Skip to content

Working with Virtual Environments#

When you run pdm init command, PDM will ask for the Python interpreter to use in the project, which is the base interpreter to install dependencies and run tasks.

Compared to PEP 582, virtual environments are considered more mature and have better support in the Python ecosystem as well as IDEs. Therefore, virtualenv is the default mode if not configured otherwise.

Virtual environments will be used if the project interpreter (the interpreter stored in .pdm-python, which can be checked by pdm info) is from a virtualenv.

Virtualenv auto-creation#

By default, PDM prefers to use the virtualenv layout as other package managers do. When you run pdm install the first time on a new PDM-managed project, whose Python interpreter is not decided yet, PDM will create a virtualenv in <project_root>/.venv, and install dependencies into it. In the interactive session of pdm init, PDM will also ask to create a virtualenv for you.

You can choose the backend used by PDM to create a virtualenv. Currently it supports three backends:

You can change it by pdm config venv.backend [virtualenv|venv|conda].

Added in version 2.13.0

Moreover, when python.use_venv config is set to true, PDM will always try to create a virtualenv when using pdm use to switch the Python interpreter.

Create a virtualenv yourself#

You can create more than one virtualenvs with whatever Python version you want.

1
2
3
4
5
6
# Create a virtualenv based on 3.8 interpreter
$ pdm venv create 3.8
# Assign a different name other than the version string
$ pdm venv create --name for-test 3.8
# Use venv as the backend to create, support 3 backends: virtualenv(default), venv, conda
$ pdm venv create --with venv 3.9

The location of virtualenvs#

If no --name is given, PDM will create the venv in <project_root>/.venv. Otherwise, virtualenvs go to the location specified by the venv.location configuration. They are named as <project_name>-<path_hash>-<name_or_python_version> to avoid name collision. You can disable the in-project virtualenv creation by pdm config venv.in_project false. And all virtualenvs will be created under venv.location.

Reuse the virtualenv you created elsewhere#

You can tell PDM to use a virtualenv you created in preceding steps, with pdm use:

1
pdm use -f /path/to/venv

Virtualenv auto-detection#

When no interpreter is stored in the project config or PDM_IGNORE_SAVED_PYTHON env var is set, PDM will try to detect possible virtualenvs to use:

  • venv, env, .venv directories in the project root
  • The currently activated virtualenv, unless PDM_IGNORE_ACTIVE_VENV is set

List all virtualenvs created with this project#

1
2
3
4
5
6
$ pdm venv list
Virtualenvs created with this project:

-  3.8.6: C:\Users\Frost Ming\AppData\Local\pdm\pdm\venvs\test-project-8Sgn_62n-3.8.6
-  for-test: C:\Users\Frost Ming\AppData\Local\pdm\pdm\venvs\test-project-8Sgn_62n-for-test
-  3.9.1: C:\Users\Frost Ming\AppData\Local\pdm\pdm\venvs\test-project-8Sgn_62n-3.9.1

Show the path or python interpreter of a virtualenv#

1
2
$ pdm venv --path for-test
$ pdm venv --python for-test

Remove a virtualenv#

1
2
3
4
$ pdm venv remove for-test
Virtualenvs created with this project:
Will remove: C:\Users\Frost Ming\AppData\Local\pdm\pdm\venvs\test-project-8Sgn_62n-for-test, continue? [y/N]:y
Removed C:\Users\Frost Ming\AppData\Local\pdm\pdm\venvs\test-project-8Sgn_62n-for-test

Activate a virtualenv#

Instead of spawning a subshell like what pipenv and poetry do, pdm venv doesn't create the shell for you but print the activate command to the console. In this way you won't leave the current shell. You can then feed the output to eval to activate the virtualenv:

1
2
$ eval $(pdm venv activate for-test)
(test-project-for-test) $  # Virtualenv entered

1
$ eval (pdm venv activate for-test)

1
PS1> Invoke-Expression (pdm venv activate for-test)

Additionally, if the project interpreter is a venv Python, you can omit the name argument following activate.

