pipgrip

pipgrip is a lightweight pip dependency resolver with deptree preview functionality based on the PubGrub algorithm, which is also used by poetry. For one or more PEP 508 dependency specifications, pipgrip recursively fetches/builds the Python wheels necessary for version solving, and optionally renders the full resulting dependency tree.

$ pipgrip --tree fastapi~=0.94

fastapi~=0.94 (0.95.1)
├── pydantic!=1.7,!=1.7.1,!=1.7.2,!=1.7.3,!=1.8,!=1.8.1,<2.0.0,>=1.6.2 (1.10.7)
│   └── typing-extensions>=4.2.0 (4.5.0)
└── starlette<0.27.0,>=0.26.1 (0.26.1)
    └── anyio<5,>=3.4.0 (3.6.2)
        ├── idna>=2.8 (3.4)
        └── sniffio>=1.1 (1.3.0)

pipgrip vs. poetry

poetry offers package management with dependency resolution, essentially replacing pip/setuptools. This means that poetry packages don't contain setup.py, and hence are not compatible with pip install -e: poetry projects would have to be converted to setuptools-based projects with e.g. dephell. To avoid such hassle, pipgrip only requires the selected package(s) + dependencies to be available to pip in the usual way.

pipgrip vs. pipdeptree

For offline usage, pipdeptree can inspect the current environment and show how the currently installed packages relate to each other. This however requires the packages to be pip-installed, and (despite warnings about e.g. cyclic dependencies) offers no form of dependency resolution since it's only based on the (single) package versions installed in the environment. Such shortcomings are avoided when using pipgrip, since packages don't need to be installed and all versions available to pip are considered.

Installation

This pure-Python, OS independent package is available on PyPI:

pip install pipgrip

Usage

This package can be used to:

Render an exhaustive dependency tree for any given pip-compatible package(s):
- pipgrip --tree requests
Alleviate Python dependency hell by resolving the latest viable combination of required packages
Avoid bugs by running pipgrip as a stage in CI pipelines
Detect version conflicts for given constraints and give human readable feedback about it
Warn for cyclic dependencies in local projects [and install them anyway]:
- pipgrip -v --tree . [--install -e]
Install complex packages without worries:
- pipgrip --install aiobotocore[awscli]
Generate a lockfile with a complete working set of dependencies for reproducible installs:
- pipgrip --lock aiobotocore[awscli] && pip install aiobotocore[awscli] --constraint ./pipgrip.lock
Combine dependency trees of multiple packages into one unified set of pinned packages:
- pipgrip .[boto3] s3transfer==0.2.1 s3fs smart_open[s3]

See also known caveats.

Optionally, the environment variable PIPGRIP_ADDITIONAL_REQUIREMENTS can be populated with space/newline separated requirements, which will be appended to the requirements passed via CLI.

$ pipgrip --help

Usage: pipgrip [OPTIONS] [DEPENDENCIES]...

  pipgrip is a lightweight pip dependency resolver with deptree preview
  functionality based on the PubGrub algorithm, which is also used by poetry. For
  one or more PEP 508 dependency specifications, pipgrip recursively
  fetches/builds the Python wheels necessary for version solving, and optionally
  renders the full resulting dependency tree.

Options:
  --install                     Install full dependency tree after resolving.
  -e, --editable                Install a project in editable mode.
  --user                        Install to the Python user install directory for
                                your platform -- typically ~/.local/, or
                                %APPDATA%\Python on Windows.
  -r, --requirements-file FILE  Install from the given requirements file. This
                                option can be used multiple times.
  --lock                        Write out pins to './pipgrip.lock'.
  --pipe                        Output space-separated pins instead of newline-
                                separated pins.
  --json                        Output pins as JSON dict instead of newline-
                                separated pins. Combine with --tree for a detailed
                                nested JSON dependency tree.
  --sort                        Sort pins alphabetically before writing out. Can
                                be used bare, or in combination with --lock,
                                --pipe, --json, --tree-json, or --tree-json-exact.
  --tree                        Output human readable dependency tree (top-down).
                                Combine with --json for a detailed nested JSON
                                dependency tree. Use --tree-json instead for a
                                simplified JSON dependency tree (requirement
                                strings as keys, dependencies as values), or
                                --tree-json-exact for exact pins as keys.
  --tree-ascii                  Output human readable dependency tree with ASCII
                                tree markers.
  --reversed-tree               Output human readable dependency tree (bottom-up).
  --max-depth INTEGER           Maximum (JSON) tree rendering depth (default -1).
  --cache-dir DIRECTORY         Use a custom cache dir.
  --no-cache-dir                Disable pip cache for the wheels downloaded by
                                pipper. Overrides --cache-dir.
  --index-url TEXT              Base URL of the Python Package Index (default
                                https://pypi.org/simple).
  --extra-index-url TEXT        Extra URLs of package indexes to use in addition
                                to --index-url.
  --threads INTEGER             Maximum amount of threads to use for running
                                concurrent pip subprocesses.
  --pre                         Include pre-release and development versions. By
                                default, pip implicitly excludes pre-releases
                                (unless specified otherwise by PEP 440).
  -v, --verbose                 Control verbosity: -v will print cyclic
                                dependencies (WARNING), -vv will show solving
                                decisions (INFO), -vvv for development (DEBUG).
  -h, --help                    Show this message and exit.

Dependency trees

Exhaustive dependency trees without the need to install any packages (at most build some wheels).

