Draft: Support add to where command

[santiweight]

This MR is still in progress because comments are handled in a breaking way and I'm not totally sure why yet...

This MR is based on the add-argument MR that is merging soon.

Here are a few examples:

foo True = _bar [True]
foo False = False
-- becomes
foo True = _bar [True]
  where
    _bar = _
foo False = False

---------------

foo True = _bar [True]
  where
    baz = 1
-- becomes
foo True = _bar [True]
  where
    _bar = _
    baz = 1

-----------------

foo True = _bar [True]
  where baz = 1
-- becomes
foo True = _bar [True]
  where
    _bar = _
    baz = 1

[santiweight]

@alanz This is a function that I wrote in order to do "the right thing" when inserting a new declaration at the start of a where clause. insertAtStart wasn't doing the right thing in a few situations, such as the following:

-- insert a "bar" decl with noLocA

foo = 1
-- becomes
foo = 1
  wherebar = 1

foo = 1
  where 
    baz = 1
-- becomes
foo
  wherebar = _ baz = 1

Inserting an L (... DifferentLine 1 0 ...) decl causes a whole different type of issues:

foo = 1
  where
    baz = 1
-- becomes
foo = 1
  where
    bar = 1
      baz = 1

Also, it appears that the insertAtStart from Exactprint does the wrong thing for comments (I've tagged you in another comment on the test that shows the problem)

[santiweight]

@alanz This test case (foo is the before, foo' is the after), shows how comments appear to cause problems when using insertAt from Exactprint. I tried some stuff to make it work but to no avail...

The issue persists even when using insertAt from exact print without doing anything except returning the old decls:

foo = 1
  where
    -- hi
    baz = 1
-- becomes
foo = 1
  where
____
    -- hibaz = 1

-- NOTE used 4 underscores to indicate 4 space characters

[alanz]

Comments are tough.

Which version of ghc-exactprint?

[alanz]

@santiweight you should be using the high-level API as demonstrated in the various addLocalDeclX tests from https://github.com/alanz/ghc-exactprint/blob/master/tests/Test/Transform.hs#L318

[alanz]

And for current ghc-exactprint master, I reproduced your example with a changer test using

addLocaLDecl7 :: Changer
addLocaLDecl7 libdir lp = do
  let
      doAddLocal = do
        decls0 <- hsDecls lp
        replaceDecls lp decls0

  (lp',_,_w) <- runTransformT doAddLocal
  debugM $ "log:[\n" ++ intercalate "\n" _w ++ "]log end\n"
  return lp'

And test file

module AddLocalDecl7 where

foo = 1
  where
    -- hi
    baz = 1

[santiweight]

Thanks for these examples @alanz

I realize now that the newer versions of ghc-exactprint are not available to me until 9.4.x is supported in HLS...

I am using 9.2.4 for the refactoring actions I have been working on up until now, which I believe corresponds to ghc-exactprint 1.6.

I am thinking it might be best to hold out for newer versions of ghc-exactprint to be available with 9.4 and then use newer versions of the API then... Alternatively, I might try again to use the exact-ring available in HLS and see if the results can be better than what I've been producing before I saw these examples in the ghc-exactprint repo!

[michaelpj]

9.4 is supported in HLS

[santiweight]

WIP here: https://github.com/alanz/ghc-exactprint/pull/116/files