5. The import system

Python code in one module gains access to the code in another module by the process of importing it. The import statement is the most common way of invoking the import machinery, but it is not the only way. Functions such as importlib.import_module() and built-in __import__() can also be used to invoke the import machinery.

The import statement combines two operations; it searches for the named module, then it binds the results of that search to a name in the local scope. The search operation of the import statement is defined as a call to the __import__() function, with the appropriate arguments. The return value of __import__() is used to perform the name binding operation of the import statement. See the import statement for the exact details of that name binding operation.

A direct call to __import__() performs only the module search and, if found, the module creation operation. While certain side-effects may occur, such as the importing of parent packages, and the updating of various caches (including sys.modules), only the import statement performs a name binding operation.

When calling __import__() as part of an import statement, the import system first checks the module global namespace for a function by that name. If it is not found, then the standard builtin __import__() is called. Other mechanisms for invoking the import system (such as importlib.import_module()) do not perform this check and will always use the standard import system.

When a module is first imported, Python searches for the module and if found, it creates a module object [1], initializing it. If the named module cannot be found, an ImportError is raised. Python implements various strategies to search for the named module when the import machinery is invoked. These strategies can be modified and extended by using various hooks described in the sections below.

Changed in version 3.3: The import system has been updated to fully implement the second phase of PEP 302. There is no longer any implicit import machinery - the full import system is exposed through sys.meta_path. In addition, native namespace package support has been implemented (see PEP 420).

5.1. importlib

The importlib module provides a rich API for interacting with the import system. For example importlib.import_module() provides a recommended, simpler API than built-in __import__() for invoking the import machinery. Refer to the importlib library documentation for additional detail.

5.2. Packages

Python has only one type of module object, and all modules are of this type, regardless of whether the module is implemented in Python, C, or something else. To help organize modules and provide a naming hierarchy, Python has a concept of packages.

You can think of packages as the directories on a file system and modules as files within directories, but don’t take this analogy too literally since packages and modules need not originate from the file system. For the purposes of this documentation, we’ll use this convenient analogy of directories and files. Like file system directories, packages are organized hierarchically, and packages may themselves contain subpackages, as well as regular modules.

It’s important to keep in mind that all packages are modules, but not all modules are packages. Or put another way, packages are just a special kind of module. Specifically, any module that contains a __path__ attribute is considered a package.

All modules have a name. Subpackage names are separated from their parent package name by dots, akin to Python’s standard attribute access syntax. Thus you might have a module called sys and a package called email, which in turn has a subpackage called email.mime and a module within that subpackage called email.mime.text.

5.2.1. Regular packages

Python defines two types of packages, regular packages and namespace packages. Regular packages are traditional packages as they existed in Python 3.2 and earlier. A regular package is typically implemented as a directory containing an __init__.py file. When a regular package is imported, this __init__.py file is implicitly executed, and the objects it defines are bound to names in the package’s namespace. The __init__.py file can contain the same Python code that any other module can contain, and Python will add some additional attributes to the module when it is imported.

For example, the following file system layout defines a top level parent package with three subpackages:

parent/
    __init__.py
    one/
        __init__.py
    two/
        __init__.py
    three/
        __init__.py

Importing parent.one will implicitly execute parent/__init__.py and parent/one/__init__.py. Subsequent imports of parent.two or parent.three will execute parent/two/__init__.py and parent/three/__init__.py respectively.

5.2.2. Namespace packages

A namespace package is a composite of various portions, where each portion contributes a subpackage to the parent package. Portions may reside in different locations on the file system. Portions may also be found in zip files, on the network, or anywhere else that Python searches during import. Namespace packages may or may not correspond directly to objects on the file system; they may be virtual modules that have no concrete representation.

Namespace packages do not use an ordinary list for their __path__ attribute. They instead use a custom iterable type which will automatically perform a new search for package portions on the next import attempt within that package if the path of their parent package (or sys.path for a top level package) changes.

