Diff displays hunks incorrectly in some cases

[mrkkrp]

When diffing these two inputs:

testPermParser :: Permutation Parser String
testPermParser =
  f <$> toPermutationWithDefault 'x' (char 'a')
    <*> toPermutationWithDefault 'y' (char 'b')
    <*> toPermutationWithDefault 'z' (char 'c')
  where
    f a b c = [a, b, c]

and

testPermParser :: Permutation Parser String
testPermParser =
  f
    <$> toPermutationWithDefault 'x' (char 'a')
    <*> toPermutationWithDefault 'y' (char 'b')
    <*> toPermutationWithDefault 'z' (char 'c')
  where
    f a b c = [a, b, c]

The result is

@@ -148,5 +148,6 @@
  testPermParser :: Permutation Parser String
  testPermParser =
-   f <$> toPermutationWithDefault 'x' (char 'a')
+   f
+     <$> toPermutationWithDefault 'x' (char 'a')
    where
      f a b c = [a, b, c]

which is not correct. Instead of the line with where it is the line with <*> toPermutationWithDefault 'y' (char 'b') that should follow. Proper diff reports:

@@ -147,7 +147,8 @@ prsp' p = prs' (runPermutation p)

 testPermParser :: Permutation Parser String
 testPermParser =
-  f <$> toPermutationWithDefault 'x' (char 'a')
+  f
+    <$> toPermutationWithDefault 'x' (char 'a')
     <*> toPermutationWithDefault 'y' (char 'b')
     <*> toPermutationWithDefault 'z' (char 'c')
   where

We should match that.

This seems to have been broken ever since I introduced the functionality for nice displaying of diffs.

[mrkkrp]

I'm surprised that this did not come up earlier and I did not notice it before...

[amesgen]

I'm surprised that this did not come up earlier and I did not notice it before...

Personally, I think I never looked at the diffs produced by Ormolu more than in a binary "oh, it is not yet formatted" fashion, but rather used the diffing of the git project I was formatting afterwards for inspection. One possibility would be to just shell out to git diff if available, and don't display diffs otherwise, but the diff printing code in Ormolu is not that large/complex that this seems like a priority.

[mrkkrp]

git diff approach is what was used before. I think there is a problem with it though, git diffing and Ormolu diffing are not always the same, in particular, they are not the same in the presence of intermediate editing. Actually IIRC that was the reason for adding built-in diffing in Ormolu: if I wanted to check on CI with Ormolu I had to use git diff, but my script also did some other editing and it interfered with the results.

[amesgen]

One could use git diff --no-index to just use the diffing part and ignoring any actual git-related functionality.