jazzband/pip-tools

pip-tools = pip-compile + pip-sync

A set of command line tools to help you keep your pip-based packages fresh, even when you've pinned them. You do pin them, right? (In building your Python application and its dependencies for production, you want to make sure that your builds are predictable and deterministic.)

Installation

Similar to pip, pip-tools must be installed in each of your project's virtual environments:

$ source /path/to/venv/bin/activate
(venv) $ python -m pip install pip-tools

Note: all of the remaining example commands assume you've activated your project's virtual environment.

Example usage for pip-compile

The pip-compile command lets you compile a requirements.txt file from your dependencies, specified in either pyproject.toml, setup.cfg, setup.py, or requirements.in.

Run it with pip-compile or python -m piptools compile (or pipx run --spec pip-tools pip-compile if pipx was installed with the appropriate Python version). If you use multiple Python versions, you can also run py -X.Y -m piptools compile on Windows and pythonX.Y -m piptools compile on other systems.

pip-compile should be run from the same virtual environment as your project so conditional dependencies that require a specific Python version, or other environment markers, resolve relative to your project's environment.

Note: If pip-compile finds an existing requirements.txt file that fulfils the dependencies then no changes will be made, even if updates are available. To compile from scratch, first delete the existing requirements.txt file, or see Updating requirements for alternative approaches.

Requirements from pyproject.toml

The pyproject.toml file is the latest standard for configuring packages and applications, and is recommended for new projects. pip-compile supports both installing your project.dependencies as well as your project.optional-dependencies. Thanks to the fact that this is an official standard, you can use pip-compile to pin the dependencies in projects that use modern standards-adhering packaging tools like Setuptools, Hatch or flit.

Suppose you have a 'foobar' Python application that is packaged using Setuptools, and you want to pin it for production. You can declare the project metadata as:

[build-system]
requires = ["setuptools", "setuptools-scm"]
build-backend = "setuptools.build_meta"

[project]
requires-python = ">=3.9"
name = "foobar"
dynamic = ["dependencies", "optional-dependencies"]

[tool.setuptools.dynamic]
dependencies = { file = ["requirements.in"] }
optional-dependencies.test = { file = ["requirements-test.txt"] }

If you have a Django application that is packaged using Hatch, and you want to pin it for production. You also want to pin your development tools in a separate pin file. You declare django as a dependency and create an optional dependency dev that includes pytest:

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[project]
name = "my-cool-django-app"
version = "42"
dependencies = ["django"]

[project.optional-dependencies]
dev = ["pytest"]

You can produce your pin files as easily as:

$ pip-compile -o requirements.txt pyproject.toml
#
# This file is autogenerated by pip-compile with Python 3.10
# by the following command:
#
#    pip-compile --output-file=requirements.txt pyproject.toml
#
asgiref==3.6.0
    # via django
django==4.1.7
    # via my-cool-django-app (pyproject.toml)
sqlparse==0.4.3
    # via django

$ pip-compile --extra dev -o dev-requirements.txt pyproject.toml
#
# This file is autogenerated by pip-compile with Python 3.10
# by the following command:
#
#    pip-compile --extra=dev --output-file=dev-requirements.txt pyproject.toml
#
asgiref==3.6.0
    # via django
attrs==22.2.0
    # via pytest
django==4.1.7
    # via my-cool-django-app (pyproject.toml)
exceptiongroup==1.1.1
    # via pytest
iniconfig==2.0.0
    # via pytest
packaging==23.0
    # via pytest
pluggy==1.0.0
    # via pytest
pytest==7.2.2
    # via my-cool-django-app (pyproject.toml)
sqlparse==0.4.3
    # via django
tomli==2.0.1
    # via pytest

This is great for both pinning your applications, but also to keep the CI of your open-source Python package stable.

Requirements from setup.py and setup.cfg

pip-compile has also full support for setup.py- and setup.cfg-based projects that use setuptools.

Just define your dependencies and extras as usual and run pip-compile as above.

Requirements from requirements.in

You can also use plain text files for your requirements (e.g. if you don't want your application to be a package). To use a requirements.in file to declare the Django dependency:

# requirements.in
django

Now, run pip-compile requirements.in:

$ pip-compile requirements.in
#
# This file is autogenerated by pip-compile with Python 3.10
# by the following command:
#
#    pip-compile requirements.in
#
asgiref==3.6.0
    # via django
django==4.1.7
    # via -r requirements.in
sqlparse==0.4.3
    # via django

And it will produce your requirements.txt, with all the Django dependencies (and all underlying dependencies) pinned.