With namespace packages, there is no parent/__init__.py file. In fact, there may be multiple parent directories found during import search, where each one is provided by a different portion. Thus parent/one may not be physically located next to parent/two. In this case, Python will create a namespace package for the top-level parent package whenever it or one of its subpackages is imported.

See also PEP 420 for the namespace package specification.

5.3. Searching

To begin the search, Python needs the fully qualified name of the module (or package, but for the purposes of this discussion, the difference is immaterial) being imported. This name may come from various arguments to the import statement, or from the parameters to the importlib.import_module() or __import__() functions.

This name will be used in various phases of the import search, and it may be the dotted path to a submodule, e.g. foo.bar.baz. In this case, Python first tries to import foo, then foo.bar, and finally foo.bar.baz. If any of the intermediate imports fail, an ImportError is raised.

5.3.1. The module cache

The first place checked during import search is sys.modules. This mapping serves as a cache of all modules that have been previously imported, including the intermediate paths. So if foo.bar.baz was previously imported, sys.modules will contain entries for foo, foo.bar, and foo.bar.baz. Each key will have as its value the corresponding module object.

During import, the module name is looked up in sys.modules and if present, the associated value is the module satisfying the import, and the process completes. However, if the value is None, then an ImportError is raised. If the module name is missing, Python will continue searching for the module.

sys.modules is writable. Deleting a key may not destroy the associated module (as other modules may hold references to it), but it will invalidate the cache entry for the named module, causing Python to search anew for the named module upon its next import. The key can also be assigned to None, forcing the next import of the module to result in an ImportError.

Beware though, as if you keep a reference to the module object, invalidate its cache entry in sys.modules, and then re-import the named module, the two module objects will not be the same. By contrast, imp.reload() will reuse the same module object, and simply reinitialise the module contents by rerunning the module’s code.

5.3.2. Finders and loaders

If the named module is not found in sys.modules, then Python’s import protocol is invoked to find and load the module. This protocol consists of two conceptual objects, finders and loaders. A finder’s job is to determine whether it can find the named module using whatever strategy it knows about. Objects that implement both of these interfaces are referred to as importers - they return themselves when they find that they can load the requested module.

Python includes a number of default finders and importers. The first one knows how to locate built-in modules, and the second knows how to locate frozen modules. A third default finder searches an import path for modules. The import path is a list of locations that may name file system paths or zip files. It can also be extended to search for any locatable resource, such as those identified by URLs.

The import machinery is extensible, so new finders can be added to extend the range and scope of module searching.

Finders do not actually load modules. If they can find the named module, they return a module spec, an encapsulation of the module’s import-related information, which the import machinery then uses when loading the module.

The following sections describe the protocol for finders and loaders in more detail, including how you can create and register new ones to extend the import machinery.

Changed in version 3.4: In previous versions of Python, finders returned loaders directly, whereas now they return module specs which contain loaders. Loaders are still used during import but have fewer responsibilities.

5.3.3. Import hooks

The import machinery is designed to be extensible; the primary mechanism for this are the import hooks. There are two types of import hooks: meta hooks and import path hooks.

Meta hooks are called at the start of import processing, before any other import processing has occurred, other than sys.modules cache look up. This allows meta hooks to override sys.path processing, frozen modules, or even built-in modules. Meta hooks are registered by adding new finder objects to sys.meta_path, as described below.

Import path hooks are called as part of sys.path (or package.__path__) processing, at the point where their associated path item is encountered. Import path hooks are registered by adding new callables to sys.path_hooks as described below.

5.3.4. The meta path

When the named module is not found in sys.modules, Python next searches sys.meta_path, which contains a list of meta path finder objects. These finders are queried in order to see if they know how to handle the named module. Meta path finders must implement a method called find_spec() which takes two arguments, a name and an import path. The meta path finder can use any strategy it wants to determine whether it can handle the named module or not.

If the meta path finder knows how to handle the named module, it returns a spec object. If it cannot handle the named module, it returns None. If sys.meta_path processing reaches the end of its list without returning a spec, then an ImportError is raised. Any other exceptions raised are simply propagated up, aborting the import process.

