Overview

The process of creating an installer consists of three steps. The first two tend to be platform-agnostic unless you have code that needs to be compiled (e.g., Cython code):

1. Build the source distribution and wheel that will be uploaded to PyPI
2. Build the conda package
3. Create the installer

The Python script, packages/build.py, facilitates the first two steps. The Python script, installers/build.py facilitates the last step.

Packaging manually

First, make sure your package works. This is important if you've been running in developer mode:

conda create -n <test_env_name> python=<python version>
cd <source_dir>
python setup.py install
<commands to make sure it works>

Now, remove the env:

conda remove -n <test_env_name> --all

You need a build environment:

conda create -n build conda-build constructor anaconda-client

Packages in the environment:

• conda-build: builds conda packages
• constructor: creates installers
• anaconda-client: automatically uploads to your conda channel

First, build your PyPI distribution:

cd <source_dir>
python setup.py sdist bdist_wheel
twine upload dist/*

At this point, it may be worthwhile to verify that installing from pip works by creating another test environment. Once you're satisfied everything works, build your conda package:

cd <recipe_dir>
conda-build <package_name>

Now, cleanup if desired:

conda-build purge

Installers

Windows

You need a build environment:

conda create -n build constructor

Useful links

• General instructions on building conda packages.
• Wiki for menuinst.