, and it checks whether such a mark already exists; if such a continuation mark does exist in the current continuation, then the exn:fail exception is raised with a message about a dependency cycle.

The default module name resolver cooperates with the default compiled-load handler: on a module-attach notification, bytecode-file information recorded by the compiled-load handler for the source namespace’s module registry is transferred to the target namespace’s module registry.

The default module name resolver also maintains a small, module registry-specific cache that maps lib and symbolic module paths to their resolutions. This cache is consulted before checking parameters such as current-library-collection-links and current-library-collection-paths, so results may “stick” even if those parameter values change. An entry is added to the cache only when the fourth argument to the module name resolver is true (indicating that a module should be loaded) and only when loading succeeds.

Finally, the default module name resolver potentially treats a submod path specially. If the module path as the first element of the submod form refers to non-existent collection, then instead of raising an exception, the default module name resolver synthesizes an uninterned symbol module name for the resulting resolved module path. This special treatment of submodule paths is consistent with the special treatment of nonexistent submodules by the compiled-load handler, so that module-declared? can be used more readily to check for the existence of a submodule.

Module loading is suppressed (i.e., #f is supplied as a fourth argument to the module name resolver) when resolving module paths in syntax objects (see Syntax Objects). When a syntax object is manipulated, the current namespace might not match the original namespace for the syntax object, and the module should not necessarily be loaded in the current namespace.

For historical reasons, the default module name resolver currently accepts three arguments, in addition to two and four. Three arguments are treated the same as four arguments with the fourth argument as #t, except that an error is also logged. Support for three arguments will be removed in a future version.

The current-module-name-resolver binding is provided as protected in the sense of protect-out.

Changed in version 6.0.1.12 of package base: Added error logging to the default module name resolver when called with three arguments.
Changed in version 7.0.0.17: Added special treatment of submod forms with a nonexistent collection by the default module name resolver.
Changed in version 8.2.0.4: Changed binding to protected.

When declaring submodules, current-module-declare-name determines the name used for the submodule’s root module, while its submodule path relative to the root module is unaffected.

14.4.2 Compiled Modules and References🔗ℹ

While expanding a module declaration, the expander resolves module paths for imports to load module declarations as necessary and to determine imported bindings, but the compiled form of a module declaration preserves the original module path. Consequently, a compiled module can be moved to another filesystem, where the module name resolver can resolve inter-module references among compiled code.

When a module reference is extracted from compiled form (see module-compiled-imports) or from syntax objects in macro expansion (see Syntax Object Content), the module reference is reported in the form of a module path index. A module path index is a semi-interned (multiple references to the same relative module tend to use the same module path index value, but not always) opaque value that encodes a module path (see module-path?) and either a resolved module path or another module path index to which it is relative.

A module path index that uses both #f for its path and base module path index represents “self”—i.e., the module declaration that was the source of the module path index—and such a module path index can be used as the root for a chain of module path indexes at compile time. For example, when extracting information about an identifier’s binding within a module, if the identifier is bound by a definition within the same module, the identifier’s source module is reported using the “self” module path index. If the identifier is instead defined in a module that is imported via a module path (as opposed to a literal module name), then the identifier’s source module will be reported using a module path index that contains the required module path and the “self” module path index. A “self” module path index has a submodule path when the module that it refers to is a submodule.

A module path index has state. When it is resolved to a resolved module path, then the resolved module path is stored with the module path index. In particular, when a module is loaded, its root module path index is resolved to match the module’s declaration-time name. This resolved path is forgotten, however, in identifiers that the module contributes to the compiled and marshaled form of other modules. The transient nature of resolved names allows the module code to be loaded with a different resolved name than the name when it was compiled.

Two module path index values are equal? when they have equal? path and base values (even if they have different resolved values).

Resolving a module path index uses the current module name resolver (see current-module-name-resolver). Depending on the kind of module paths encapsulated by mpi, the computed resolved name can depend on the value of current-load-relative-directory or current-directory. The load? argument is propagated as the last argument to the module name resolver, while the src-stx argument is propagated as the next-to-last argument.

Beware that concurrent resolution in namespaces that share a module registry can create race conditions when loading modules. See also namespace-call-with-registry-lock.

If mpi represents a “self” (see above) module path that was not created by the expander as already resolved, then module-path-index-resolve raises exn:fail:contract without calling the module name resolver.

See also resolve-module-path-index.

Changed in version 6.90.0.16 of package base: Added the load? optional argument.
Changed in version 8.2: Added the src-stx optional argument.

A #f second result means that the path is relative to an unspecified directory (i.e., its resolution depends on the value of current-load-relative-directory and/or current-directory).

A #f for the first result implies a #f for the second result, and means that mpi represents “self” (see above). Such a module path index may have a non-#f submodule path as reported by module-path-index-submodule.

