Avoid CPP in Config.hs

[fendor]

In haskell/lsp#360, compatibility with aeson >= 2 is achieved without using CPP, we want that too, because no one likes CPP!

[fendor]

Rewriting should be possible, however, the parsers of Config.hs are hard to follow, so this might be an opportunity to rewrite the parsers.

[ThomasCrevoisier]

👋 I can try to give a shot at this :) Though I would need some pointers before doing so.

From what I see, the only impacted points are parsing HashMap and the use of KeyMap in parseSingleOrMultiple, right ?

however, the parsers of Config.hs are hard to follow, so this might be an opportunity to rewrite the parsers

Do you have something specific in mind ?

[fendor]

Hello! Happy to hear it!

True, I think so as well, these are the only locations where we access the Keys.
To be frank, I am not fully sure right now, whether we can avoid all CPP, but any reduction would be great!

Do you have something specific in mind ?

I feel like we have a number of code smells, for example, we dispatch manually the cradle type in https://github.com/haskell/hie-bios/blob/master/src/HIE/Bios/Config.hs#L158.

Then the individual parsers feel a little off as well, take https://github.com/haskell/hie-bios/blob/master/src/HIE/Bios/Config.hs#L203 where we explicitly check the size of the Aeson object to determine what values are possible and provide a proper error message for the user (e.g. when we fail to parse a hie.yaml). Maybe we can improve this parser? Or can at least generate the same error message without hard-coding the available values?

In general, everything you would consider a code-smell can and should be re-worked!

The structure and code of the types CradleConfig and CradleType can be changed if you think it improves the code-quality.

If you are not sure about something, you can either ask here or ping me in irc/matrix (libera.chat #haskell-language-server, I have the same name)

[fendor]

Closing, has been fixed by #329