The find_spec() method of meta path finders is called with two arguments. The first is the fully qualified name of the module being imported, for example foo.bar.baz. The second argument is the path entries to use for the module search. For top-level modules, the second argument is None, but for submodules or subpackages, the second argument is the value of the parent package’s __path__ attribute. If the appropriate __path__ attribute cannot be accessed, an ImportError is raised.

The meta path may be traversed multiple times for a single import request. For example, assuming none of the modules involved has already been cached, importing foo.bar.baz will first perform a top level import, calling mpf.find_spec("foo", None) on each meta path finder (mpf). After foo has been imported, foo.bar will be imported by traversing the meta path a second time, calling mpf.find_spec("foo.bar", foo.__path__). Once foo.bar has been imported, the final traversal will call mpf.find_spec("foo.bar.baz", foo.bar.__path__).

Some meta path finders only support top level imports. These importers will always return None when anything other than None is passed as the second argument.

Python’s default sys.meta_path has three meta path finders, one that knows how to import built-in modules, one that knows how to import frozen modules, and one that knows how to import modules from an import path (i.e. the path based finder).

Changed in version 3.4: The find_spec() method of meta path finders replaced find_module(). which is now deprecated. While it will continue to work without change, the import machinery will try it only if the finder does not implement find_spec().

5.4. Loading

If and when a module spec is found, the import machinery will use it (and the loader it contains) when loading the module. Here is an approximation of what happens during the loading portion of import:

module = None
if spec.loader is not None and hasattr(spec.loader, 'create_module'):
    module = spec.loader.create_module(spec)
if module is None:
    module = ModuleType(spec.name)
# The import-related module attributes get set here:
_init_module_attrs(spec, module)

if spec.loader is None:
    if spec.submodule_search_locations is not None:
        # namespace package
        sys.modules[spec.name] = module
    else:
        # unsupported
        raise ImportError
elif not hasattr(spec.loader, 'exec_module'):
    module = spec.loader.load_module(spec.name)
else:
    sys.modules[spec.name] = module
    try:
        spec.loader.exec_module(module)
    except BaseException:
        try:
            del sys.modules[spec.name]
        except KeyError:
            pass
        raise
module_to_return = sys.modules[spec.name]

Note the following details:

  • If there is an existing module object with the given name in sys.modules, import will have already returned it.
  • The module will exist in sys.modules before the loader executes the module code. This is crucial because the module code may (directly or indirectly) import itself; adding it to sys.modules beforehand prevents unbounded recursion in the worst case and multiple loading in the best.
  • If loading fails, the failing module – and only the failing module – gets removed from sys.modules. Any module already in the sys.modules cache, and any module that was successfully loaded as a side-effect, must remain in the cache. This contrasts with reloading where even the failing module is left in sys.modules.
  • After the module is created but before execution, the import machinery sets the import-related module attributes (“init_module_attrs”), as summarized in a later section.
  • Module execution is the key moment of loading in which the module’s namespace gets populated. Execution is entirely delegated to the loader, which gets to decide what gets populated and how.
  • The module created during loading and passed to exec_module() may not be the one returned at the end of import [2].

Changed in version 3.4: The import system has taken over the boilerplate responsibilities of loaders. These were previously performed by the load_module() method.

5.4.1. Loaders

Module loaders provide the critical function of loading: module execution. The import machinery calls the exec_module() method with a single argument, the module object to execute. Any value returned from exec_module() is ignored.

Loaders must satisfy the following requirements:

  • If the module is a Python module (as opposed to a built-in module or a dynamically loaded extension), the loader should execute the module’s code in the module’s global name space (module.__dict__).
  • If loader cannot execute the module, it should raise an ImportError, although any other exception raised during exec_module() will be propagated.

In many cases, the finder and loader can be the same object; in such cases the finder.find_spec() would just return a spec with the loader set to self.

Module loaders may opt in to creating the module object during loading by implementing a create_module() method. It takes one argument, the module spec, and returns the new module object to use during loading. create_module() does not need to set any attributes on the module object. If the loader does not define create_module(), the import machinery will create the new module itself.

