Namespace packages in Python 3.3+

Question

Accepted Answer

Before PEP 420 (Python 3.3), a directory needed __init__.py to be a package. Namespace packages remove that requirement. The real power is that a single logical package can span multiple directories. Suppose two separate library distributions both put files under 'acme/'. With namespace packages, Python merges them: 'import acme.billing' finds it in one directory, 'import acme.shipping' finds it in another, and they all look like one 'acme' package to the caller. The practical impact for data scientists: if you install two packages that both contribute to the same top-level namespace (common in large organizations with monorepo distributions), namespace packages make it work automatically. The tradeoff is there is no __init__.py to run on first import, so you cannot put initialization code at the package level - you have to put it in submodules instead.