(updating-requirements)=

Updating requirements

pip-compile generates a requirements.txt file using the latest versions that fulfil the dependencies you specify in the supported files.

If pip-compile finds an existing requirements.txt file that fulfils the dependencies then no changes will be made, even if updates are available.

To force pip-compile to update all packages in an existing requirements.txt, run pip-compile --upgrade.

To update a specific package to the latest or a specific version use the --upgrade-package or -P flag:

# only update the django package
$ pip-compile --upgrade-package django

# update both the django and requests packages
$ pip-compile --upgrade-package django --upgrade-package requests

# update the django package to the latest, and requests to v2.0.0
$ pip-compile --upgrade-package django --upgrade-package requests==2.0.0

You can combine --upgrade and --upgrade-package in one command, to provide constraints on the allowed upgrades. For example to upgrade all packages whilst constraining requests to the latest version less than 3.0:

$ pip-compile --upgrade --upgrade-package 'requests<3.0'

Understanding output file behavior

When you run pip-compile, it reads any existing output file (e.g., requirements.txt) and uses the versions specified there as constraints. This ensures that the output is stable -- pip-compile will prefer to use existing pinned versions.

When upgrading packages or adding new packages to the input, it is possible for the existing pins to conflict with new packages. In such cases, pip-compile will update the pins and emit a warning (e.g., Discarding foo=1.2.3 to proceed).

In order to resolve dependencies afresh, pip-compile provides the following options:

• --upgrade / -U flag to upgrade all packages
• --upgrade-package <package>/-P <package> to upgrade specific packages

Using hashes

If you would like to use Hash-Checking Mode available in pip since version 8.0, pip-compile offers --generate-hashes flag:

$ pip-compile --generate-hashes requirements.in
#
# This file is autogenerated by pip-compile with Python 3.10
# by the following command:
#
#    pip-compile --generate-hashes requirements.in
#
asgiref==3.6.0 \
    --hash=sha256:71e68008da809b957b7ee4b43dbccff33d1b23519fb8344e33f049897077afac \
    --hash=sha256:9567dfe7bd8d3c8c892227827c41cce860b368104c3431da67a0c5a65a949506
    # via django
django==4.1.7 \
    --hash=sha256:44f714b81c5f190d9d2ddad01a532fe502fa01c4cb8faf1d081f4264ed15dcd8 \
    --hash=sha256:f2f431e75adc40039ace496ad3b9f17227022e8b11566f4b363da44c7e44761e
    # via -r requirements.in
sqlparse==0.4.3 \
    --hash=sha256:0323c0ec29cd52bceabc1b4d9d579e311f3e4961b98d174201d5622a23b85e34 \
    --hash=sha256:69ca804846bb114d2ec380e4360a8a340db83f0ccf3afceeb1404df028f57268
    # via django

Output File

To output the pinned requirements in a filename other than requirements.txt, use --output-file. This might be useful for compiling multiple files, for example with different constraints on django to test a library with both versions using tox:

$ pip-compile --upgrade-package 'django<1.0' --output-file requirements-django0x.txt
$ pip-compile --upgrade-package 'django<2.0' --output-file requirements-django1x.txt

Or to output to standard output, use --output-file=-:

$ pip-compile --output-file=- > requirements.txt
$ pip-compile - --output-file=- < requirements.in > requirements.txt

Forwarding options to pip

Any valid pip flags or arguments may be passed on with pip-compile's --pip-args option, e.g.

$ pip-compile requirements.in --pip-args "--retries 10 --timeout 30"

Workflow for layered requirements

If you have different environments that you need to install different but compatible packages for, then you can create layered requirements files and use one layer to constrain the other.

For example, if you have a Django project where you want the newest 2.1 release in production and when developing you want to use the Django debug toolbar, then you can create two *.in files, one for each layer:

# requirements.in
django<2.2

At the top of the development requirements dev-requirements.in you use -c requirements.txt to constrain the dev requirements to packages already selected for production in requirements.txt.

