Typemap reimplementation

[fendor]

Under the orchestration of @mpickering, reimplements the genTypeMap function in a new module.
A Typechecked module is only traveresed once.
This reimplementation should have a lower memory footprint but is likely to increase the first startup time of modules.
Some parts may no longer have a type on hover.

Implementation has already been merged into #1126. There it is said to be working!

• More Documentation
• Way more tests

Closes #1156

[fendor]

@mpickering Anything missing for this PR? Or do you think we should rather wait for your PR to be merged?

[mpickering]

You should quickly check the ghc-mod source to see if there are any other hover points it supports so this doesn't regress behaviour?

Also, are there any tests for hover? It would be good to add some if there aren't but it shouldn't block merging this MR.

[fendor]

You should quickly check the ghc-mod source to see if there are any other hover points it supports so this doesn't regress behaviour?

Will do, thanks!

Also, are there any tests for hover? It would be good to add some if there aren't but it shouldn't block merging this MR.

I think there are some tests, but I will investigate

[mpickering]

Thanks for this @fendor the patch looks really good now. The tests are a big improvement as well!

[mpickering]

Can just have the else clause here I think.

[fendor]

Yeah, I wanted to ask that anyway, do we prefer easy to read, e.g. always three clauses although they often completely overlap, or conciseness when it comes to CPP?

[mpickering]

I prefer not having the duplication and just having two clauses when possible.

[fendor]

When the tests run through, this should be ready to merge, imo.

[alanz]

Given what a big change this is, I suggest we merge it after the April release, which is due in a few days.

[mpickering]

CI looks to be failing still on 8.2.2 and 8.2.1.

@alanz Are you sure you don't want to merge this? It improves one of the big complaints about HIE which is high memory usage.

[fendor]

@mpickering Tests fail for the type information for deriving clauses. Perhaps the code generation API has changed over one GHC version to the next?
Should the test case be solved via CPP?

However, failing tests are unrelated to the feature itself.

[fendor]

CI fails are imo unrelated to this PR.

[mpickering]

This can be merged now I think?

[alanz]

I think so, I presume the CI is transient.

[alanz]

In other words, I guess I should actually take a look at it now.