Usage

Grimp provides an API in the form of an ImportGraph that represents all the imports within one or more top-level Python packages. This object has various methods that make it easy to find out information about the packages’ structures and interdependencies.

Terminology

The terminology around Python packages and modules can be a little confusing. Here are the definitions we use, taken in part from the official Python docs:

Module: A file containing Python definitions and statements. This includes ordinary .py files and __init__.py files.
Package: A Python module which can contain submodules or recursively, subpackages.
Top Level Package: A package that is not a subpackage of another package.
Graph: A graph in the mathematical sense of a collection of items with relationships between them. Grimp’s ImportGraph is a directed graph of imports between modules.
Direct Import: An import from one module to another.
Import Chain: A chain of direct imports between two modules, possibly via other modules. For example, if mypackage.foo imports mypackage.bar, which in turn imports mypackage.baz, then there is an import chain between mypackage.foo and mypackage.baz.
Squashed Module: A module in the graph that represents both itself and all its descendants. Squashed modules allow parts of the graph to be simplified. For example, if you include external packages when building the graph, each external package will exist in the graph as a single squashed module.

Building the graph

import grimp

# Single package
graph = grimp.build_graph('mypackage')

# Multiple packages
graph = grimp.build_graph('mypackage', 'anotherpackage', 'onemore')

# Include imports of external packages
graph = grimp.build_graph('mypackage', include_external_packages=True)

# Exclude imports within a TYPE_CHECKING guard
graph = grimp.build_graph('mypackage', exclude_type_checking_imports=True)

# Use a different cache directory, or disable caching altogether
graph = grimp.build_graph('mypackage', cache_dir="/path/to/cache")
graph = grimp.build_graph('mypackage', cache_dir=None)

grimp.build_graph(package_name, *additional_package_names, include_external_packages=False, exclude_type_checking_imports=False)

Build and return an ImportGraph for the supplied package or packages.

Parameters:

package_name (str) – The name of an importable package, for example 'mypackage'. For regular packages, this must be the top level package (i.e. one with no dots in its name). However, in the special case of namespace packages, the name of the portion should be supplied, for example 'mynamespace.foo'.
additional_package_names (tuple[str, ...]) – Tuple of any additional package names. These can be supplied as positional arguments, as in the example above.
include_external_packages (bool, optional) –
Whether to include external packages in the import graph. If this is True, any other top level packages (including packages in the standard library) that are imported by this package will be included in the graph as squashed modules (see Terminology above).

The behaviour is more complex if one of the internal packages is a namespace portion. In this case, the squashed module will have the shallowest name that doesn’t clash with any internal modules. For example, in a graph with internal packages namespace.foo and namespace.bar.one.green, namespace.bar.one.orange.alpha would be added to the graph as namespace.bar.one.orange. However, in a graph with only namespace.foo as an internal package, the same external module would be added as namespace.bar.

Note: external packages are only analysed as modules that are imported; any imports they make themselves will not be included in the graph.
exclude_type_checking_imports (bool, optional) – Whether to exclude imports made in type checking guards. If this is True, any import made under an if TYPE_CHECKING: statement will not be added to the graph. See the typing module documentation for reference. (The type checking guard is detected purely by looking for a statement in the form if TYPE_CHECKING or if {some_alias}.TYPE_CHECKING. It does not check whether TYPE_CHECKING is actually the attribute from the typing module.)
cache_dir (str, optional) – The directory to use for caching the graph. Defaults to .grimp_cache. To disable caching, pass None. See Caching.

Returns:

An import graph that you can use to analyse the package.

Return type:

ImportGraph

Methods for analysing the module tree

ImportGraph.modules

All the modules contained in the graph.

return:

Set of module names.

rtype:

A set of strings.

ImportGraph.find_children(module)

Return all the immediate children of the module, i.e. the modules that have a dotted module name that is one level below.

param str module:

The importable name of a module in the graph, e.g. 'mypackage' or 'mypackage.foo.one'. This may be any non-squashed module. It doesn’t need to be a package itself, though if it isn’t, it will have no children.

return:

Set of module names.

rtype:

A set of strings.

raises:

ValueError if the module is a squashed module, as by definition it represents both itself and all of its descendants.