New in version 3.4: The create_module() method of loaders.

Changed in version 3.4: The load_module() method was replaced by exec_module() and the import machinery assumed all the boilerplate responsibilities of loading.

For compatibility with existing loaders, the import machinery will use the load_module() method of loaders if it exists and the loader does not also implement exec_module(). However, load_module() has been deprecated and loaders should implement exec_module() instead.

The load_module() method must implement all the boilerplate loading functionality described above in addition to executing the module. All the same constraints apply, with some additional clarification:

  • If there is an existing module object with the given name in sys.modules, the loader must use that existing module. (Otherwise, imp.reload() will not work correctly.) If the named module does not exist in sys.modules, the loader must create a new module object and add it to sys.modules.
  • The module must exist in sys.modules before the loader executes the module code, to prevent unbounded recursion or multiple loading.
  • If loading fails, the loader must remove any modules it has inserted into sys.modules, but it must remove only the failing module, and only if the loader itself has loaded it explicitly.

5.4.2. Module spec

The import machinery uses a variety of information about each module during import, especially before loading. Most of the information is common to all modules. The purpose of a module’s spec is to encapsulate this import-related information on a per-module basis.

Using a spec during import allows state to be transferred between import system components, e.g. between the finder that creates the module spec and the loader that executes it. Most importantly, it allows the import machinery to perform the boilerplate operations of loading, whereas without a module spec the loader had that responsibility.

See ModuleSpec for more specifics on what information a module’s spec may hold.

New in version 3.4.

5.4.4. module.__path__

By definition, if a module has an __path__ attribute, it is a package, regardless of its value.

A package’s __path__ attribute is used during imports of its subpackages. Within the import machinery, it functions much the same as sys.path, i.e. providing a list of locations to search for modules during import. However, __path__ is typically much more constrained than sys.path.

__path__ must be an iterable of strings, but it may be empty. The same rules used for sys.path also apply to a package’s __path__, and sys.path_hooks (described below) are consulted when traversing a package’s __path__.

A package’s __init__.py file may set or alter the package’s __path__ attribute, and this was typically the way namespace packages were implemented prior to PEP 420. With the adoption of PEP 420, namespace packages no longer need to supply __init__.py files containing only __path__ manipulation code; the import machinery automatically sets __path__ correctly for the namespace package.

5.4.5. Module reprs

By default, all modules have a usable repr, however depending on the attributes set above, and in the module’s spec, you can more explicitly control the repr of module objects.

If the module has a spec (__spec__), the import machinery will try to generate a repr from it. If that fails or there is no spec, the import system will craft a default repr using whatever information is available on the module. It will try to use the module.__name__, module.__file__, and module.__loader__ as input into the repr, with defaults for whatever information is missing.

Here are the exact rules used:

  • If the module has a __spec__ attribute, the information in the spec is used to generate the repr. The “name”, “loader”, “origin”, and “has_location” attributes are consulted.
  • If the module has a __file__ attribute, this is used as part of the module’s repr.
  • If the module has no __file__ but does have a __loader__ that is not None, then the loader’s repr is used as part of the module’s repr.
  • Otherwise, just use the module’s __name__ in the repr.

Changed in version 3.4: Use of loader.module_repr() has been deprecated and the module spec is now used by the import machinery to generate a module repr.

For backward compatibility with Python 3.3, the module repr will be generated by calling the loader’s module_repr() method, if defined, before trying either approach described above. However, the method is deprecated.

5.5. The Path Based Finder

As mentioned previously, Python comes with several default meta path finders. One of these, called the path based finder, searches an import path, which contains a list of path entries. Each path entry names a location to search for modules.

The path based finder itself doesn’t know how to import anything. Instead, it traverses the individual path entries, associating each of them with a path entry finder that knows how to handle that particular kind of path.

