ARM64 on Bioconda

ARM64 support is increasingly important as cloud providers offer ARM-based instances (like AWS Graviton) that are 20-40% cheaper than x86 alternatives with comparable performance. Enabling ARM64 compatibility allows users to leverage these cost-effective platforms.

This page explains how to enable existing Bioconda and Conda-forge packages to work on ARM64 architecture (linux/arm64).

Bioconda and Conda-forge

Bioconda is a conda channel for bioinformatics packages, while conda-forge provides general-purpose recipes. Both systems differ in CI infrastructure and recipe structure, though they share commonalities.

This documentation addresses adapting existing bioconda/conda-forge packages to support linux/arm64 (ARM processors like AWS Graviton) alongside the standard linux/amd64 (Intel architecture).

What you need to do

The goal is to enable users to run conda install <package> and receive the package on ARM systems.

Set up your environment

Install Bioconda

To use Bioconda on ARM systems:

1. Install Conda:

   wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-aarch64.sh
   sh ./Miniforge3-Linux-aarch64.sh

   Select “Yes” at installation completion.

2. Install Bioconda:

   conda config --add channels bioconda
   conda config --add channels conda-forge
   conda config --set channel_priority strict

Create a development environment (optional)

For local development on ARM machines:

1. Create a development environment:

   conda create -n bioconda-builds -c conda-forge -c bioconda bioconda-utils
   conda activate bioconda-builds

Enable packages

Test locally

On ARM machines, test whether a package installs correctly:

conda install <name>

If the package fails to install, it needs ARM64 support.

Test with Wave

If you do not have access to ARM hardware, use Wave through Seqera Containers or the Wave CLI:

wave --conda <name> --platform linux/arm64 --freeze --await

• --freeze persists generated images to Seqera Containers
• --await maintains execution until completion

Build failures display error messages with links to detailed build pages.

Interpret test output

Build failures typically involve two categories:

1. “Not found” errors that require package fixes
2. Missing dependencies that need investigation

Most fixable packages reside in Bioconda. However, conda-forge packages occupy individual GitHub repositories under the conda-forge organisation, formatted as <package-name>-feedstock.

Bioconda recipes

Generic recipes

Generic packages may have missing non-generic dependencies that prevent functionality until dependencies are resolved.

Generic package indicators in recipes/<name>/meta.yaml:

build:
  ..
  noarch: generic
  ..

When generic packages fail, dependencies require fixing. This is often resolved through simple build number increments and pull requests. Since generic packages lock to specific build versions that pre-date bioconda ARM support, they may need rebuilding.

Non-generic recipes

Non-generic packages compile native code or download binaries (typically x86-only). Examine meta.yaml and adjacent build.sh files.

Enable ARM builds by adding to meta.yaml:

extra:
  additional-platforms:
    - linux-aarch64
    - osx-arm64

Bump the build number by adding 1 to the existing value. Add a package versioning pin if missing:

build:
  number: 2
  noarch: generic
  run_exports:
    - { { pin_subpackage(name, max_pin='x.x') } }

Test in the CI

Open a pull request and monitor CI results. Common solutions include:

• Source file patching: meta.yaml handles patches
• Platform-specific logic: build.sh supports conditional bash logic. Use CFLAGS, CC, and ${PREFIX} (dependency base directory containing bin, lib, include)
• Build isolation: CI output shows conda packages built in isolated environments

Test builds locally

Build locally for faster iteration:

bioconda-utils build --packages <package>

Consult the Bioconda dev instructions for detailed guidance.

Conda-forge

Quick fixes

For conda-forge packages (for example, a package named perl-nonsense):

1. Navigate to https://github.com/conda-forge/<package-name>-feedstock
2. Check for existing ARM migration pull requests. If a stale PR exists, comment: @conda-forge-admin please rerender
3. If no outstanding PR exists, edit the migrations list alphabetically. Bots will automatically initiate a PR.
4. When builds succeed, maintainers merge and binaries become available within a few days.
5. For unresponsive maintainers, comment: @conda-forge-admin, please ping team or @conda-forge/core please help review and merge this PR

Manual fixes

Clone the feedstock:

git clone git@github.com:<your-github-username>/<package-name>-feedstock.git
cd <package-name>-feedstock
git remote add bot https://github.com/regro-cf-autotick-bot/<package-name>-feedstock
git fetch bot
git checkout -b aarch64-fixes bot/bot-pr_arch_<branch-suffix>

Replace <branch-suffix> with the actual migration branch name. Use tab completion after typing bot/bot-pr_arch_ to find available branches.

Fix issues by editing meta.yaml or conda-forge.yml, then:

git commit --all
git push --set-upstream origin aarch64-fixes

Create a new PR from your branch. Request rerendering by commenting: @conda-forge-admin please rerender