ImportGraph.find_descendants(module)

Return all the descendants of the module, i.e. the modules that have a dotted module name that is below the supplied module, to any depth.

param str module:

The importable name of the module, e.g. 'mypackage' or 'mypackage.foo.one'. As with find_children, this doesn’t have to be a package, though if it isn’t then the set will be empty.

return:

Set of module names.

rtype:

A set of strings.

raises:

ValueError if the module is a squashed module, as by definition it represents both itself and all of its descendants.

Methods for analysing direct imports

ImportGraph.direct_import_exists(importer, imported, as_packages=False)

Parameters:

importer (str) – A module name.
imported (str) – A module name.
as_packages (bool) – Whether or not to treat the supplied modules as individual modules, or as entire packages (including any descendants).

Returns:

Whether or not the importer directly imports the imported module.

Return type:

True or False.

ImportGraph.find_modules_directly_imported_by(module)

Parameters:: module (str) – A module name.
Returns:: Set of all modules in the graph are imported by the supplied module.
Return type:: A set of strings.

ImportGraph.find_modules_that_directly_import(module)

Parameters:: module (str) – A module name.
Returns:: Set of all modules in the graph that directly import the supplied module.
Return type:: A set of strings.

ImportGraph.get_import_details(importer, imported)

Provides a way of seeing any available metadata about direct imports between two modules. Usually the list will consist of a single dictionary, but it is possible for a module to import another module more than once.

This method should not be used to determine whether an import is present: some of the imports in the graph may have no available metadata. For example, if an import has been added by the add_import method without the line_number and line_contents specified, then calling this method on the import will return an empty list. If you want to know whether the import is present, use direct_import_exists.

The details returned are in the following form:

[
    {
        'importer': 'mypackage.importer',
        'imported': 'mypackage.imported',
        'line_number': 5,
        'line_contents': 'from mypackage import imported',
    },
    # (additional imports here)
]

If no such import exists, or if there are no available details, an empty list will be returned.

Parameters:

importer (str) – A module name.
imported (str) – A module name.

Returns:

A list of any available metadata for imports between two modules.

Return type:

List of dictionaries with the structure shown above. If you want to use type annotations, you may use the grimp.DetailedImport TypedDict for each dictionary.

ImportGraph.count_imports()

Returns:: The number of direct imports in the graph.
Return type:: Integer.

Methods for analysing import chains

ImportGraph.find_downstream_modules(module, as_package=False)

Parameters:

module (str) – A module name.
as_package (bool) – Whether or not to treat the supplied module as an individual module, or as an entire package (including any descendants). If treating it as a package, the result will include downstream modules external to the supplied module, and won’t include modules within it.

Returns:

All the modules that import (even indirectly) the supplied module.

Return type:

A set of strings.

Examples:

# Returns the modules downstream of mypackage.foo.
graph.find_downstream_modules('mypackage.foo')

# Returns the modules downstream of mypackage.foo, mypackage.foo.one and
# mypackage.foo.two.
graph.find_downstream_modules('mypackage.foo', as_package=True)

ImportGraph.find_upstream_modules(module, as_package=False)

Parameters:

module (str) – A module name.
as_package (bool) – Whether or not to treat the supplied module as an individual module, or as a package (i.e. including any descendants, if there are any). If treating it as a subpackage, the result will include upstream modules external to the package, and won’t include modules within it.

Returns:

All the modules that are imported (even indirectly) by the supplied module.

Return type:

A set of strings.

ImportGraph.find_shortest_chain(importer, imported)

Parameters:

importer (str) – The module at the start of a potential chain of imports between importer and imported (i.e. the module that potentially imports imported, even indirectly).
imported (str) – The module at the end of the potential chain of imports.

Returns:

The shortest chain of imports between the supplied modules, or None if no chain exists.

Return type:

A tuple of strings, ordered from importer to imported modules, or None.

ImportGraph.find_shortest_chains(importer, imported)

Parameters:

importer (str) – A module or subpackage within the graph.
imported (str) – Another module or subpackage within the graph.

Returns:

The shortest import chains that exist between the importer and imported, and between any modules contained within them. Only one chain per upstream/downstream pair will be included. Any chains that are contained within other chains in the result set will be excluded.

