shebang lines are lost in exactprint

[kokobd]

In hls, the exactprint doesn't preserve lines starting with #!. I've written a failing test case for this in #2725.

Your environment

Which OS do you use:

Ubuntu 20.04

Which LSP client (editor/plugin) do you use:

VSCode

Describe your project (alternative: link to the project):

hls itself

Compiler Version:

Currently checked on ghc 8.10.7

Steps to reproduce

Given this code:

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

class Semigroup a => SomeData a
instance SomeData All

We can place the cursor on All and select code action "Add All to the import list of Data.Monoid", and see how the file content changes.

Expected behaviour

This is the correct code after extending the import, without losing the shebang line.

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

class Semigroup a => SomeData a
instance SomeData All

Actual behaviour

This is what we are getting now.

import

    Data.Monoid ( Monoid, All )

class Semigroup a => SomeData a
instance SomeData All

[eddiemundo]

I think it's only really legit to have the shebangs as the first line of the file, but maybe preserving everything even stuff that makes no sense is worth.

Not sure if it's related to what you're trying to do but I wrote a parser that finds the last pragma location before the first declaration or module statement that uses the ghc lexer. I think with ghc 9.2 annotation stuff some of it can be deleted but not sure.

https://github.com/haskell/haskell-language-server/blob/master/ghcide/src/Development/IDE/Spans/Pragmas.hs

[kokobd]

I found this when working on #2694, trying to exact print the whole module. We have existing test cases for shebang lines in the middle of a source file. That's why I found this issue.

parsePreDecl looks better than ghc-exactprint's naive parser, but it may require some modifications to be applicable in this use case.

[michaelpj]

@alanz do you know if this is still the case or if it ever got fixed?

[alanz]

The changelog says it was fixed in ghc-exactprint 0.5.6, courtesy of @lspitzner