wz1000

I think you can also configure stack to download the 9.0.2a version manually if it doesn't automatically do so: https://docs.haskellstack.org/en/stable/yaml_configuration/#setup-info-locations Note that if you do this, please delete the cached/installed...

Looks like the `aarch64-darwin` build job silently failed to produce binaries for 9.0.2: https://gitlab.haskell.org/haskell/haskell-language-server/-/jobs/1168557 I'll fix the ci job so this doesn't happen in the future. Meanwhile, could someone try...

I've got a patch for this which I just implemented today. Sorry for not updating the ticket!.

I think this is triggered by hlint. I can't reproduce this with just ghcide.

I've also disabled the `call-hierarchy` plugin tests on windows.

@pepeiborra you might want to look the changes to `showAstData` in `ghcide/src/Development/IDE/GHC/Dump.hs` and see if you are comfortable with those. We could add a flag to use exactprint for that...

Also the async tests now use code lenses instead of code actions, I believe this still preserves the expected test behavior, is that correct @pepeiborra?

@pepeiborra I've also replaced code actions in the benchmarks with code lenses. This is unfortunate, but we should fix this in the future by benchmarking the entire IDE with plugins...

> I don't _love_ the idea of dynamically looking up another plugin just in order to get access to some helper code. I think that could do with some clear...

All the plugins that use the `GetAnnotatedParsedSource` rule now have to link it in via `pluginRules` to ensure they have access to it even if hls-refactor-plugin is disabled.