Python Packaging milestones & Internship projects across the scientific Python ecosystem

The Quansight Quarterly

Covering the latest in scientific python and data science

OPEN SOURCE PROJECT & COMMUNITY HIGHLIGHTS

Python Packaging

PEP 825 – Wheel Variants: Package Format has been provisionally accepted. This marks a milestone for the Wheel Variants effort where we, together with many collaborators across companies and open source projects, aim to bring first-class support of accelerators and other performance-relevant build configurations to Python packaging.

Building on this, PEP 850 – Wheel Variants: Providers, which defines how variant properties are governed and how their compatibility is determined, is now open for community review. Additionally, PEP 817 – Wheel Variants: Beyond Platform Tags, an informational PEP, has been updated to provide a high-level overview of the current design and a comparison of different solutions.

In governance news, the inaugural Python Packaging Council (PPC) was elected in September. With decision-making authority over Python’s packaging standards, processes, and documentation, this governance body is a meaningful step towards creating a more consistent packaging experience for everyone. We’re delighted that our Co-CEO, Ralf Gommers, was elected to the Council, among its first members.

FROM THE TEAM

Blog Highlights

Scaling NumPy on Free-Threaded Pythonby Kumar Aditya

Describes fixing bottlenecks in NumPy and CPython, to help NumPy scale well on free-threaded Python. One test workload dropped from 44 seconds to about 1.5 seconds across 32 cores.

The Polars vs SQL differences nobody is talking aboutby Marco Gorelli

Explains subtle differences between Polars and SQL through their contrasting mental models. It covers row-order preservation, the risks of column independence, and more, with tips for writing code that ports cleanly between them.

Write It Like C: What V8's Optimizing Compilers Explain About the Zen of stdlibby Mara Averick

Shares the reasons behind stdlib’s coding principles and finds that much of what reads like style advice is really a description of how the V8 JavaScript engine optimizes code.

OPEN SOURCE PROJECTS & COMMUNITY HIGHLIGHTS

• The CPython Limited C API and the Stable ABI are now supported for free-threaded Python, with abi3t support in Python 3.15 and in a number of build backends and binding generators.

• PyTorch 2.14+ migrated its build machinery from setuptools (with custom build scripts) to scikit-build-core, maintaining its existing CMake configuration and wheels. Driven by the deprecation of setup.py commands like develop and install, this migration provides a reusable pattern for modernizing builds.

COMMUNITY ENGAGEMENT

Featured Talks & Videos

EuroPython: What every Python developer should know about the CPython ABIby Nathan Goldbaum

Explains what the CPython ABI is and how it differs from the CPython C API, how it determines whether a PyPI package is installable with a given CPython, how it is evolving for free-threading and a new stable ABI in Python 3.15, and more.

EuroPython: Building your DSL compiler in Pythonby Mateusz Sokół

Introduces Finch, a language and compiler for sparse and structured multidimensional arrays, and shares lessons on compiler design learned from porting Finch from its original Julia implementation to pure Python.

Openscapes Community Call: Open Communities in the Age of AIby Mara Averick (Quansight), Hadley Wickham (Posit), Yanina Bellini Saibene (rOpenSci), Mark Padgham (rOpenSci)

Talks and facilitated panel discussion on the impact of AI on the growth and maintenance of open source projects and communities.