[TODO] Rewrite the plugin fetcher as a Sphinx extension

[webknjaz]

Hey @nicoddemus @bluetech @RonnyPfannschmidt @The-Compiler are y'all married to the idea of committing this table to the Git repo as RST? I know how to write an in-repo Sphinx extension that would do this on build with proper caching. This would eliminate the need to keep updating RST.

Originally posted by @webknjaz in #12607 (comment)

[webknjaz]

I lost track of this and Bruno's endorsement of the idea (#12607 (comment)). So I'm filing this issue for visibility.

[webknjaz]

This will need a new directive like https://github.com/ansible/ansible-lint/blob/a85f191/docs/rules_table_generator_ext.py#L38-L76. And some cache processing, perhaps: https://github.com/sphinx-contrib/sphinxcontrib-towncrier/blob/01bb987/src/sphinxcontrib/towncrier/ext.py#L295-L326.

[Pierre-Sassoulas]

Generally I like committing the resulting rst because it's easier to review the changes and you can see problems in the generation that you would miss otherwise. In that case the plugin list is very stable, automated, and barely need a review so +1 from me.

[RonnyPfannschmidt]

I prefer the split of generate and doc build

I don't want to make general doc building dependent on a potentially failing process

[webknjaz]

@RonnyPfannschmidt not even with caching?

[RonnyPfannschmidt]

There already is caching

[The-Compiler]

Personally I don't get the benefit of this. There is a lot more complexity, more moving parts, potentially more stuff to take care of on Sphinx upgrades, more reasons for doc builds to fail locally, more cache stuff to take care of (e.g. for CI), etc. etc.

What's the benefit? The only I can see is:

• No periodic upgrade PRs to review anymore (but those are trivial to review, and then we can catch easily if something goes wrong)
• Possibly more recent data as the list gets rebuilt more often maybe (but the caching would probably destroy that again?)
• No long autogenerated file in the git history / searches / etc.

[nicoddemus]

Actually @The-Compiler makes good points, perhaps @webknjaz may comment on them. I'm changing my previous "vote" to -0 for now.

[webknjaz]

I personally wanted to make it more integrated with the underlying tooling (Sphinx) rather than non-local stuff like CI. And yes, I had those benefits that Florian mentioned in mind. But OTOH, I hear your concerns and can't do anything about them other than acknowledge them.

At this point, it's probably safe to say that this is being rejected, so I'm closing the issue to declutter the tracker. But if y'all change your minds — reopen/ping me and I can re-add this into my backlog then.