Join the forge!

Overview

• History of conda-forge
• Submitting your own recipes (i.e. packages)
• How to get involved as a volunteer

• The community was moving from easy_install/eggs to pip, but there was no easy way to ship pre-compiled packages
• Wheels weren't formalized until 2013 (PEP 427); on Windows, Christoph Gohlke's builds were often your only option
• Linux users had to compile from source — manylinux wheels didn't exist until PEP 513 in 2016

• 2012: Continuum Analytics announces Anaconda 0.8 at SciPy; in September, conda 1.0 ships as a cross-platform, language-agnostic package manager for pre-compiled artifacts
• conda packages could ship compiled shared libraries and Python itself, not just Python code
• By June 2013: conda gets a SAT solver, conda build ships, Miniconda and Binstar.org (hosting for user-built conda packages) launch

• ContinuumIO/conda-recipes became the central place to contribute conda recipes — but quality varied, most recipes only worked on 1–2 platforms, and there was no CI
• Groups adopted Binstar/Anaconda.org for their own needs: Omnia (Molecular Dynamics, Mar 2014), the Met Office's SciTools (Jun 2014), IOOS (Jul 2014)
• Phil Elson (SciTools) built CI + tooling (conda-build-all); Filipe Fernandes (IOOS) borrowed it — lots of cross-pollination, but duplicated recipes and unpredictable compatibility when mixing channels

• conda-forge steadily grows its own feedstocks while Anaconda's main channel is born
• Pressure is mounting to move faster by adopting new build toolchains
• conda-forge becomes the public base that Anaconda's main channel is built on top of
• Allows conda-forge to move independently and freely

context:
  name: zensus2pgsql
  version: 0.2.1
  python_min: "3.10"

package:
  name: ${{ name|lower }}
  version: ${{ version }}

source:
  url: https://pypi.org/packages/source/${{ name[0] }}/${{ name }}/zensus2pgsql-${{ version }}.tar.gz
  sha256: 00517129cf7747e7580f57c680e143de5e5accca928407cd79bc88c114502ec5

build:
  number: 0
  noarch: python
  script: ${{ PYTHON }} -m pip install . -vv --no-deps --no-build-isolation
  python:
    entry_points:
      - zensus2pgsql = zensus2pgsql.cli:app
      - zpg = zensus2pgsql.cli:app

requirements:
  host:
    - python ==${{ python_min }}
    - hatchling >=1.27.0
    - pip
  run:
    - python >=${{ python_min }}
    - aiocsv >=1.4.0
    - aiofiles >=25.1.0
    - asyncpg >=0.30.0
    - httpx >=0.28.1
    - platformdirs >=4.5.0
    - rich >=14.2.0
    - typer >=0.15.1
    - types-aiofiles >=25.1.0.20251011

tests:
  - python:
      imports:
        - zensus2pgsql
        - zensus2pgsql.commands.create
        - zensus2pgsql.commands.cache
        - zensus2pgsql.commands.drop
        - zensus2pgsql.commands.list
        - zensus2pgsql.cache
        - zensus2pgsql.constants
        - zensus2pgsql.errors
        - zensus2pgsql.logging
        - zensus2pgsql.retry
      pip_check: true
  - requirements:
      run:
        - python ==${{ python_min }}
    script:
      - zensus2pgsql --help
      - zpg --help

about:
  summary: A Python CLI program that collects Zensus GIS data from Germany and imports it into PostgreSQL
  license: GPL-3.0-only
  license_file: COPYING
  homepage: https://travishathaway.github.io/zensus2pgsql

extra:
  recipe-maintainers:
    - travishathaway

context:
  version: 2.3.1

recipe:
  name: osm2pgsql
  version: ${{ version }}

source:
  - url: https://github.com/osm2pgsql-dev/osm2pgsql/archive/refs/tags/${{ version }}.tar.gz
    sha256: e90461bd78787fd5cf82f73ac4bc0cfccbc4f498f35dce59eb4d168a380d6ac0

build:
  number: 1

package:
  name: osm2pgsql
  version: ${{ version }}

build:
  number: 1
  script:
    - if: unix
      then: build.sh
      else: build.bat

requirements:
  build:
    - ${{ compiler('cxx') }}
    - ${{ stdlib('c') }}
    - cmake
    - ninja
    - pkg-config
  host:
    - expat
    - proj
    - bzip2
    - zlib
    - postgresql
    - lua
    - protozero
    - libosmium
    - cli11
    - fmt <12
    - nlohmann_json
    - libboost-devel
  run:
    - lua
    - osm2pgsql-replication

tests:
  - files:
      recipe:
        - test.sh
        - liechtenstein-2013-08-03.osm.pbf
    requirements:
      run:
        - postgresql
        - if: unix
          then: postgis
    script:
      - osm2pgsql --version
      - osm2pgsql --help
      - if: unix
        then: bash test.sh

about:
  homepage: https://osm2pgsql.org
  license: GPL-2.0-only
  license_file:
    - COPYING
  summary: Import OpenStreetMap data into a PostgreSQL/PostGIS database
  description: osm2pgsql is a tool for loading OpenStreetMap data into a PostgreSQL / PostGIS database suitable for applications like rendering into a map, geocoding with Nominatim, or general analysis.
  documentation: https://osm2pgsql.org
  repository: https://github.com/osm2pgsql-dev/osm2pgsql

extra:
  recipe-maintainers:
    - travishathaway

• At minimum: existence tests — verify the package installed the files you expect (package_contents in v1)
• Python packages: a python: test with imports: + automatic pip_check: true
• R packages: an r: test with libraries:
• Command-line tools: run <tool> --help to prove it's on PATH and executable
• The goal is testing that the conda package built and installed correctly — not re-running the upstream project's full test suite

• License uses the correct SPDX identifier and case (Apache-2.0, not APACHE 2.0)
• A license_file is included, unless the upstream tarball already ships one
• Build from a tarball (url:), never git_url: — tarballs are checksummable and smaller
• Remove the example's explainer comments from your copy of the recipe
• Leave the recipes/example-v1/ folder itself untouched

• It's one of the highest-traffic repos in conda-forge — every new package waits here for a review
• Review times vary from days to weeks to months 😱 depending on how many volunteers are active on a given language team
• Straight from the maintainers: “We are always looking for more staged-recipe reviewers”

• These ecosystems tend to have more compiler-heavy, platform-specific build logic
• Fewer reviewers are comfortable evaluating those recipes, so PRs in these languages wait longer
• Python is the most popular language in the world, so it's easier to find reviewers

• Interested volunteers can reach out to a member of
  @conda-forge/core
• On top of this, the best way is to dig in and start commenting and reviewing on pull requests yourself
• Hangout with us on Zulip and ping me! (Travis Hathaway)

• Can take as little as 5 minutes a day (ditch your morning crossword for a pull request review! 😛)
• Find a build that's failing, examine the logs and find ways to get it working
• Making sure submitted recipes check all the boxes (including licenses, no or limited use of vendoring, passes linter checks)
• Scientifically proven to stimulate brain growth and foster new neural pathway connections*