Python Package and Wheel

An import package is code organized under a name Python can import. A distribution package is the installable project described by packaging metadata. Their names need not match: a distribution called orders-pipeline can provide an import package called orders_pipeline, and one distribution can provide more than one import package.

A wheel is a built distribution archive containing installable files and metadata. It is not a full virtual environment. Required libraries still need to be installed, and some wheels are compatible only with particular interpreters or platforms. A build backend creates the artifact from source and configuration; the presence of a source directory alone does not establish what will be included.

During development, an editable installation exposes the source so edits become visible without repeatedly rebuilding a wheel. That is useful for iteration but can hide packaging omissions. Build the wheel, install it in a separate environment, and run from outside the source checkout to avoid accidentally importing local files instead of the installed artifact.

Entry points are another part of the contract. A configured console command calls its designated function. Running an import package with python -m uses its __main__ module. Exercise the intended entry points and required data files in the installed artifact. A successful startup is a smoke test; it does not replace tests of parsing, business rules, or integration with actual services.

Reference: Python Packaging User Guide. For worked examples, see Python for Data Engineers.