conda-forge / conda-smithy

The tool for managing conda-forge feedstocks.

Home Page:

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool


conda-smithy is a tool for combining a conda recipe with configurations to build using freely hosted CI services into a single repository, also known as a feedstock. conda-smithy is still a work-in-progress, but when complete, conda-smithy will:

  • Create a git repo with a conda recipe and the files to run conda builds via CI services.
  • Register the repo on github and push it.
  • Connect the repo to the CI services,,, (For, configure your org or user to enable the service for all repos)

tests Coverage Status Code style: black


The easiest way to install conda-smithy is to use conda and conda-forge:

conda install -n root -c conda-forge conda-smithy

To install conda-smithy from source, see the requirements file in requirements.txt, clone this repo, and python -m pip install ..


You need a token from github,, and to try out conda-smithy. The commands which need this will tell you where to get these tokens and where to place them. If you need help getting tokens please ask on the conda-forge google group.

You should be able to test parts of conda-smithy with whatever tokens you have. For example, you should be able to conda smithy register-github without the CI service tokens. Re-rendering an existing feedstock is also possible without CI service tokens set.

Re-rendering an existing feedstock

Periodically feedstocks need to be upgraded to include new features. To do this we use conda-smithy to go through a process called re-rendering. Make sure you have installed conda-smithy before proceeding. Re-rendering an existing feedstock is possible without CI service tokens set.

  1. cd <feedstock directory>
  2. conda smithy rerender [--commit]
  3. Commit and push all changes

Optionally one can commit the changes automatically with conda-smithy version 1.4.1+. To do this just use the --commit/-c option. By default this will open an editor to make a commit. It will provide a default commit message and show the changes to be added. If you wish to do this automatically, please just use --commit auto/-c auto and it will use the stock commit message.

Making a new feedstock

  1. Make the feedstock repo: conda smithy init <directory_of_conda_recipe>. For a recipe called foo, this creates a directory called foo-feedstock, populates it with CI setup skeletons, adds the recipe under recipe and initializes it as a git repo.

  2. Create a github repo: conda smithy register-github --organization conda-forge ./foo-feedstock. This requires a github token. You can try it out with a github user account instead of an organization by replacing the organization argument with --user github_user_name. If you are interested in adding teams for your feedstocks, you can provide the --add-teams option to create them. This can be done when creating the feedstock or after.

  3. Register the feedstock with CI services: conda smithy register-ci --organization conda-forge --feedstock_directory ./foo-feedstock. This requires tokens for the CI services. You can give the name of a user instead of organization with --user github_user_name. By default this command requires an Anaconda/Binstar token to be available in ~/.conda-smithy/anaconda.token, or as BINSTAR_TOKEN in the environment. This can be opted out of by specifying --without-anaconda-token, as such execpted package uploads will not be attempted.

    • For Azure, you will have to create a service connection with the same name as your github user or org
    • For Azure builds, you will have to export the environment variable AZURE_ORG_OR_USER to point to your Azure org
    • If this is your first build on Azure, make sure to add Library Variable Group containing your BINSTAR_TOKEN for automated anaconda uploads.
  4. Specify the feedstock channel and label: Optionally, you can specify source channels and choose a channel to upload to in recipe/conda_build_config.yaml.

      - mysourcechannel1,mysourcechannel2,conda-forge,defaults
      - target_channel target_label

    Default source channels are conda-forge,defaults. Default for channel targets is conda-forge main.

  5. Specify your branding in the Optionally, you can specify the branding on the file by adding the following the conda-forge.yml file:

      user_or_org: YOUR_GITHUB_USER_OR_ORG
  6. Re-render the feedstock: conda smithy rerender --feedstock_directory ./foo-feedstock

  7. Commit the changes: cd foo-feedstock && git commit, then push git push upstream master.

Running a build

When everything is configured you can trigger a build with a push to the feedstock repo on github.

Releasing conda-smithy

Before making a release, consult @conda-forge/core and wait some time for objections.

To release a new version of conda-smithy, you can use the rever release managment tool. Run rever in the root repo directory with the version number you want to release. For example,

$ rever 0.1.2

Conda-smithy in a nutshell

xkcd 1629: Tools

xkcd 1629: Tools

Titletext: I make tools for managing job-hunting sites for people who make tools for managing job-hunting sites for people who make tools for ...


The tool for managing conda-forge feedstocks.

License:BSD 3-Clause "New" or "Revised" License


Language:Python 94.7%Language:Shell 4.6%Language:Jinja 0.5%Language:Xonsh 0.1%Language:Batchfile 0.0%