DOC: clarify NpzFile handling of non-.npy archive members

[LarytheLord]

Summary

• update numpy.lib.npyio.NpzFile docs to match current runtime behavior
• clarify that .npy members are loaded as arrays, while non-.npy members are returned as raw bytes
• clarify files attribute wording: it lists all archive members, with .npy suffix removed where present

Closes #26004.

Validation

• python3 -m compileall numpy/lib/_npyio_impl.py

[jorenham]

I'm not sure what python3 -m compileall numpy/lib/_npyio_impl.py does, but we use spin for building numpy and the docs: https://numpy.org/devdocs/dev/spin.html

[Copilot]

Pull request overview

This PR updates the NpzFile class documentation to accurately reflect its runtime behavior regarding non-.npy archive members. The original documentation incorrectly stated that files without a .npy extension are ignored, but the actual implementation loads them as raw bytes. This fixes issue #26004 by clarifying that .npy files are loaded as arrays while other files can be read as raw bytes.

Changes:

• Clarified that files with .npy extension are loaded as NumPy arrays, while other files are returned as raw bytes
• Updated the files attribute description to indicate it lists all archive members (not just .npy files), with .npy extensions removed where present
• Updated the description text to accurately reflect that .npy extension removal happens for all files in the archive

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[LarytheLord]

Thanks for the note. You’re right: compileall was only a quick syntax sanity check on my side, not a NumPy build/doc validation step. I’ll use spin for validation notes on NumPy PRs going forward.

[LarytheLord]

CI note: the current GCC / riscv64 failures appear infra-related (Could not resolve ports.ubuntu.com during apt install), before test/build steps run. This does not look specific to the doc change in this PR.

[LarytheLord]

Closing this to reduce review load after opening too many PRs too quickly. I apologize for the noise.

[rkern]

I'm not sure what python3 -m compileall numpy/lib/_npyio_impl.py does, but we use spin for building numpy and the docs: https://numpy.org/devdocs/dev/spin.html

It's a Claudism. It likes to check that the Python it wrote is syntactically valid with compileall.

[LarytheLord]

Fair point, and thanks for calling it out.

I should have validated this using NumPy’s documented spin workflow and reported that, instead of using a generic syntax check. I’ve stopped the batch-PR behavior and I’m being more deliberate with project-specific validation and communication.