The default set of path entry finders implement all the semantics for finding modules on the file system, handling special file types such as Python source code (.py files), Python byte code (.pyc and .pyo files) and shared libraries (e.g. .so files). When supported by the zipimport module in the standard library, the default path entry finders also handle loading all of these file types (other than shared libraries) from zipfiles.

Path entries need not be limited to file system locations. They can refer to URLs, database queries, or any other location that can be specified as a string.

The path based finder provides additional hooks and protocols so that you can extend and customize the types of searchable path entries. For example, if you wanted to support path entries as network URLs, you could write a hook that implements HTTP semantics to find modules on the web. This hook (a callable) would return a path entry finder supporting the protocol described below, which was then used to get a loader for the module from the web.

A word of warning: this section and the previous both use the term finder, distinguishing between them by using the terms meta path finder and path entry finder. These two types of finders are very similar, support similar protocols, and function in similar ways during the import process, but it’s important to keep in mind that they are subtly different. In particular, meta path finders operate at the beginning of the import process, as keyed off the sys.meta_path traversal.

By contrast, path entry finders are in a sense an implementation detail of the path based finder, and in fact, if the path based finder were to be removed from sys.meta_path, none of the path entry finder semantics would be invoked.

5.5.1. Path entry finders

The path based finder is responsible for finding and loading Python modules and packages whose location is specified with a string path entry. Most path entries name locations in the file system, but they need not be limited to this.

As a meta path finder, the path based finder implements the find_spec() protocol previously described, however it exposes additional hooks that can be used to customize how modules are found and loaded from the import path.

Three variables are used by the path based finder, sys.path, sys.path_hooks and sys.path_importer_cache. The __path__ attributes on package objects are also used. These provide additional ways that the import machinery can be customized.

sys.path contains a list of strings providing search locations for modules and packages. It is initialized from the PYTHONPATH environment variable and various other installation- and implementation-specific defaults. Entries in sys.path can name directories on the file system, zip files, and potentially other “locations” (see the site module) that should be searched for modules, such as URLs, or database queries. Only strings and bytes should be present on sys.path; all other data types are ignored. The encoding of bytes entries is determined by the individual path entry finders.

The path based finder is a meta path finder, so the import machinery begins the import path search by calling the path based finder’s find_spec() method as described previously. When the path argument to find_spec() is given, it will be a list of string paths to traverse - typically a package’s __path__ attribute for an import within that package. If the path argument is None, this indicates a top level import and sys.path is used.

The path based finder iterates over every entry in the search path, and for each of these, looks for an appropriate path entry finder for the path entry. Because this can be an expensive operation (e.g. there may be stat() call overheads for this search), the path based finder maintains a cache mapping path entries to path entry finders. This cache is maintained in sys.path_importer_cache (despite the name, this cache actually stores finder objects rather than being limited to importer objects). In this way, the expensive search for a particular path entry location’s path entry finder need only be done once. User code is free to remove cache entries from sys.path_importer_cache forcing the path based finder to perform the path entry search again [3].

If the path entry is not present in the cache, the path based finder iterates over every callable in sys.path_hooks. Each of the path entry hooks in this list is called with a single argument, the path entry to be searched. This callable may either return a path entry finder that can handle the path entry, or it may raise ImportError. An ImportError is used by the path based finder to signal that the hook cannot find a path entry finder for that path entry. The exception is ignored and import path iteration continues. The hook should expect either a string or bytes object; the encoding of bytes objects is up to the hook (e.g. it may be a file system encoding, UTF-8, or something else), and if the hook cannot decode the argument, it should raise ImportError.

If sys.path_hooks iteration ends with no path entry finder being returned, then the path based finder’s find_spec() method will store None in sys.path_importer_cache (to indicate that there is no finder for this path entry) and return None, indicating that this meta path finder could not find the module.

If a path entry finder is returned by one of the path entry hook callables on sys.path_hooks, then the following protocol is used to ask the finder for a module spec, which is then used when loading the module.