Return type:

A set of tuples of strings. Each tuple is ordered from importer to imported modules.

ImportGraph.chain_exists(importer, imported, as_packages=False)

Parameters:

importer (str) – The module at the start of the potential chain of imports (as in find_shortest_chain).
imported (str) – The module at the end of the potential chain of imports (as in find_shortest_chain).
as_packages (bool) – Whether to treat the supplied modules as individual modules, or as packages (including any descendants, if there are any). If treating them as packages, all descendants of importer and imported will be checked too.

Returns:

Return whether any chain of imports exists between importer and imported, even indirectly; in other words, does importer depend on imported?

Return type:

bool

Higher level analysis

ImportGraph.find_illegal_dependencies_for_layers(layers, containers=None)

Find dependencies that don’t conform to the supplied layered architecture.

Parameters:

layers (Sequence[Layer | str | set[str]]) – A sequence of layers ordered from the highest to the lowest. The module names passed are relative to any containers passed in: for example, to specify mypackage.foo, you could either pass it in directly, or pass mypackage as the container (see the containers argument) and foo as the module name. A layer may optionally consist of multiple module names. If it does, the layer will by default treat each module as ‘independent’ (see below), though this can be overridden by passing independent=False when instantiating the Layer. For convenience, if a layer consists only of one module name then a string may be passed in place of the Layer object. Additionally, if the layer consists of multiple independent modules, that can be passed as a set of strings instead of a Layer object. Any modules specified that don’t exist in the graph will be silently ignored.
containers (set[str]) – The parent modules of the layers, as absolute names that you could import, such as mypackage.foo. (Optional.)

Returns:

The illegal dependencies in the form of a set of PackageDependency objects. Each package dependency is for a different permutation of two layers for which there is a violation, and contains information about the illegal chains of imports from the lower layer (the ‘importer’) to the higher layer (the ‘imported’).

Return type:

set[PackageDependency].

Raises:

grimp.exceptions.NoSuchContainer – if a container is not a module in the graph.

Overview

‘Layers’ is a software architecture pattern in which a list of modules/packages have a dependency direction from high to low. In other words, a higher layer would be allowed to import a lower layer, but not the other way around.

In this diagram, mypackage has a layered architecture in which the subpackage d is the highest layer and the subpackage a is the lowest layer. a would not be allowed to import from any of the modules above it, while d can import from everything. In the middle, c could import from a and b, but not d.

These layers can be individual .py modules or subpackages; if they’re subpackages then the architecture is enforced for all modules within the subpackage, so mypackage.a.one would not be allowed to import from mypackage.b.two.

Here’s how the architecture shown can be checked using Grimp:

dependencies = graph.find_illegal_dependencies_for_layers(
    layers=(
        "mypackage.d",
        "mypackage.c",
        "mypackage.b",
        "mypackage.a",
    ),
)

Containers

Containers allow for a less repetitive way of specifying layers, and are particularly useful if you want to specify a recurring pattern of layers in different places in the graph.

Example with containers:

dependencies = graph.find_illegal_dependencies_for_layers(
    layers=(
        "high",
        "medium",
        "low",
    ),
    containers={
        "mypackage.foo",
        "mypackage.bar",
    },
)

This call will check that, for example, mypackage.foo.low doesn’t import from mypackage.foo.medium. There is no checking between the containers, though, so mypackage.foo.low would be able to import mypackage.bar.high.

Layers containing multiple siblings

Grimp supports the presence of multiple sibling modules or packages within the same layer. In the diagram below, the modules blue and green are ‘independent’ in the same layer, meaning that, in addition to not being allowed to import from layers above them, they are not allowed to import from each other.

Architecture with a layer containing independent siblings.

An architecture like this can be checked by passing a set of module names:

dependencies = graph.find_illegal_dependencies_for_layers(
    layers=(
        "mypackage.d",
        {"mypackage.blue", "mypackage.green"},
        "mypackage.b",
        "mypackage.a",
    ),
)

Alternatively, siblings can be designated as non-independent, meaning that they are allowed to import from each other, as shown:

Architecture with a layer containing non-independent siblings.

