@georgefst it would be great to merge this + release 0.5.1 to fix the regression quickly

I'm reluctant to merge this as-is, since unfortunately while it fixes one regression, in a sense it introduces a new one. I, for one, am already using Fourmolu 0.5 and would rather avoid the churn that this would entail. I also don't quite understand all of your concerns, so I could do with a little explanation.

With --indentation=4, --comma-style=trailing and --comma-style=leading will indent the two comments differently:

I don't really see why this is an issue? Why should different options be consistent in this way?

But in addition to this, using --indentation=2, it also causes multiline types to be deindented

That certainly is problematic.

I haven't tested this extensively, but I think we can keep the best of both worlds if we hack in an indentation decrease by changing line 279 to:

mapM_ (inciBy (-2) . (newline >>) . p_hsDocString Caret False) cd_fld_doc

It also shows up with multiline tuples:

I'm unsure what I'm supposed to be looking at here. I think the first output is a bit ugly, but that isn't affected by removing or re-adding sitcc.

With --indentation=4, --comma-style=trailing and --comma-style=leading will indent the two comments differently:

I don't really see why this is an issue? Why should different options be consistent in this way?

I guess, personally, I would expect that the comments, regardless of being -- | or -- ^, should always be either at the same indentation level as the initial { or indented past {. But I guess I do recognize this is more on the subjective end, so I won't push much more here.

But in addition to this, using --indentation=2, it also causes multiline types to be deindented

That certainly is problematic.

I haven't tested this extensively, but I think we can keep the best of both worlds if we hack in an indentation decrease by changing line 279 to:

mapM_ (inciBy (-2) . (newline >>) . p_hsDocString Caret False) cd_fld_doc

I can try that

It also shows up with multiline tuples:

I'm unsure what I'm supposed to be looking at here. I think the first output is a bit ugly, but that isn't affected by removing or re-adding sitcc.

If you look at the first example, the do block is not indented in the second element of the tuple. it's indented in every other example

If you look at the first example, the do block is not indented in the second element of the tuple. it's indented in every other example

Sure, but this output isn't affected by the changes in #124. It appears to be a separate issue. Is it a regression? If so, we should try to bisect.

ah you're right... i'm not sure why i connected that issue to this. i'll move that discussion to a separate branch

@georgefst Fixed! Tested that the below does not reformat with --indentation=2 --comma-style=leading

newtype App a = App
  { unApp ::
      JobExecutorT
        ( ReaderT
            MyConfig
            (LoggingT IO)
        )
        a
  , a :: Int
  -- ^ asfd
  , b :: Bool
  -- ^ asdf
  }

I didn't add this as a test yet, because I'm not sure what we want to do about testing (did you get my latest email?). I'm personally of the mindset that data/examples/ should ONLY contain ormolu test cases, and any test cases fourmolu adds should be in a separate directory like data/examples-fourmolu/, to make it clear what tests are upstream and what tests we added.

@georgefst Thanks! I'll be making a release shortly