5.5.2. Path entry finder protocol

In order to support imports of modules and initialized packages and also to contribute portions to namespace packages, path entry finders must implement the find_spec() method.

find_spec() takes one argument, the fully qualified name of the module being imported. find_spec() returns a fully populated spec for the module. This spec will always have “loader” set (with one exception).

To indicate to the import machinery that the spec represents a namespace portion. the path entry finder sets “loader” on the spec to None and “submodule_search_locations” to a list containing the portion.

Changed in version 3.4: find_spec() replaced find_loader() and find_module(), but of which are now deprecated, but will be used if find_spec() is not defined.

Older path entry finders may implement one of these two deprecated methods instead of find_spec(). The methods are still respected for the sake of backward compatibility. Howevever, if find_spec() is implemented on the path entry finder, the legacy methods are ignored.

find_loader() takes one argument, the fully qualified name of the module being imported. find_loader() returns a 2-tuple where the first item is the loader and the second item is a namespace portion. When the first item (i.e. the loader) is None, this means that while the path entry finder does not have a loader for the named module, it knows that the path entry contributes to a namespace portion for the named module. This will almost always be the case where Python is asked to import a namespace package that has no physical presence on the file system. When a path entry finder returns None for the loader, the second item of the 2-tuple return value must be a sequence, although it can be empty.

If find_loader() returns a non-None loader value, the portion is ignored and the loader is returned from the path based finder, terminating the search through the path entries.

For backwards compatibility with other implementations of the import protocol, many path entry finders also support the same, traditional find_module() method that meta path finders support. However path entry finder find_module() methods are never called with a path argument (they are expected to record the appropriate path information from the initial call to the path hook).

The find_module() method on path entry finders is deprecated, as it does not allow the path entry finder to contribute portions to namespace packages. If both find_loader() and find_module() exist on a path entry finder, the import system will always call find_loader() in preference to find_module().

5.6. Replacing the standard import system

The most reliable mechanism for replacing the entire import system is to delete the default contents of sys.meta_path, replacing them entirely with a custom meta path hook.

If it is acceptable to only alter the behaviour of import statements without affecting other APIs that access the import system, then replacing the builtin __import__() function may be sufficient. This technique may also be employed at the module level to only alter the behaviour of import statements within that module.

To selectively prevent import of some modules from a hook early on the meta path (rather than disabling the standard import system entirely), it is sufficient to raise ImportError directly from find_spec() instead of returning None. The latter indicates that the meta path search should continue. while raising an exception terminates it immediately.

5.7. Open issues

XXX It would be really nice to have a diagram.

XXX * (import_machinery.rst) how about a section devoted just to the attributes of modules and packages, perhaps expanding upon or supplanting the related entries in the data model reference page?

XXX runpy, pkgutil, et al in the library manual should all get “See Also” links at the top pointing to the new import system section.

5.8. References

The import machinery has evolved considerably since Python’s early days. The original specification for packages is still available to read, although some details have changed since the writing of that document.

The original specification for sys.meta_path was PEP 302, with subsequent extension in PEP 420.

PEP 420 introduced namespace packages for Python 3.3. PEP 420 also introduced the find_loader() protocol as an alternative to find_module().

PEP 366 describes the addition of the __package__ attribute for explicit relative imports in main modules.

PEP 328 introduced absolute and explicit relative imports and initially proposed __name__ for semantics PEP 366 would eventually specify for __package__.

PEP 338 defines executing modules as scripts.

PEP 451 adds the encapsulation of per-module import state in spec objects. It also off-loads most of the boilerplate responsibilities of loaders back onto the import machinery. These changes allow the deprecation of several APIs in the import system and also addition of new methods to finders and loaders.

Footnotes

[1]See types.ModuleType.
[2]The importlib implementation avoids using the return value directly. Instead, it gets the module object by looking the module name up in sys.modules. The indirect effect of this is that an imported module may replace itself in sys.modules. This is implementation-specific behavior that is not guaranteed to work in other Python implementations.
[3]In legacy code, it is possible to find instances of imp.NullImporter in the sys.path_importer_cache. It is recommended that code be changed to use None instead. See Porting Python code for more details.