$ pipgrip --tree pipgrip

pipgrip (0.10.6)
├── anytree>=2.4.1 (2.9.0)
│   └── six (1.16.0)
├── click>=7 (8.1.6)
├── packaging>=17 (23.1)
├── pip>=22.2 (23.2.1)
├── setuptools>=38.3 (68.0.0)
└── wheel (0.41.1)

For more details/further processing, combine --tree with --json for a detailed nested JSON dependency tree. See also --tree-ascii (no unicode tree markers), and --tree-json & --tree-json-exact (simplified JSON dependency trees).

Lockfile generation

Using the --lock option, resolved (pinned) dependencies are additionally written to ./pipgrip.lock.

$ pipgrip --tree --lock botocore==1.13.48 'boto3>=1.10,<1.10.50'

botocore==1.13.48 (1.13.48)
├── docutils<0.16,>=0.10 (0.15.2)
├── jmespath<1.0.0,>=0.7.1 (0.9.5)
├── python-dateutil<3.0.0,>=2.1 (2.8.1)
│   └── six>=1.5 (1.14.0)
└── urllib3<1.26,>=1.20 (1.25.8)
boto3<1.10.50,>=1.10 (1.10.48)
├── botocore<1.14.0,>=1.13.48 (1.13.48)
│   ├── docutils<0.16,>=0.10 (0.15.2)
│   ├── jmespath<1.0.0,>=0.7.1 (0.9.5)
│   ├── python-dateutil<3.0.0,>=2.1 (2.8.1)
│   │   └── six>=1.5 (1.14.0)
│   └── urllib3<1.26,>=1.20 (1.25.8)
├── jmespath<1.0.0,>=0.7.1 (0.9.5)
└── s3transfer<0.3.0,>=0.2.0 (0.2.1)
    └── botocore<2.0.0,>=1.12.36 (1.13.48)
        ├── docutils<0.16,>=0.10 (0.15.2)
        ├── jmespath<1.0.0,>=0.7.1 (0.9.5)
        ├── python-dateutil<3.0.0,>=2.1 (2.8.1)
        │   └── six>=1.5 (1.14.0)
        └── urllib3<1.26,>=1.20 (1.25.8)

$ cat ./pipgrip.lock

botocore==1.13.48
docutils==0.15.2
jmespath==0.9.5
python-dateutil==2.8.1
six==1.14.0
urllib3==1.25.8
boto3==1.10.48
s3transfer==0.2.1

NOTE: Since the selected botocore version is older than the one required by the recent versions of boto3, all boto3 versions will be checked for compatibility with botocore==1.13.48.

Version conflicts

If version conflicts exist for the given (ranges of) package version(s), a verbose explanation is raised.

$ pipgrip auto-sklearn~=0.6 dragnet==2.0.4

Error: Because dragnet (2.0.4) depends on scikit-learn (>=0.15.2,<0.21.0)
 and auto-sklearn (0.6.0) depends on scikit-learn (<0.22,>=0.21.0), dragnet (2.0.4) is incompatible with auto-sklearn (0.6.0).
And because no versions of auto-sklearn match >0.6.0,<1.0, dragnet (2.0.4) is incompatible with auto-sklearn (>=0.6.0,<1.0).
So, because root depends on both auto-sklearn (~=0.6) and dragnet (==2.0.4), version solving failed.

NOTE: If older versions of auto-sklearn are allowed, PubGrub will try all acceptable versions of auto-sklearn. In this case, auto-sklearn==0.5.2 requires scikit-learn (<0.20,>=0.19), making it compatible with dragnet==2.0.4.

Cyclic dependencies

If cyclic dependencies are found, it is noted in the resulting tree.

$ pipgrip --tree -v keras==2.2.2

WARNING: Cyclic dependency found: keras depends on keras-applications and vice versa.
WARNING: Cyclic dependency found: keras depends on keras-preprocessing and vice versa.
keras==2.2.2 (2.2.2)
├── h5py (2.10.0)
│   ├── numpy>=1.7 (1.18.1)
│   └── six (1.14.0)
├── keras-applications==1.0.4 (1.0.4)
│   ├── h5py (2.10.0)
│   │   ├── numpy>=1.7 (1.18.1)
│   │   └── six (1.14.0)
│   ├── keras>=2.1.6 (2.2.2, cyclic)
│   └── numpy>=1.9.1 (1.18.1)
├── keras-preprocessing==1.0.2 (1.0.2)
│   ├── keras>=2.1.6 (2.2.2, cyclic)
│   ├── numpy>=1.9.1 (1.18.1)
│   ├── scipy>=0.14 (1.4.1)
│   │   └── numpy>=1.13.3 (1.18.1)
│   └── six>=1.9.0 (1.14.0)
├── numpy>=1.9.1 (1.18.1)
├── pyyaml (5.3)
├── scipy>=0.14 (1.4.1)
│   └── numpy>=1.13.3 (1.18.1)
└── six>=1.9.0 (1.14.0)

Known caveats

PubGrub doesn't support version epochs, the main reason PyPA chose resolvelib over PubGrub for their new resolver.
Package names are canonicalised in wheel metadata, resulting in e.g. path.py -> path-py and keras_preprocessing -> keras-preprocessing in output.
VCS Support: combining VCS requirements with --editable, as well as the @ -e svn+ pattern are not supported.
Similar to setuptools' install_requires, omitting the projectname @ prefix is not supported neither for VCS requirements (like pip install git+https...), nor for PEP 440 direct references (like pip install https...).
Parsing requirements files (-r) does not support: custom file encodings, line continuations, global/per-requirement options
--reversed-tree isn't implemented yet.
Since pip install -r does not accept . as requirement, it is omitted from --lock output. So when installing local projects, either --pipe or --install should be used (the latter basically does pipgrip --lock . && pip install . --constraint ./pipgrip.lock).
Local paths are not supported (like pip install -e ../aiobotocore[boto3]), except for the current directory (like pipgrip --install -e .[boto3]).