Create conda package across many versions

Question

I have a very simple Pure Python package on PyPI that I'd like to make available on binstar. My package targets Python 2.6+ and 3.2+ with a single codebase. I also expect that it works equally well on Windows as well as Unix.

Is there a simple way to comprehensively build my package and upload it to binstar for many Python versions?

I've tried a naive use of conda skeleton pypi as suggested in this article. I'd like to replicate this process across many different versions.

Answer 1

If you want to build recipes for many different versions of the package, use the --version flag to conda skeleton pypi. I recommend using package-version as a naming convention for the recipes.

If you want to build the same package for many different Python versions, use the --py flag to conda build, like conda build --py 3.4 recipe. The flag can be specified multiple times, like conda build --py 3.3 --py 3.4 recipe, or you can use --py all to build against Python 2.6, 2.7, 3.3, and 3.4.

To convert the package to other platforms, use conda convert (see conda convert -h for usage; be sure to run conda update conda-build, as the API changed a little bit recently).

The easiest way to get the binstar uploading is to run conda config --set binstar_upload yes. This will cause conda build and conda convert to upload the packages to binstar automatically when they are done.

Answer 2

For pure Python packages that can run on any Python version, you can use the noarch: python syntax in the recipe.
Something like:

requirements:
  build:
    - python
    - setuptools

  run:
    - python

build:
  noarch: python
  script: python setup.py install --single-version-externally-managed --record=record.txt

Link to the conda-build documentation: https://docs.conda.io/projects/conda-build/en/latest/resources/define-metadata.html#architecture-independent-packages