User Guide#

Initializing a new project#

The purpose of conda project init is to provide a command like conda create that creates the environment.yml, conda-project.yml and conda-lock.default.yml files before installing the environment. The --install flag can be used to build the files and then install the environment.

The init command will always write channels and platforms into the environment.yml file.

From within an existing project directory, run:

conda project init

Packages can also be specified when creating a project:

conda project init python=3.10

To specify pip dependencies on initialization, use the @pip:: prefix:

conda project init python=3.10 @pip::numpy

This will initialize your project with a new conda-project.yml, environment.yml, and local .condarc file.

CLI help#

The following is the output of conda project init --help:

usage: conda-project init [-h] [--directory PROJECT_DIR] [-n NAME] [-c CHANNEL] [--platforms PLATFORMS] [--conda-configs CONDA_CONFIGS]
                            [--lock] [--install]
                            [dependencies [dependencies ...]]

Initialize a new project

positional arguments:
  dependencies          Packages to add to the environment.yml in MatchSpec format.

optional arguments:
  -h, --help            show this help message and exit
  --directory PROJECT_DIR
                        Project directory (defaults to current directory)
  -n NAME, --name NAME  Name for the project.
  -c CHANNEL, --channel CHANNEL
                        Additional channel to search for packages. The default channel is 'defaults'. Multiple channels are added with repeated
                        use of this argument.
  --platforms PLATFORMS
                        Comma separated list of platforms for which to lock dependencies. The default is win-64,linux-64,osx-64,osx-arm64
  --conda-configs CONDA_CONFIGS
                        Comma separated list of conda configuration parameters to write into the .condarc file in the project directory. The
                        format for each config is key=value. For example --conda-configs experimental_solver=libmamba,channel_priority=strict
  --lock                Create the conda-lock.<env>.yml file(s)
  --install             Create the local conda environment for the current platform.

If I already have an environment.yml#

If the user writes a minimal environment.yml file as shown below, conda-project will make two assumptions:

  1. all packages come from the defaults channel, and

  2. the dependencies will be locked for win-64, linux-64, mac-64 and your current platform if it is not one of those three.

The conda-project.yml file#

An optional conda-project.yml file is defined that supports multiple conda environments per project and each environment can be built from multiple conda environment YAML sources, which uses conda-lock compound specification.

For example:

name: project-name

environments:
  main:
    - environment.yml
  development:
    - environment.yml
    - ../dev-extras.yml

variables: {}
commands: {}

Note

Environment YAML files are specified as relative to the location of the conda-project.yml file. Each key in environments: can be utilized in conda project lock <env-name> or conda project install <env-name>. These commands also accept --all to lock and prepare each of the defined environments.

If no env name is supplied lock, prepare, and clean assume the default environment is the first environment listed in the conda-project.yml file.

Locking dependencies#

conda-project uses conda-lock to lock environment dependencies. To manually lock your environments, run:

conda project lock

Locking of dependencies completely ignores the user’s channel settings in ~/.condarc and will only use channels supplied by the environment.yml file.

conda project lock utilizes the --check-input-hash feature of conda-lock. When you run conda project lock multiple times, the lock file will only be updated if the environment.yml has changed. To force a re-lock use conda project lock --force.

Installing your environments#

conda project install enforces the use of conda-lock. If a conda-lock.<env>.yml file is not present it will be created by install with the above assumptions if necessary. If a conda-lock.<env>.yml file is found but the locked platforms do not match your current platform it will raise an exception.

The live conda environment is built from a rendered lockfile (explicit type) for your current platform, similar to how conda lock install works.

Adding packages to an environment#

The conda project add command works similar to conda install to add packages. Like init you can specify pip packages with the @pip:: prefix. The add command will re-lock and install your environment each time it is run.

For example:

conda project init
conda project add -c defaults python=3.10
conda project add conda-forge::pandas requests @pip::pydantic
conda project add "pandas<2"

Note that in the above commands pandas was added twice. Adding a package that already exists in your environment.yml file will replace that entry with the new one.

To add packages to an environment other than the first one in the conda-project.yml file use the --environment <name> flag.

Here’s the full help for the add command.

usage: conda-project add [-h] [--directory PROJECT_DIR] [--project-archive PROJECT_ARCHIVE_FILE_OR_URL]
                         [--archive-storage-options ARCHIVE_STORAGE_OPTIONS] [--environment ENVIRONMENT] [-c CHANNEL]
                         [PACKAGE_SPECIFICATION [PACKAGE_SPECIFICATION ...]]

