This repository contains documents, tooling and other content related to the API standard for arrays (or tensors).

These are relevant documents related to the content in this repository:

• Rendered HTML docs for latest version
• Consortium announcement blog post
• Blog post on first release of draft array API standard

See CONTRIBUTING.md for how to go about contributing to this array API standard.

To coordinate activities and facilitate discussion across the Python ecosystem, the Consortium for Python Data API Standards holds an open meeting every other week on Thursdays throughout the year.

This meeting is open to the public. So if you have questions about the array API standard, pain points, feature requests, or are just wanting to listen in, we encourage you to join the public call to connect with Consortium members, stay updated on the latest Consortium activities, and chat with others in the community.

Everyone is welcome—drop in and say hello!

UTC 17:00 (05:00 PM) every other Thursday:

• Public calendar: https://calendar.google.com/calendar/embed?src=8fe9013a2cb5d3409bb236d04eca73fa5227eac01c02ea8f6bc4a6a3cf982fa3%40group.calendar.google.com

• See the public calendar for information on how to join!

• Have specific things to discuss? Feel free to comment on data-apis#853 or the linked agenda document!

To install the local stubs and additional dependencies of the Sphinx docs, you can use pip install -r doc-requirements.txt. Then just running make at the root of the repository should build the whole spec website.

$ pip install -r doc-requirements.txt
$ make
$ ls _site/
2021.12/  draft/  index.html  latest/  versions.json

The spec website is comprised of multiple Sphinx docs (one for each spec version), all of which exist in spec/ and rely on the modules found in src/ (most notably array_api_stubs). For purposes of building the docs, these src/ modules do not need to be installed as they are added to the sys.path at runtime.

To build specific versions of the spec, run sphinx-build on the respective folder in spec/, e.g.

$ sphinx-build spec/2012.12/ _site/2012.12/

Additionally, make draft aliases

$ sphinx-build spec/draft/ _site/draft/

To build the whole website, which includes every version of the spec, you can utilize make spec.

The Sphinx doc at spec/draft/ should be where the in-development spec resides, with src/array_api_stubs/_draft/ containing its respective stubs. A spec release should involve:

• Renaming src/array_api_stubs/_draft/ to src/array_api_stubs/_YYYY_MM

• Renaming spec/draft/ to spec/YYYY.MM

• Updating spec/YYYY.MM/conf.py

  ...
  - from array_api_stubs import _draft as stubs_mod
  + from array_api_stubs import _YYYY_MM as stubs_mod
  ...
  - release = "DRAFT"
  + release = "YYYY.MM"
  ...

• Updating spec/_ghpages/versions.json

  {
  +     "YYYY.MM": "YYYY.MM",
  ...

• Updating Makefile

  ...
  	-sphinx-build "$(SOURCEDIR)/PREVIOUS.VER" "$(BUILDDIR)/PREVIOUS.VER" $(SPHINXOPTS)
  + 	-sphinx-build "$(SOURCEDIR)/YYYY.MM" "$(BUILDDIR)/YYYY.MM" $(SPHINXOPTS)
  - 	-cp -r "$(BUILDDIR)/PREVIOUS.VER" "$(BUILDDIR)/latest"
  + 	-cp -r "$(BUILDDIR)/YYYY.MM" "$(BUILDDIR)/latest"
  ...

These changes should be committed and tagged. The next draft should then be created. To preserve git history for both the new release and the next draft:

1. Create and checkout to a new temporary branch.

$ git checkout -b tmp

2. Make an empty commit. This is required so merging the temporary branch (4.) is not automatic.

$ git commit --allow-empty -m "Empty commit for draft at YYYY.MM "

3. Checkout back to the branch you are making a spec release in.

$ git checkout YYYY.MM-release

4. Merge the temporary branch, specifying no commit and no fast-forwarding.

$ git merge --no-commit --no-ff tmp
Automatic merge went well; stopped before committing as requested

5. Checkout the spec/draft/ files from the temporary branch.

$ git checkout tmp -- spec/draft/

6. Commit your changes.

$ git commit -m "Copy YYYY.MM as draft with preserved git history"

You can run git blame on both spec/YYYY.MM and spec/draft files to verify we've preserved history. See this StackOverflow question for more background on the approach we use.

Thanks goes to these wonderful people (emoji key):

Saul Shanabrook 🔧 🤔 🔬	Athan 🖋 🔣 🔧 🔬	Stephannie Jimenez Gacha 🔣 🖋 🔬	Aaron Meurer 🖋 ⚠️ 🔧	Tony Fast 🚧	Ralf Gommers 📝 💼 💻 🖋 📖 🔍 🚧 🤔 📆 📢	Travis E. Oliphant 💼 🔍 🤔
Leo Fang 👀 🤔 🖋	Tianqi Chen 🤔 👀	Stephan Hoyer 🤔 👀 💬	Alexandre Passos 🤔 👀	Paige Bailey 🔍	Adam Paszke 🤔 👀 📢	Andreas Mueller 🤔 👀
Sheng Zha 🤔	kkraus 🤔 👀 📢	Tom Augspurger 👀 💬	edloper 👀 💬	Areg Melik-Adamyan 👀 🔍	Oleksandr Pavlyk 👀 💬	tdimitri 🤔
Jack Pappas 🤔	Ashish Agarwal 👀 💬	Edward Z. Yang 🤔	Mike Ruberry 🤔	Eric Wieser 🤔	Carol Willing 🤔	Alex Rogozhnikov 🤔
Matthew Honnibal 🤔	Mario Lezcano Casado 🤔	Bas van Beek 🤔	Sebastian Berg 🤔	Isaac Breen 🤔	Kenichi Maehashi 🤔	Chris Pryer 🤔
Tirth Patel 🤔	Kshiteej K 🤔	Anirudh Dagar 🤔	Tom White 🤔	Matthew Barber 🤔 🖋	Philip Meier 🔬 💻	Zac Hatfield-Dodds 🤔 💻
Daniel Lenton 💻	Simone G 💻 🤔	Tyler Reddy 🤔	Matt Barrett 🤔	Jatin Prakash 🤔	Ishtiaq Hussain 🤔	sherry30 🤔
João Lobo 🤔	Neil Girdhar 🤔	Nathaniel Starkman 🤔	jakirkham 🤔	RickSanchezStoic 🤔	Talley Lambert 🤔	Juan Nunez-Iglesias 🤔
Christian Kothe 🤔	Carlos Ramos Carreño 🤔	Gilad 🤔	Thomas J. Fan 🤔	Conchylicultor 🤔	Franck Charras 🤔	Keith Kraus 🤔
Lucas Colley 🚧 🐛

This project follows the all-contributors specification. Contributions of any kind welcome!