🕵️ PyDependence 🐍

Python local package dependency discovery and resolution

Contributions are welcome!

---

Table Of Contents

• Overview
  • Why
  • How This Works
• Configuration
• Usage
  • Usage - Pre-Commit
  • Usage - CLI
• Help
  • Version Mapping
  • Scopes
    • Sub-Scopes
  • Output Resolvers
  • Example

---

Overview

If multiple dependencies are listed in a project, only some of them may actually be required! This project finds those dependencies!

Why

This project was created for multiple reasons

• Find missing dependencies
• Generate optional dependencies lists, eg. for pyproject.toml
• Create minimal dockerfiles with only the dependencies that are needed for a specific entrypoint

How This Works

1. Specify root python packages to search through (we call this the namespace)
   • This can either be modules under a folder, similar to PYTHONPATH
   • Or actual paths to modules
2. The AST of each python file is parsed, and import statements are found
3. Finally, dependencies are resolved using graph traversal and flattened.
   • imports that redirect to modules within the current scope are flattened and replaced with imports not in the scope.

---

Configuration

Check the pyproject.toml for detailed explanations of various config options and a working example of pydependence applied to itself.

It is recommended to specify the config inside your projects existing pyproject.toml file, however, pydepedence will override whatever is specified here if a .pydependence.cfg file exists in the same folder. (This behavior is needed if for example a project is still using a setup.py file, or migrating from this.)

Configuration using the pyproject.toml should be placed under the [tool.pydependence] table, while configuration for the .pydependence.cfg should be placed under the root table.

Here is a minimal example:

# ... rest of pyproject.toml file ...

[tool.pydependence]  # exclude table definition if inside `.pydependence.cfg`, place all attributes at root instead.
versions = ["tomlkit>=0.12,<1"]
scopes = [{name = "pydependence", pkg_paths = "./pydependence"}]
resolvers = [
    {strict_requirements_map=false, scope='pydependence', output_mode='dependencies'},
    {strict_requirements_map=false, scope='pydependence', output_mode='optional-dependencies', output_name='all', visit_lazy=true},
]

# ... rest of pyproject.toml file ...

---

Usage

pydependence can be triggered from both the CLI and using pre-commit, and currently requires python>=3.8, however, it should still be able to run in a virtual environment over legacy python code.

Usage - Pre-Commit

Add a pre-commit entry pointing to your pyproject.toml file or configuration file.

  - repo: https://github.com/nmichlo/pydependence
    rev: v0.5.0
    hooks:
      - id: pydependence
        args: ["pyproject.toml"]

Usage - CLI

Manually invoke pydependence

# install
pip install pydependence

# manual invocation - help
python -m pydependence --help

# manual invocation
python -m pydependence <path_to_config.toml>

---

Help

pydependence is an AST imports analysis tool that is used to discover the imports of a package and generate a dependency graph and requirements/pyproject sections.

pydependence is NOT a package manager or a dependency resolver. This is left to the tool of your choice, e.g. pip, poetry, pip-compile, uv, etc.

Check the pyproject.toml for detailed explanations of various config options and a working example of pydependence applied to itself.

Version Mapping

Versions are used to specify the version of a package that should be used when generating output requirements.

• If a version is not specified, an error will be raised.

Versions are also used to construct mappings between package names and import names.

• e.g. Pillow is imported as PIL, so the version mapping is {package="pillow", version="*", import="PIL"}

Scopes

A scope is a logical collection of packages. It is a way to group packages together for the purpose of dependency resolution.

• NOTE: there cannot be conflicting module imports within a scope.

Scopes can inherit from other scopes. Scopes can have filters applied to them, include & exclude. Scopes must have unique names.

The order of constructing a single scope is important.

1. parents, search_paths, pkg_paths
   • parents: inherit all modules from the specified scopes
   • search_paths: search for packages inside the specified paths (like PYTHONPATH)
   • pkg_paths: add the packages at the specified paths
2. limit, include, exclude
   • limit: limit the search space to children of the specified packages
   • include: include packages that match the specified patterns
   • exclude: exclude packages that match the specified patterns

The order of evaluation when constucting multiple scopes is important, and can be used to create complex dependency resolution strategies.