Add packages to an environment

positional arguments:
  PACKAGE_SPECIFICATION
                        Packages to add to the environment.yml. The format for each package is '[<prefix>::]<name>[<op><version>]'
                        where <op> can be =, <, >, <=, or >=.Most commonly `<prefix>::` declares the conda channel from which to
                        install packages. Use the prefix `@pip::` to add pip package dependencies with support for full pip package
                        specification syntax.

optional arguments:
  -h, --help            show this help message and exit
  --directory PROJECT_DIR
                        Project directory (defaults to current directory)
  --project-archive PROJECT_ARCHIVE_FILE_OR_URL
                        EXPERIMENTAL: Extract and run directly from a project archive. The archive can be a local file or a fsspec
                        compatible URL. You may need to install appropriate driver packages to work with remote archives. Optionally,
                        use --directory to set the destination directory of the extracted project.
  --archive-storage-options ARCHIVE_STORAGE_OPTIONS
                        EXPERIMENTAL: Comma separated list of fsspec storage_options for accessing a remote archive For example
                        --archive-storage-options username=<user>,password=<pass>
  --environment ENVIRONMENT
  -c CHANNEL, --channel CHANNEL
                        Additional channel to search for packages. The default channel is 'defaults'. Multiple channels are added with
                        repeated use of this argument.

Removing a package from an environment#

The inverse of add is conda project remove. Removing a package will also re-lock and re-install the environment. Only the name of the package is required to remove it and you can remove a pip package with the @pip:: prefix.

Start from where we left of in the previous section we initialized the project and added packages.

conda project init
conda project add -c defaults python=3.10
conda project add "conda-forge::pandas<2" requests @pip::pydantic

In the end our environment.yml now looks like:

name:
channels:
  - defaults
dependencies:
  - python=3.10
    - conda-forge::pandas<2
    - requests
    - pip
    - pip:
      - pydantic
variables:
prefix:
platforms:
  - osx-arm64
  - linux-64
  - osx-64
  - win-64

To remove packages we need only specify the name of package

conda project remove pandas @pip::pydantic

Then we are left with:

name:
channels:
  - defaults
dependencies:
  - python=3.10
  - requests
  - pip
  - pip: []
variables:
prefix:
platforms:
  - osx-arm64
  - linux-64
  - osx-64
  - win-64

Here’s the help output for the remove command:

usage: conda-project remove [-h] [--directory PROJECT_DIR]
                            [--project-archive PROJECT_ARCHIVE_FILE_OR_URL]
                            [--archive-storage-options ARCHIVE_STORAGE_OPTIONS]
                            [--environment ENVIRONMENT]
                            [PACKAGE_SPECIFICATION [PACKAGE_SPECIFICATION ...]]

Remove packages to an environment

positional arguments:
  PACKAGE_SPECIFICATION
                        Packages to remove from the environment.yml. Only the
                        name of the package is required here. To remove a pip
                        package use the pypyi:: prefix.

optional arguments:
  -h, --help            show this help message and exit
  --directory PROJECT_DIR
                        Project directory (defaults to current directory)
  --project-archive PROJECT_ARCHIVE_FILE_OR_URL
                        EXPERIMENTAL: Extract and run directly from a project
                        archive. The archive can be a local file or a fsspec
                        compatible URL. You may need to install appropriate
                        driver packages to work with remote archives.
                        Optionally, use --directory to set the destination
                        directory of the extracted project.
  --archive-storage-options ARCHIVE_STORAGE_OPTIONS
                        EXPERIMENTAL: Comma separated list of fsspec
                        storage_options for accessing a remote archive For
                        example --archive-storage-options
                        username=<user>,password=<pass>
  --environment ENVIRONMENT

Activating environments in the shell#

conda project activate [environment] will launch a shell and activate the named conda environment. If no environment name is supplied the first environment is activated. The activate command will force updating the lock and install the environment if it has not already been completed. Unlike conda activate, which is capable of adjusting your current shell process, conda project activate starts a new shell so it preferable to exit the shell back to the parent rather than running conda deactivate.

Minimal example#

❯ conda project init python=3.8
Locking dependencies for default: done
Locked dependencies for win-64, osx-64, osx-arm64, linux-64 platforms
Project created at /Users/adefusco/Development/conda-incubator/conda-project/examples/new-project

❯ tree -a ./
./
├── .condarc
├── conda-project.yml
├── conda-lock.default.yml
└── environment.yml

❯ cat environment.yml
name: new-project
channels:
  - defaults
dependencies:
  - python=3.8
platforms:
  - win-64
  - osx-64
  - osx-arm64
  - linux-64

❯ cat conda-project.yml
name: new-project
environments:
  default:
    - environment.yml
variables: {}
commands: {}

Environment variables#

Conda Project supports defining environment variables in the conda-project.yml file that will be set upon environment activation or when running commands. Variables are defined in the variables: key and can have an optional default value. Starting from the minimal example above I’ll add a variable FOO with the default value has-default-value and the variable BAR with no default value.

name: new-project
environments:
  default:
    - environment.yml

variables:
  FOO: has-default-value
  BAR:

commands: {}

The values of the variables can be set or overridden by the use of a .env file (see the python-dotenv documentation for more details) or by setting the variable in your shell.

