What is a Python package?
A Python package is a collection of related code modules (files) bundled with metadata describing how the package should be installed and used.
There are three common formats for Python packages, each serving different purposes:
- Wheels: Pre-built binary distributions that can be quickly installed with pip or uv.
- Source distributions (sdists): Raw source code that may need to be built during installation
- Conda packages: A format specific to the Conda ecosystem that can include non-Python components
Package vs Application Comparison
Packages and applications serve fundamentally different purposes:
Packages:
- Provide reusable code modules/libraries
- Focus on API design and documentation
- Published to package indexes like PyPI
- Carefully manage dependencies for compatibility
- Designed to be imported by other Python code
Applications:
- Have a main entry point (e.g.,
main.pyor command-line interface) - Focus on end-user functionality
- Often include configuration files and user settings
- May bundle dependencies for deployment
- Typically not imported by other Python code
Compiled vs Non-Compiled Python Packages
Python packages may or may not include extensions written in languages other than Python, e.g., C, C++, and Rust. The distinction between wheels and source distributions is most pertinent when extensions are included. Source distributions include the raw extension source code, which must be compiled at installation time, while wheels (as well as Conda packages) include pre-compiled binary versions.
Build Backends, Frontends, and Package Managers
Several types of tools are involved in the construction and installation of packages:
- A build backend (like setuptools or hatchling) is the engine that knows how to construct a package from source code
- A build frontend (like build or pip) is the tool that coordinates the build process and handles user interaction
- A package manager (like uv or pip) focuses on installing and managing dependencies in environments