The name is a symbol for a top-level module, or a symbol paired with a list of symbols where the list reflects the submodule path to the module starting with the top-level module’s declared name.

Each associated list, which is represented by list? in the result contracts above, more precisely matches the contract

(listof (list/c symbol?

(listof

(or/c module-path-index?

(list/c module-path-index?

phase+space?

symbol?

phase+space?)))

; only if verbosity is 'defined-names:

symbol?))

For each element of the list, the leading symbol is the name of the export.

The second part—the list of module path index values, etc.—describes the origin of the exported identifier. If the origin list is null, then the exported identifier is defined in the module. If the exported identifier is re-exported, instead, then the origin list provides information on the import that was re-exported. The origin list has more than one element if the binding was imported multiple times from (possibly) different sources.

The last part, a symbol, is included only if verbosity is 'defined-names. In that case, the included symbol is the name of the definition within its defining module (which may be different than the name that is exported).

For each origin, a module path index by itself means that the binding was imported with a phase level shift of 0 (i.e., a plain require without for-meta, for-syntax, etc.) into the default binding space (i.e., without for-space), and the imported identifier has the same name as the re-exported name. An origin represented with a list indicates explicitly the import, the phase level plus binding space where the imported identifier is bound (see phase+space? for more information on the representation), the symbolic name of the import as bound in the importing module, and the phase level plus binding space of the identifier from the exporting module.

Changed in version 7.5.0.6 of package base: Added the verbosity argument.
Changed in version 8.2.0.3: Generalized results to phase–space combinations.

Added in version 6.5.0.5 of package base.

Returns information intended to reflect the “language” of the module’s implementation as originally attached to the syntax of the module’s declaration though the 'module-language syntax property. See also module.

If no information is available for the module, the result is #f. Otherwise, the result is (vector mp name val) such that ((dynamic-require mp name) val) should return function that takes two arguments. The function’s arguments are a key for reflected information and a default value. Acceptable keys and the interpretation of results is up to external tools, such as DrRacket. If no information is available for a given key, the result should be the given default value.

See also module->language-info and racket/language-info.

Added in version 8.4.0.2 of package base.

14.4.3 Dynamic Module Access🔗ℹ

Dynamically instantiates the module specified by mod in the current namespace’s registry at the namespace’s base phase, if it is not yet instantiated. The current module name resolver may load a module declaration to resolve mod (see current-module-name-resolver); the path is resolved relative to current-load-relative-directory and/or current-directory. Beware that concurrent dynamic-requires in namespaces that share a module registry can create race conditions; see also namespace-call-with-registry-lock.

If provided is #f, then the result is #<void>, and the module is not visited (see Module Expansion, Phases, and Visits) or even made available (for on-demand visits) in phases above the base phase.

The double quoted ''a evaluates to the root-module-path 'a (see the grammar for require). Using 'a for mod won’t work, because that evaluates to root-module-path a, and the example is not a module installed in a collection. Using a won’t work, because a is an undefined variable.

Declaring (module a ....) within another module, instead of in the read-eval-print loop, would create a submodule. In that case, (dynamic-require ''a #f) would not access the module, because ''a does not refer to a submodule.

When provided is a symbol, the value of the module’s export with the given name is returned, and still the module is not visited or made available in higher phases.

If the module exports provided as syntax, then a use of the binding is expanded and evaluated in a fresh namespace to which the module is attached, which means that the module is visited in the fresh namespace. The expanded syntax must return a single value.

If the module has no such exported variable or syntax, then fail-thunk is called; the default fail-thunk raises exn:fail:contract. If the variable named by provided is exported protected (see Code Inspectors), then the exn:fail:contract exception is raised.

If provided is 0, then the module is instantiated but not visited, the same as when provided is #f. With 0, however, the module is made available in higher phases.

If provided is #<void>, then the module is visited but not instantiated (see Module Expansion, Phases, and Visits), and the result is #<void>.

More examples using different module-path grammar expressions are given below:

The last line in the above example could instead have been written as

which is equivalent.

If load? is #t and mod is not a resolved module path, the module is loaded in the process of resolving mod (as for dynamic-require and other functions). Checking for the declaration of a submodule does not trigger an exception if the submodule cannot be loaded because it does not exist, either within a root module that does exist or because the root module does not exist.

A module can be declared by using dynamic-require.

Changed in version 7.5.0.6 of package base: Added the verbosity argument.
Changed in version 8.2.0.3: Generalized results to phase–space combinations.

Added in version 6.5.0.5 of package base.

Added in version 8.4.0.2 of package base.

14.4.4 Module Cache🔗ℹ

The expander keeps a place-local module cache in order to save time while loading modules that have been previously declared.

Added in version 8.4.0.5 of package base.