In this example the BAR environment variable is unset so conda project activate will not start unless a value is provided by the shell or in a .env file.

❯ conda project activate
CondaProjectError: The following variables do not have a default value and values
were not provided in the .env file or set on the command line when executing 'conda project run':
BAR

On Unix you can do the following

❯ BAR=set-on-cli conda project activate
## Project environment default activated in a new shell.
## Exit this shell to de-activate.
❯ conda activate /Users/adefusco/Development/conda-incubator/conda-project/examples/p/envs/default

On Windows in either cmd.exe you can use set variable=value

> set BAR=set-in-shell
> conda project activate

and finally, in Powershell you would use $env

> $env:BAR = 'set-in-shell'
> conda project activate

Defining and running commands#

Conda Project supports running commands as if the desired environment were activated, similar to conda run.

In Conda Project the run command can be used to execute ad-hoc commands or those defined in the conda-project.yml file. Note that conda project run should not be utilized after conda project activate. The run command will not work from within an activated environment.

Let’s start with a simple python script called vars.py, that prints environment variables and optionally takes an argument --version to also print the Python version.

import os
import sys

if len(sys.argv) > 1 and sys.argv[1] == '--version':
    print("0.0.1")
    sys.exit(1)
else:
    print(f"The value of FOO is {os.environ.get('FOO')}")
    print(f"The value of BAR is {os.environ.get('BAR')}")
    print(f"The value of BAZ is {os.environ.get('BAZ')}")

Note that when the --version flag is provided the return code for this script is 1, meaning a failure.

We can execute this script with the default environment as an ad-hoc command using the Python interpreter provided by the default environment. Here a .env file is utilized to provide the value of the BAR variable.

> BAR=set-on-cli conda project run python vars.py
The value of FOO is has-default-value
The value of BAR is set-on-cli
The value of BAZ is None

On Linux, Mac, and Windows the return code of the conda project run is set to the return code of the command it is executing. For example the above invocation returns 0, known as a successful execution. The --version flag, however exits with code 1, typically meaning a failed execution.

> BAR=set-on-cli conda project run python vars.py --version
0.0.1
> echo $?
1

On Windows cmd.exe we can echo the `%ERRORLEVEL% variable.

> set BAR=set-on-cli
> conda project run python vars.py --version
> echo %ERRORLEVEL%
1

And in Powershell we print the $LASTEXITCODE variable

> $env:BAR = 'set-on-cli'
> conda project run python vars.py --version
0.0.1
> $LASTEXITCODE
1

Defined commands in the conda-project.yml are placed under the commands: key. Commands written in one line are set to execute over the default (first) environment.

commands:
  print-vars: python vars.py

Now you can use conda project run without arguments and it will execute the first named command in the conda-project.yml file.

> BAR=set-on-cli conda project run
The value of FOO is has-default-value
The value of BAR is set-on-cli
The value of BAZ is None

Defined commands also support extra arguments, but the name of the command must be supplied.

> BAR=set-on-cli conda project run print-vars --version
0.0.1

Here is an example of a full specified named command that declares the environment from which it will run, and variables. Command variables can override project-level variables or define new variables. Note that even though command-level-variables override project-level variables the final value of the variable can still be override by the .env file and that can be overridden by the shell.

commands:
  print-vars:
    cmd: python vars.py
    environment: default
    variables:
      BAR: set-in-cmd
      BAZ: a-new-var

Here the command is run by name without any extra variables. This is the same as conda project run

> conda project run print-vars
The value of FOO is bar
The value of BAR is set-on-cli
The value of BAZ is a-new-var

Python API#

The Python API provides full support for the above workflows by creating a CondaProject object. CondaProject takes a single optional argument to supply the path to the project. The default value is the current working directory, . Every CondaProject has at least one conda environment.

A project directory containing only an environment.yml file will create a single environment of the name default, which can be locked or installed. If multiple environments are defined in a conda-project.yml the .environments attribute provides dictionary-style syntax for each named environment. The first defined conda environment in the conda-project.yml is accessible as project.default_environment

from conda_project import CondaProject

project = CondaProject()
project.default_environment.lock()
prefix = project.default_environment.install()

## alternative, use the name 'default'

project.environments['default'].lock()
prefix = project.environments['default'].install()

To create a new project directory the CondaProject.init() method follows the CLI arguments described above. Projects are automatically locked. See the docstring for .init() for more details.

from conda_project import CondaProject

project = CondaProject.init(
   directory='new-project',
   dependencies=['python=3.8'],
)