Define tiers of plugins for HLS

[michaelpj]

It would be helpful to identify certain HLS plugins as more important or better supported than others. For example:

• We advertise the support for newer GHC versions as "partial", meaning that not all plugins work for that version. But as we get more and more plugins, not all of which have highly active maintainers, "all plugins work" may become an unachievable goal. Indeed, the newly added stan plugin may never support GHCs other than 8.10!
• Well-Typed is dong some work to update HLS for 9.4. We're asking them to continue that so that some plugins are supported... but which ones? It would be nice to have a clear set of "core" plugins that we really care about.

Here's a proposal. We define three tiers of plugin support:

1. Tier 1. These are "core" plugins, without which we don't think HLS is worth using. e.g. completions, hovers. This mostly corresponds to the ghcide plugins today, maybe a few more?
2. Tier 2. These are additional plugins which are well-maintained and quickly get updated to support new versions of GHC.
3. Tier 3. Everything else.

This interacts with GHC support as follows:

• We should not release a version of HLS supporting GHC X unless all Tier 1 plugins work.
  • All Tier 1 plugins work = "basic support for GHC X"
• We should try not to release a version of HLS supporting GHC X unless all Tier 2 plugins work, but we can if we need to.
  • All Tier 2 plugins work = "support for GHC X"
  • If a plugin fails to meet the bar for some period of time we should demote it to Tier 3.
• Tier 3 plugins are supported at a best-effort basis depending on their maintainers.
  • We don't say anything different about our support for GHC X based on which Tier 3 plugins work.

Tentative tier lists:

• Tier 1
  • ghcide plugins
  • hls-call-hierarchy-plugin
  • hls-code-range-plugin
  • hls-pragmas-plugin
  • hls-explicit-imports-plugin
• Tier 2
  • hls-alternate-number-format-plugin
  • hls-chnage-type-signature-plugin
  • hls-class-plugin
  • hls-eval-plugin
  • hls-module-name-plugin
  • hls-hlint-plugin
  • hls-ormolu-plugin
  • hls-fourmolu-plugin
  • hls-stylish-haskell-plugin
  • hls-qualify-imported-names-plugin
  • hls-refine-imports-plugin
  • hls-rename-plugin
  • hls-gadt-plugin
• Tier 3
  • hls-brittany-plugin
  • hls-haddock-comments-plugin
  • hls-stan-plugin
  • hls-retrie-plugin
  • hls-splice-plugin
  • hls-tactics-plugin

My rubric here was basically: if it provides a relatively core LSP feature and it's fairly stable, it goes in Tier 1; if it's not currently up-to-date with the latest GHC we support, then clearly that's hard and it goes in Tier 3; everything else in Tier 2.

Thoughts?

[drsooch]

hls-change-type-signature-plugin and alternate-number-format-plugin can go in tier 2?

I would think refine-imports should be a Tier 1 feature, it seems like it would be the dual to explicit-imports, just my opinion

[michaelpj]

I think refine-imports is just not quite so established at the moment.

[pepeiborra]

I would bump hls-eval-plugin to Tier 1, under the premise that that's where it belongs even if its currently half-working or under-maintained. What do others think?

[michaelpj]

I think it's better for the tiers to reflect reality rather than aspiration. Otherwise we wouldn't be able to say we had more than basic support for 9.2, just because the eval plugin is not going anywhere. I think that's more misleading than the alternative, which is to say that the eval plugin is tier 2 but we'd like it to be tier 1. That suggests some concrete actions: can we find more maintainers, can we make it easier to upgrade?

Personally, I wouldn't be surprised if the eval plugin stays stuck in tier 2 since it's pretty complex and not so core to the HLS experience.

[michaelpj]

I guess I was also thinking that there is also some kind of difficulty modifier. We don't want things in tier 1 that are (in practice) difficult to update, even if they're very nice features.

[pepeiborra]

to say that the eval plugin is tier 2 but we'd like it to be tier 1 ... suggests some concrete actions: can we find more maintainers, can we make it easier to upgrade?

Good point, agreed

[July541]

We added hls-explicit-fixity-plugin recently, I think it belongs to tier 2.

[georgefst]

I'm strongly in support of this, as I've seen multiple examples (e.g. today on Reddit) of people assuming that GHC 9.2 isn't supported, despite it being so almost-fully for months.

[michaelpj]

PR: #3189