# dev-requirements.in
-c requirements.txt
django-debug-toolbar<2.2

First, compile requirements.txt as usual:

$ pip-compile
#
# This file is autogenerated by pip-compile with Python 3.10
# by the following command:
#
#    pip-compile
#
django==2.1.15
    # via -r requirements.in
pytz==2023.3
    # via django

Now compile the dev requirements and the requirements.txt file is used as a constraint:

$ pip-compile dev-requirements.in
#
# This file is autogenerated by pip-compile with Python 3.10
# by the following command:
#
#    pip-compile dev-requirements.in
#
django==2.1.15
    # via
    #   -c requirements.txt
    #   django-debug-toolbar
django-debug-toolbar==2.1
    # via -r dev-requirements.in
pytz==2023.3
    # via
    #   -c requirements.txt
    #   django
sqlparse==0.4.3
    # via django-debug-toolbar

As you can see above, even though a 2.2 release of Django is available, the dev requirements only include a 2.1 version of Django because they were constrained. Now both compiled requirements files can be installed safely in the dev environment.

To install requirements in production stage use:

$ pip-sync

You can install requirements in development stage by:

$ pip-sync requirements.txt dev-requirements.txt

Version control integration

You might use pip-compile as a hook for the pre-commit. See pre-commit docs for instructions. Sample .pre-commit-config.yaml:

repos:
  - repo: https://github.com/jazzband/pip-tools
    rev: 7.4.1
    hooks:
      - id: pip-compile

You might want to customize pip-compile args by configuring args and/or files, for example:

repos:
  - repo: https://github.com/jazzband/pip-tools
    rev: 7.4.1
    hooks:
      - id: pip-compile
        files: ^requirements/production\.(in|txt)$
        args: [--index-url=https://example.com, requirements/production.in]

If you have multiple requirement files make sure you create a hook for each file.

repos:
  - repo: https://github.com/jazzband/pip-tools
    rev: 7.4.1
    hooks:
      - id: pip-compile
        name: pip-compile setup.py
        files: ^(setup\.py|requirements\.txt)$
      - id: pip-compile
        name: pip-compile requirements-dev.in
        args: [requirements-dev.in]
        files: ^requirements-dev\.(in|txt)$
      - id: pip-compile
        name: pip-compile requirements-lint.in
        args: [requirements-lint.in]
        files: ^requirements-lint\.(in|txt)$
      - id: pip-compile
        name: pip-compile requirements.in
        args: [requirements.in]
        files: ^requirements\.(in|txt)$

Example usage for pip-sync

Now that you have a requirements.txt, you can use pip-sync to update your virtual environment to reflect exactly what's in there. This will install/upgrade/uninstall everything necessary to match the requirements.txt contents.

Run it with pip-sync or python -m piptools sync. If you use multiple Python versions, you can also run py -X.Y -m piptools sync on Windows and pythonX.Y -m piptools sync on other systems.

pip-sync must be installed into and run from the same virtual environment as your project to identify which packages to install or upgrade.

Be careful: pip-sync is meant to be used only with a requirements.txt generated by pip-compile.

$ pip-sync
Uninstalling flake8-2.4.1:
    Successfully uninstalled flake8-2.4.1
Collecting click==4.1
    Downloading click-4.1-py2.py3-none-any.whl (62kB)
    100% |................................| 65kB 1.8MB/s
    Found existing installation: click 4.0
    Uninstalling click-4.0:
        Successfully uninstalled click-4.0
Successfully installed click-4.1

To sync multiple *.txt dependency lists, just pass them in via command line arguments, e.g.

$ pip-sync dev-requirements.txt requirements.txt

Passing in empty arguments would cause it to default to requirements.txt.

Any valid pip install flags or arguments may be passed with pip-sync's --pip-args option, e.g.

$ pip-sync requirements.txt --pip-args "--no-cache-dir --no-deps"

Note: pip-sync will not upgrade or uninstall packaging tools like setuptools, pip, or pip-tools itself. Use python -m pip install --upgrade to upgrade those packages.

Should I commit requirements.in and requirements.txt to source control?

Generally, yes. If you want a reproducible environment installation available from your source control, then yes, you should commit both requirements.in and requirements.txt to source control.

