Ensure all plugins are compatible with Datasette 1.0a37 and above

[simonw]

In preparation for Datasette 1.0 stable we need to make sure the test suites pass for every one of our plugins.

[simonw]

I built a repo that uses GitHub Actions to run plugin test suites against as many plugins as possible.

Here's the generated dashboard (thanks, GPT-5.6 Sol): https://datasette.github.io/ready-for-datasette/

It runs out of this repo: https://github.com/datasette/ready-for-datasette

Currently it will only work for plugins that publish a .tar.gz source package on PyPI which includes their test suite. Any plugin that was created using the datasette-plugin cookiecutter template should do this.

[simonw]

I think a bunch of plugins are missing the tests from their source distribution, possibly because there are configurations of pyproject.toml that do not produce this correctly. Maybe ones using uv-build?

[simonw]

As part of this I'm having to finish the upgrade from setup.py to pyproject.toml, and since I'm doing that I may as well make sure trusted publishing is setup for everything too.

[simonw]

I think the next step for this project is to create a skill-style upgrade guide that Codex/Claude Code can use to rattle through all of the failures and fix the most obvious errors.

[simonw]

Here's an older plugin (still setup.py) and what it takes to upgrade it (on my laptop): https://github.com/simonw/datasette-debug-asgi

1. cd ~/dev/ecosystem/datasette-debug-asgi
2. git remote get-url origin - check that it's git@ and not https://
   • git remote set-url origin git@github.com:simonw/datasette-debug-asgi.git (if previous test failed)
3. If no pyproject.toml:

   codex exec 'follow instructions from curl https://raw.githubusercontent.com/simonw/skills/refs/heads/main/setup-to-pyproject/SKILL.md'

4. Check .github/workflows/*.yml to see if they are the most recent pattern - if not:

   curl 'https://raw.githubusercontent.com/simonw/datasette-plugin-template-demo/refs/heads/main/.github/workflows/publish.yml' -o .github/workflows/publish.yml
   curl 'https://raw.githubusercontent.com/simonw/datasette-plugin-template-demo/refs/heads/main/.github/workflows/test.yml' -o .github/workflows/test.yml
   git commit -m "Upgrade GitHub Actions"

5. Run uv run purest
6. Fix any failures.

[simonw]

I'm skeptical of https://github.com/simonw/skills/blob/main/setup-to-pyproject/SKILL.md - I'm not sure it reliably produces source distributions that bundle the tests.

[simonw]

Something needs to add uv.lock to .gitignore too.

[simonw]

Here's a fun twist: because datasette-debug-asgi has a test_debug_asgi.py file in the root of the directory and not in a tests/ directory none of the default behavior of the different build tools bundle that test.

Easiest solution is probably to start using the tests/*.py format even for modules with only one module and only one test.

[simonw]

Might need to use:

codex exec - < prompt.txt

And it's not clear if Codex can edit files without:

codex exec --sandbox workspace-write

Need to test that - also make sure it can exec uv run pytest.

[simonw]

Maybe I should switch to Hatchling:

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

No extra configuration should be necessary:

• Its sdist includes all files not ignored by Git.
• Its wheel builder recognizes the project’s single datasette_debug_asgi.py module.
• Therefore the test enters the sdist but not the wheel.

Those are documented Hatchling defaults: sdist selection (https://hatch.pypa.io/dev/plugins/builder/sdist/) and wheel
selection (https://hatch.pypa.io/dev/plugins/builder/wheel/).