Working around ghc’s lexer’s layout rule

While implementing coloring for Haskell files I noticed that lines with more closing braces (either ‘)’ or ‘}’) were not being colored.

After doing some digging around I found out the following:











So apparently they were throwing lexical errors, but why?

After contacting Mr. Simon Marlow I was told that this is the handling of GHC’s layout rule. to quote

“You’re probably encountering the lexer’s handling of the Haskell “layout” rule.  When the lexer sees a ‘}’ token, it pops the current layout stack, and if the layout stack is empty then this is a lexical error.”

This left me with 3 choices

  1. Use a custom lexer much like the original visual haskell did
  2. Replace all {, },( and ) with 1 whitespace character so that they won’t be colored, but the rest of the input will, but the positions would be preserved.
  3. left pad the input with enough opening braces to have the lexer succeed in parsing then adjust the ranges.

Option 1 was the least maintainable, since I would have to keep updating the lexer everytime the one in ghc changes. So I didn’t want to do this.

Option 2 was a possibility, one which I tried out before, But I noticed that having the braces colored really did help.

Option 3 was then chosen by process of elimination. It turned out to not be that much work at all.

private int prepareLine(ref string str)
    int round= 0, brace = 0;

    for (int i = 0; i < str.Length; i++)
        switch (str[i])
            case ‘}’:
                if(i==0 || !(str[i-1]==’-‘))
            case ‘)’:

    if(round > 0)
        str = str.PadLeft(str.Length + round, ‘(‘);

    if (brace > 0)
        str = str.PadLeft(str.Length + brace, ‘{‘);

    return round + brace;


is the full implementation. Now I know what you’re thinking, By doing this I’ll create more opening than closing braces. So a balanced line like (Int) becomes unbalanced ((Int). However this is not a problem, Since for my coloring braces carry no semantics. I don’t care what they mean (as in, when interpreted) all I care about is what they are (as in the token type).

With that in place, the only other code needed is to skip the first n number of tokens returned from the lexer, where n is the result of calling the prepareLine function.

And that’s all, Now we have perfect line coloring everywhere🙂