Note

venv activate does not switch the Python interpreter used by the project. It only changes the shell by injecting the virtualenv paths to environment variables. For the forementioned purpose, use the pdm use command.

For more CLI usage, see the pdm venv documentation.

Looking for pdm shell?

PDM doesn't provide a shell command because many fancy shell functions may not work perfectly in a subshell, which brings a maintenance burden to support all the corner cases. However, you can still gain the ability via the following ways:

  • Use pdm run $SHELL, this will spawn a subshell with the environment variables set properly. The subshell can be quit with exit or Ctrl+D.
  • Add a shell function to activate the virtualenv, here is an example of BASH function that also works on ZSH:
1
2
3
4
5
6
7
8
9
pdm() {
  local command=$1

  if [[ "$command" == "shell" ]]; then
      eval $(pdm venv activate)
  else
      command pdm $@
  fi
}

Copy and paste this function to your ~/.bashrc file and restart your shell.

For fish shell you can put the following into your ~/fish/config.fish or in ~/.config/fish/config.fish

1
2
3
4
5
6
7
8
9
  function pdm
      set cmd $argv[1]

      if test "$cmd" = "shell"
          eval (pdm venv activate)
      else
          command pdm $argv
      end
  end

Now you can run pdm shell to activate the virtualenv. The virtualenv can be deactivated with deactivate command as usual.

Prompt customization#

By default when you activate a virtualenv, the prompt will show: {project_name}-{python_version}.

For example if your project is named test-project:

1
2
$ eval $(pdm venv activate for-test)
(test-project-3.10) $  # {project_name} == test-project and {python_version} == 3.10

The format can be customized before virtualenv creation with the venv.prompt configuration or PDM_VENV_PROMPT environment variable (before a pdm init or pdm venv create). Available variables are:

  • project_name: name of your project
  • python_version: version of Python (used by the virtualenv)
1
2
3
$ PDM_VENV_PROMPT='{project_name}-py{python_version}' pdm venv create --name test-prompt
$ eval $(pdm venv activate test-prompt)
(test-project-py3.10) $

Run a command in a virtual environment without activating it#

1
2
3
4
5
6
# Run a script
$ pdm run --venv test test
# Install packages
$ pdm sync --venv test
# List the packages installed
$ pdm list --venv test

There are other commands supporting --venv flag or PDM_IN_VENV environment variable, see the CLI reference. You should create the virtualenv with pdm venv create --name <name> before using this feature.

Switch to a virtualenv as the project environment#

By default, if you use pdm use and select a non-venv Python, the project will be switched to PEP 582 mode. We also allow you to switch to a named virtual environment via the --venv flag:

1
2
3
4
# Switch to a virtualenv named test
$ pdm use --venv test
# Switch to the in-project venv located at $PROJECT_ROOT/.venv
$ pdm use --venv in-project

Disable virtualenv mode#

You can disable the auto-creation and auto-detection for virtualenv by pdm config python.use_venv false. If venv is disabled, PEP 582 mode will always be used even if the selected interpreter is from a virtualenv.

Including pip in your virtual environment#

By default PDM will not include pip in virtual environments. This increases isolation by ensuring that only your dependencies are installed in the virtual environment.

To install pip once (if for example you want to install arbitrary dependencies in CI) you can run:

1
2
3
4
5
# Install pip in the virtual environment
$ pdm run python -m ensurepip
# Install arbitrary dependencies
# These dependencies are not checked for conflicts against lockfile dependencies!
$ pdm run python -m pip install coverage

Or you can create the virtual environment with --with-pip:

1
$ pdm venv create --with-pip 3.9

See the ensurepip docs for more details on ensurepip.

If you want to permanently configure PDM to include pip in virtual environments you can use the venv.with_pip configuration.