New Recipes

Conda includes a helper function to quickly build recipes from existing repositories, like PyPI and CRAN:

conda skeleton cran <package-name>

creates a directory r-<package-name> with a recipe

conda skeleton pypi <package-name>

creates a directory <package-name> with a recipe

Hard-coding Versions

To minimize incompatibilities later on, we are using hard-coded versions in all meta.yaml dependencies. So if skeleton returns a recipe with these dependencies:

numpy
scipy

we would fix them to a specific version, e.g.:

numpy ==1.13.0
scipy ==0.19.0

Docker Build Image

A CentOS/6.8 build environment is available from Docker:

docker pull mhowison/conda-build

To build a new package, use an interactive session with this image (make sure to map a local path containing your recipe directory, e.g. in agalma/dev/conda):

docker run -it -v $HOME/agalma/dev/conda:/recipes mhowison/conda-build bash

cd /recipes/<package-name>
conda build -c dunnlab .

Anaconda Channel

Built packages are pushed to the dunnlab channel.

First, checkout an anaconda token (needs to be done once per Docker container):

anaconda login

Push the package:

anaconda upload -u dunnlab /path/to/built/package-name.tar.bz2

(Note: the conda build command will print the path to the built tarball, which you can copy and paste into the command above.)