• all scopes are constructed in order of definition

Sub-Scopes

A subscope is simply an alias for constructing a new scope, where:

• the parent scope is the current scope
• a filter is applied to limit the packages

e.g.

[[tool.pydependence.scopes]]
name = "my_pkg"
pkg_paths = ["my_pkg"]
subscopes = {mySubPkg="my_pkg.my_sub_pkg"}

is the same as:

[[tool.pydependence.scopes]]
name = "my_pkg"
pkg_paths = ["my_pkg"]

[[tool.pydependence.scopes]]
name = "mySubPkg"
parents = ["my_pkg"]
limit = ["my_pkg.my_sub_pkg"]

why?

• This simplifies syntax for the common pattern of when you want to resolve optional dependencies across an entire package, but only want to traverse starting from the subscope.

Output Resolvers

Resolvers are used to specify how to resolve dependencies, and where to output the results.

options:

• scope:
  • is used to determine the search space for the resolver.
• start_scope:
  • is used to determine the starting point for the resolver, i.e. BFS across all imports occurs from this point.
• env
  • used to select a specific set of versions that are tagged with the same env key when resolving.
• raw
  • manually specify requirements and versions to output, overwriting what was resolved if conflicting.
• output_mode:
  • is used to determine where to output the results.
  • valid options are: dependencies, optional-dependencies, or requirements
• output_file:
  • is used to specify the file to output the results to, by default this is the current pyproject.toml file. this usually only needs to be specified when outputting to a different file like requirements.txt
• output_name
  • only applied if using output_mode="optional-dependencies", specifies the extras group name.

Note: We can have multiple resolvers to construct different sets of outputs. For example if you have a library with core dependencies and optional dependencies, you can construct a resolver for each. And limit the results for the optional dependencies to only output the optional dependencies for that resolver.

Example

# general settings / overrides
# * [tool.pydependence] in pyproject.toml
# * [pydependence] in a standalone config file
[tool.pydependence]

# defaults [don't need to specifiy in practice, unless you want different default behavior]:
# - if a relative path, then relative to the *parent* of this file
# - the resolved root is prepended to all relative paths in the rest of this config.
#   e.g. for the `pydependence` repo, with files `pydependence/pyproject.toml` and `pydependence/pydependence`, then
#        * ".." would mean relative files need to be references from the parent of the repo root e.g. "pydependence/pydependence/__main__.py",
#          this is useful when resolving across multiple repos.
#        * "." would mean relative files need to be reference from the repo root e.g. "pydependence/__main__.py",
#          this is more useful when resolving across a single repo (default).
default_root = "."

# defaults [don't need to specifiy in practice]:
# - these settings can be overridden on individual output resolvers.
# * visit_lazy:
#   | If true, then vist all the lazy imports. Usually the lazy imports are removed from
#   | the import graph and we don't traverse these edges. This on the other-hand allows
#   | all these edges to be traversed. This is often useful if you want to create
#   | packages that only require some minimal set of requirements, and everything else
#   | that you define should be optional. Also useful if you want to generate a minimal
#   | dependencies list, and then in optional dependency lists you want to create a full
#   | set of requirements for everything!
# * re_add_lazy: [A.K.A. shallow_include_lazy]
#   | only applicable when `visit_lazy=False`, then in this case we re-add the lazy
#   | imports that are directly referenced in all the traversed files, i.e. it is a
#   | shallow include of lazy imports without continuing to traverse them. This means
#   | that we add the lazy imports but we don't continue traversing. This is often not
#   | useful with complex lazy import graphs that continue to reference more module
#   | within the same scope as this could cause missing imports, rather specify
#   | `visit_lazy=True` in this case.
# * exclude_unvisited:
#   | If true, then exclude imports that were not encountered as we traversed the import
#   | graph. [NOTE]: this is probably useful if you don't want to include all imports
#   | below a specific scope, but only want to resolve what is actually encountered.
#   | Not entirely sure this has much of an effect?
# * exclude_in_search_space:
#   | If true, then exclude all imports that are part of the current scope. This usually
#   | should not have any effect because imports are replaced as we traverse the graph
#   | through the current scope, [NOTE] thus not entirely sure that this has any effect,
#   | should it be a bug if we encounter any of these?
# * exclude_builtins:
#   | If true, then exclude all the python builtin package names from being output in
#   | the requirements files. This usually should be true unless you are trying to debug
#   | as this would generate invalid requirements list as these would not exist on pypi.
# * strict_requirements_map:
#   | Check that generated imports and requirements have entries in the versions list.
#   | If strict mode is enabled, then an error is thrown if a version entry is missing.
#   | If strict mode is disabled, then a warning should be given, and the root import
#   | name is used instead of the requirement name, which may or may not match up
#   | to an actual python package.
default_resolve_rules = {visit_lazy=false, re_add_lazy=false, exclude_unvisited=true, exclude_in_search_space=true, exclude_builtins=true, strict_requirements_map=true}

