rewrite import completion with ghc-exactprint

[kokobd]

This will solve #2414.

TODO List:

• handle the exact case of Minor "import action" formatting bug #2414 properly
• support other existing cases (make tests pass)
• support ghc 9.2
• minimize the TextEdit sent to language client

I will request code review when it's feature complete.

[kokobd]

Both Completions.hs and CodeAction.hs has some code for adding new imports. I will try to eliminate this duplication.

[kokobd]

If we want to know where to insert the new import declaration, we will have to exact print the whole Located HsModule, because the only Located structure containing a Located ImportDecl is the whole module.

Then I encountered a problem (at least for ghc 8.10.7. I haven't tested it on ghc 9.2). In hls, the exactprint doesn't preserve lines starting with #!. As a quick test, you can try to extend import for this code (extending import already uses exactprint, both in master branch and latest release)

import
#! nix-shell --pure -i runghc -p "haskellPackages.ghcWithPackages (hp: with hp; [ turtle ])"
    Data.Monoid ( Monoid )

class Semigroup a => SomeData a
instance SomeData All

If we place the cursor on All and select code action "Add All to the import list of Data.Monoid", then the code becomes

import

    Data.Monoid ( Monoid, All )

class Semigroup a => SomeData a
instance SomeData All

I'm wondering if losing the #! line is intended and having some special considerations, or if it's just a previously unknown bug that need to be fixed.

[pepeiborra]

Then I encountered a problem (at least for ghc 8.10.7. I haven't tested it on ghc 9.2). In hls, the exactprint doesn't preserve lines starting with #!.

I believe ghc-exactprint comes with a round tripping test suite. Have you been able to repro the issue using it?

[kokobd]

@pepeiborra I've tried ghc-exactprint 0.6.4 in a standalone project. It exactprints perfectly. This problem seems to be related to hls (maybe we have some wrong configs for parsing).

I'm still investigating and trying to fix it in hls. However as changing fundamental parts like parsing might potentially affect other parts of hls, any related context information is welcome ;)

[kokobd]

I might have identified the root cause of the shebang exact printing issue. Created another draft PR #2725 to fix it seperately.

[pepeiborra]

@kokobd still around? Would be nice to land this

[kokobd]

Sorry for being absent these days. It has taken me a few months to get rid of my tiresome job.

I will probably fix this in the next few days, but less aggressively. Using exact print info to print the whole module incurs too much risk (performance, compatibility, etc.) and a lot of code change. Instead, using exact print information to locate the where keyword of the module is an easier way to fix this issue.

[kokobd]

Closing in favor of #2981.