To check this architecture, use the grimp.Layer class, specifying that the modules are not independent:

dependencies = graph.find_illegal_dependencies_for_layers(
    layers=(
        "mypackage.d",
        grimp.Layer("mypackage.blue", "mypackage.green", independent=False),
        "mypackage.b",
        "mypackage.a",
    ),
)

Return value

The method returns a set of PackageDependency objects that describe different illegal imports.

Note: each returned PackageDependency does not include all possible illegal Route objects. Instead, once an illegal Route is found, the algorithm will temporarily remove it from the graph before continuing with its search. As a result, any illegal Routes that have sections in common with other illegal Routes may not be returned.

Unfortunately the Routes included in the PackageDependencies are not, currently, completely deterministic. If there are multiple illegal Routes of the same length, it is not predictable which one will be found first. This means that the PackageDependencies returned can vary for the same graph.

class Layer

A layer within a layered architecture.

module_tails

set[str]: A set, each element of which is the final component of a module name. This ‘tail’ is combined with any container names to provide the full module name. For example, if a container is "mypackage" then to refer to "mypackage.foo" you would supply "foo" as the module tail.

independent

bool: Whether the sibling modules within this layer are required to be independent.

class PackageDependency

A collection of import dependencies from one Python package to another.

importer

str: The full name of the package within which all the routes start; the downstream package. E.g. “mypackage.foo”.

imported

str: The full name of the package within which all the routes end; the upstream package. E.g. “mypackage.bar”.

routes

frozenset[grimp.Route]: A set of Route objects from importer to imported.

class Route

A set of import chains that share the same middle.

The route fans in at the head and out at the tail, but the middle of the chain just links individual modules.

Example: the following Route represents a chain of imports from mypackage.orange -> mypackage.utils -> mypackage.helpers -> mypackage.green, plus an import from mypackage.red to mypackage.utils, and an import from mypackage.helpers to mypackage.blue:

Route(
    heads=frozenset(
        {
            "mypackage.orange",
            "mypackage.red",
        }
    ),
    middle=(
        "mypackage.utils",
        "mypackage.helpers",
    ),
    tails=frozenset(
        {
            "mypackage.green",
            "mypackage.blue",
        }
    ),
)

heads: frozenset[str]: The importer modules at the start of the chain.

middle: tuple[str]: A sequence of imports that link the head modules to the tail modules.

tails: frozenset[str]: Imported modules at the end of the chain.

Methods for manipulating the graph

ImportGraph.add_module(module, is_squashed=False)

Add a module to the graph.

Parameters:

module (str) – The name of a module, for example 'mypackage.foo'.
is_squashed (bool) – If True, the module should be treated as a ‘squashed module’ (see Terminology above).

Returns:

None

ImportGraph.remove_module(module)

Remove a module from the graph.

If the module is not present in the graph, no exception will be raised.

Parameters:: module (str) – The name of a module, for example 'mypackage.foo'.
Returns:: None

ImportGraph.add_import(importer, imported, line_number=None, line_contents=None)

Add a direct import between two modules to the graph. If the modules are not already present, they will be added to the graph.

Parameters:

importer (str) – The name of the module that is importing the other module.
imported (str) – The name of the module being imported.
line_number (int) – The line number of the import statement in the module.
line_contents (str) – The line that contains the import statement.

Returns:

None

ImportGraph.remove_import(importer, imported)

Remove a direct import between two modules. Does not remove the modules themselves.

Parameters:

importer (str) – The name of the module that is importing the other module.
imported (str) – The name of the module being imported.

Returns:

None

ImportGraph.squash_module(module)

‘Squash’ a module in the graph (see Terminology above).

Squashing a pre-existing module will cause all imports to and from the descendants of that module to instead point directly at the module being squashed. The import details (i.e. line numbers and contents) will be lost for those imports. The descendants will then be removed from the graph.

Parameters:: module (str) – The name of a module, for example 'mypackage.foo'.
Returns:: None

ImportGraph.is_module_squashed(module)

Return whether a module present in the graph is ‘squashed’ (see Terminology above).

Parameters:: module (str) – The name of a module, for example 'mypackage.foo'.
Returns:: bool