Conda concepts

conda is a Python package and environment manager, used widely in the Python data science ecosystem. conda-store build on conda and other supporting libraries in the conda community. This page briefly covers some key conda concepts necessary to use conda-store. For detailed explanations, check out the conda documentation.

Python package

Open source software projects (sometimes called libraries) are shared with users as packages. You need to "install" the package on your local workspace to use it.

[pip][pip-docs] and conda are popular package management tools in the Python ecosystem.

pip ships with the Python programming language, and can install packages from the PyPI (Python Package Index) - a community managed collection of packages, public/private PyPI mirrors, GitHub sources, and local directories.

conda needs to be downloaded separately (through a distribution like Anaconda or Miniconda), and can install packages from conda channels and local builds.

Some Python packages depend on non-Python code (for example, NumPy includes some C libraries). Installing such packages from PyPI using pip can be un-reliable and sometimes it can be your responsibility to separately install the non-Python libraries. However, conda provides a package management solution that includes both Python and other underlying non-Python code.

Dependencies

Modern open source software (and software in general) is created using or builds on other libraries, which are called the dependencies of the project. For example, pandas uses NumPy's ndarrays and is written partially in Python, hence, NumPy and Python are dependencies of pandas. Specifically, they are the direct dependencies. The dependencies of NumPy and pandas, and the dependencies of those dependencies, and so on creates a complete dependency graph for pandas.

Since conda-store focuses on environments, the terms dependencies usually refers to the full set of compatible dependencies for all the packages specified in an environment.

Channels (conda)

The conda documentation defines:

Conda channels are the locations where packages are stored. They serve as the base for hosting and managing packages. Conda packages are downloaded from remote channels, which are URLs to directories containing conda packages.

Similar to PyPI, conda channels are URLs of remote servers that manage packages.

In conda-store, packages are installed from the conda-forge channel by default. conda-forge is a community maintained channel for hosting open source libraries.

note

This behavior is different from conda downloaded from Anaconda/Miniconda distribution, that gets packages from the "default" channel by default.

Other distributions like Miniforge also use conda-forge as the default channel.

Environments

conda-store helps create and manage "conda environments", sometimes also referred to as "data science environments" or simply "environments" in conda-store spaces.

An environment is an isolated set of installed packages. The official conda documentation states:

A conda environment is a directory that contains a specific collection of conda packages that you have installed.

If you change one environment, your other environments are not affected. You can easily activate or deactivate environments, which is how you switch between them.

In data science and development workflows, you often use different environments for different projects and sub-projects. It gives you a clean space for development with only the packages and versions that you need for the specific project. You can also use different versions of the same package in different environments depending on your project needs.

Using isolated environments is a good practice to follow. The alternative, where requirements for all projects are added to a single "base" environment can not only give you un-reliable results but also be very tedious to manage across projects.

Environment specification (spec)

conda environments are specified through a YAML file, which is called the environment specification and has the following major components:

name: my-cool-env # name of your environment
channels:   # conda channels to get packages from, in order of priority
   - conda-forge
   - default
dependencies:  # list of packages required for your work
   - python >=3.10
   - numpy
   - pandas
   - matplotlib
   - scikit-learn
   - nodejs # conda can install non-Python packages as well, if it's available on a channel
   - pip
   - pip:   # Optionally, conda can also install packages using pip if needed
      - pytest

conda uses this file to create a conda environment.

tip

In some cases, installing packages using pip through conda can cause issues dependency conflicts. We suggest you use the pip: section only if the package you need is not available on conda-forge.

Learn more in the conda documentation about created an environment file manually

Environment creation

Given an environment.yaml file, this is how conda performs a build (in brief):

1. Conda downloads channeldata.json, a metadata file from each of the channels which list the available architectures.

2. Conda then downloads repodata.json for each of the architectures it is interested in (specifically your particular compute architecture along with noarch¹). The repodata.json has fields like package name, version, and dependencies.

tip

You may notice that the channels listed in the YAML do not have a URL. This is because in general , non-URL channels are expected to be present at https://conda.anaconda.org/<channel-name>.

3. Conda then performs a solve to determine the exact version and sha256 of each package to download.

4. The specific packages are downloaded.

For a detailed walkthrough, check out the conda install deep dive in the conda documentation.

Understand how conda-store builds on conda for improved reproducibility in conda-store concepts page.

Conda configuration (conda config)

You can configure various behaviors in conda through the .condarc configuration file.

conda-store needs to configure some parts of conda without modifying your conda configuration file, for this conda-store (internally) sets some conda configuration variables using environment variables.

The impact of this is that if you try to print your conda configuration with conda config --show CLI command, some configuration settings displayed by that command will not reflect the values that are actually used by conda-store.

In particular, conda-store internally sets CONDA_FLAGS=--strict-channel-priority, overriding the channel priority in the conda configuration file. Keep this in mind when using conda config to inspect your conda configuration and when viewing the build logs.

Footnotes

1. noarch is a cross-platform architecture which has no OS-specific files. Read noarch packages in the conda documentation for more information. ↩