Note that if you are deploying on multiple Python environments (read the section below), then you must commit a separate output file for each Python environment. We suggest to use the {env}-requirements.txt format (ex: win32-py3.7-requirements.txt, macos-py3.10-requirements.txt, etc.).

Cross-environment usage of requirements.in/requirements.txt and pip-compile

The dependencies of a package can change depending on the Python environment in which it is installed. Here, we define a Python environment as the combination of Operating System, Python version (3.7, 3.8, etc.), and Python implementation (CPython, PyPy, etc.). For an exact definition, refer to the possible combinations of PEP 508 environment markers.

As the resulting requirements.txt can differ for each environment, users must execute pip-compile on each Python environment separately to generate a requirements.txt valid for each said environment. The same requirements.in can be used as the source file for all environments, using PEP 508 environment markers as needed, the same way it would be done for regular pip cross-environment usage.

If the generated requirements.txt remains exactly the same for all Python environments, then it can be used across Python environments safely. But users should be careful as any package update can introduce environment-dependent dependencies, making any newly generated requirements.txt environment-dependent too. As a general rule, it's advised that users should still always execute pip-compile on each targeted Python environment to avoid issues.

Maximizing reproducibility

pip-tools is a great tool to improve the reproducibility of builds. But there are a few things to keep in mind.

• pip-compile will produce different results in different environments as described in the previous section.
• pip must be used with the PIP_CONSTRAINT environment variable to lock dependencies in build environments as documented in #8439.
• Dependencies come from many sources.

Continuing the pyproject.toml example from earlier, creating a single lock file could be done like:

$ pip-compile --all-build-deps --all-extras --output-file=constraints.txt --strip-extras pyproject.toml
#
# This file is autogenerated by pip-compile with Python 3.9
# by the following command:
#
#    pip-compile --all-build-deps --all-extras --output-file=constraints.txt --strip-extras pyproject.toml
#
asgiref==3.5.2
    # via django
attrs==22.1.0
    # via pytest
backports-zoneinfo==0.2.1
    # via django
django==4.1
    # via my-cool-django-app (pyproject.toml)
editables==0.3
    # via hatchling
hatchling==1.11.1
    # via my-cool-django-app (pyproject.toml::build-system.requires)
iniconfig==1.1.1
    # via pytest
packaging==21.3
    # via
    #   hatchling
    #   pytest
pathspec==0.10.2
    # via hatchling
pluggy==1.0.0
    # via
    #   hatchling
    #   pytest
py==1.11.0
    # via pytest
pyparsing==3.0.9
    # via packaging
pytest==7.1.2
    # via my-cool-django-app (pyproject.toml)
sqlparse==0.4.2
    # via django
tomli==2.0.1
    # via
    #   hatchling
    #   pytest

Some build backends may also request build dependencies dynamically using the get_requires_for_build_ hooks described in PEP 517 and PEP 660. This will be indicated in the output with one of the following suffixes:

• (pyproject.toml::build-system.backend::editable)
• (pyproject.toml::build-system.backend::sdist)
• (pyproject.toml::build-system.backend::wheel)

Other useful tools

• pip-compile-multi - pip-compile command wrapper for multiple cross-referencing requirements files.
• pipdeptree to print the dependency tree of the installed packages.
• requirements.in/requirements.txt syntax highlighting:
  • requirements.txt.vim for Vim.
  • Python extension for VS Code.
  • pip-requirements.el for Emacs.

Deprecations

This section lists pip-tools features that are currently deprecated.

• In the next major release, the --allow-unsafe behavior will be enabled by default (https://github.com/jazzband/pip-tools/issues/989). Use --no-allow-unsafe to keep the old behavior. It is recommended to pass --allow-unsafe now to adapt to the upcoming change.
• The legacy resolver is deprecated and will be removed in future versions. The new default is --resolver=backtracking.
• In the next major release, the --strip-extras behavior will be enabled by default (https://github.com/jazzband/pip-tools/issues/1613). Use --no-strip-extras to keep the old behavior.

A Note on Resolvers

You can choose from either default backtracking resolver or the deprecated legacy resolver.

The legacy resolver will occasionally fail to resolve dependencies. The backtracking resolver is more robust, but can take longer to run in general.

You can continue using the legacy resolver with --resolver=legacy although note that it is deprecated and will be removed in a future release.