A collection of Conda recipes used internally at Memfault but shared broadly.

The packages can be found on the Anaconda Package Repository

Using

To use any of these packages in your own Conda environments, just add memfault to the top of the environment.yml in your project:

channels:
  - memfault
  - conda-forge
  - nodefaults

Since all of these packages are built using Conda Forge's package pinnings (https://github.com/conda-forge/conda-forge-pinning-feedstock/blob/master/recipe/conda_build_config.yaml), using Conda Forge as the base is heavily suggested.

Building via GitHub Action

Not heavily tested, but it's possible to build packages from github actions, see .github/workflows/build.yml.

To trigger it, set the appropriate PACKAGE_DIR when making a pull request.

Note that this may not work if the above CONDA_BUILD_SYSROOT is set; you'll have to add a step to install the appropriate tools into that location if you want to go that route.

From the Github UI, you can trigger the build by going here, then:

• Click "Run Workflow".
• Enter the package directory in the designated input field.
• Hit "Run".

Once the Github action has built the packages, they still need to be uploaded to anaconda.org manually. Go to the detail page of your workflow run and download the "packages" artifact.

Unzip the packages.zip and then run:

PACKAGE=<package_name> anaconda upload **/$PACKAGE*.tar.bz2 --user memfault

Building Locally

To build any of the following packages (macOS and Linux Ubuntu 18.04 tested):

# Create build environment
$ conda create -n build conda-build anaconda-client
$ conda activate build

# Build specific recipe
$ cd <some_recipe_dir>
$ conda build -c conda-forge .

# Successful build prints an upload command
$ anaconda upload ...

You can optionally install + use the mambabuild build command, see here:

https://boa-build.readthedocs.io/en/latest/mambabuild.html

It can significantly speed up package dependency resolution during the build.

Docker for Linux

So you don't want to build on your native machine? That's fine!

$ docker run -ti -v $(pwd):/conda-recipes condaforge/miniforge3  /bin/bash
# force an architecture, e.g. building linux amd64 on macOS Apple Silicon
$ docker run --platform=linux/amd64 -ti -v $(pwd):/conda-recipes condaforge/miniforge3  /bin/bash

$ apt update && apt install -y build-essential
$ conda create -n build conda-build anaconda-client
$ conda activate build
$ cd /conda-recipes/<recipe>
$ conda build -c conda-forge .

# Successful build prints an upload command
$ anaconda upload ...

Building on macOS

We follow Conda Build's (and Conda Forge's) strategy for building macOS packages.

As noted in the conda_build_config.yaml of each recipe, we use the MacOS 10.9 SDK.

CONDA_BUILD_SYSROOT:
  - /opt/MacOSX10.9.sdk # [osx]

To download and install this SDK, you can find the package here: https://github.com/phracker/MacOSX-SDKs/releases

$ sudo mv <10.9 SDK> /opt/MacOSX10.9.sdk

Apple Silicon

If you're on Apple Silicon, you can build for both ARM64 and X86_64 via Rosetta. The default environment is osx-arm64, but you can explicitly create them with CONDA_SUBDIR:

# create an Apple Silicon environment
CONDA_SUBDIR=osx-arm64 conda create -n build-silicon conda-build anaconda-client
conda activate build-silicon
conda config --env --set subdir osx-arm64

# create a Rosetta environment
CONDA_SUBDIR=osx-64 conda create -n build-rosetta conda-build anaconda-client
conda activate build-rosetta
conda config --env --set subdir osx-64

Then follow the Building Locally instructions at the top.

Uploading Packages

It's nice to convert packages to the new .conda archive format, see here for details:

https://docs.conda.io/projects/conda/en/latest/user-guide/concepts/packages.html

Only the .conda package needs to be uploaded (conda clients 4.7 (2019-05-17) and later support the .conda package format).

You can make this the default package format by adding the following to your ~/.condarc:

conda_build:
  pkg_format: 2
  zstd_compression_level: 19

Or using the conda config command:

conda config --set conda_build.pkg_format 2
conda config --set conda_build.zstd_compression_level 19

Reference:

conda/conda-docs#796 (comment)

If you built a package as a .tar.bz2 but want to convert it to a .conda package, you can do so with:

cph transmute /path/to/package.tar.bz2 .conda

Useful Resources

• Creating patch files for fixing builds: https://www.anaconda.com/blog/patching-source-code-to-conda-build-recipes