# defaults [don't need to specifiy in practice, unless you want different default behavior]:
# - these settings can be overridden on individual scopes.
# * unreachable_mode
#   | Specify how to handle modules that are unreachable, e.g. if there is no `__init__.py`
#   | file in all the parents leading up to importing this module. If this is the case
#   | then the module/package does not correctly follow python/PEP convention and is
#   | technically invalid. By default, for `error`, we raise an exception and do not allow
#   | the scope to be created, but this can be relaxed to `skip` or `keep` these files.
default_scope_rules = {unreachable_mode="error"}

# map requirements and resolved imports to specific packages and version requirements.
# - to generate dependency lists for conflicting package versions you can specify
#   requirements more than once as long as you add a unique `env` entry. In the
#   corresponding resolver, you can then reference this `env` entry to switch out the
#   version you are after. If no `env` is set, then by default the env="default".
versions = [
    "pydantic>=2.0.0",
    {requirement="pydependence", scope="pydependence"},
]

# Output resolvers define the resolution and generation of output requirements by
#   1. traversing a scope's directed module and import graph
#   2. collecting all imports and mapping these using the version list above to specific pypi requirements.
# Note that the starting point can be set to some subset of the scope
#   e.g. if you script that you want to generate requirements for uses the `main.py` file,
#        then you can generate requirements for this script specifically by setting it as
#        the starting point, instead of generating requirements for the entire scope. Then
#        only imports that this script uses and traverses to will actually be included in
#        the output.
# Note that by default lazy imports are not traversed or included.
#   * Lazy imports are imports that are contained inside functions, methods or TYPE_CHECKING blocks.
#     This means that some coding convention has to be followed when pydependence is used within
#     projects to generate minimal requirements as it is expected that these imports should not be
#     called at import time, instead possibly only at runtime if specific features of the library or
#     framework are requested.
# Note that by default we use a strict requirements mapping mode
#   * if resolved imports and requirements do not have a corresponding `versions` entry, then an error will
#     be thrown in this case. This is to ensure that devs know exactly what versions are being used within their
#     program.
resolvers = [
    {                   output_mode='dependencies',          scope='pydependence', visit_lazy=false, strict_requirements_map=false},
    {output_name='all', output_mode='optional-dependencies', scope='pydependence', visit_lazy=true,  strict_requirements_map=false},
]

# Scopes represent graphs of modules (nodes) and their interconnecting
# import statements (directed edges) that reference themselves or other modules.
# - scopes are traversed by the resolvers in different ways to generate lists of requirements.
# - scopes can be constructed from various different sources.
#   * `pkg_paths` / `search_paths`
# - scopes can inherit from eachother with `parents`
# - scopes can be filtered down with `limit`
# - scopes cannot have conflicting module names, each module name should have a one to one mapping to a specific file.
scopes = [
    {name = "pydependence", pkg_paths = "./pydependence"},
]