Emacs/Indentation

Indentation Approaches

Emacs can indent Haskell in various ways. The most common is the tab cycle.

Indentation using tab cycle

Haskell-mode offers intelligent indentation. As Haskell source code uses indentation aware code blocks, there is usually more than one column for which indentation makes sense.

Hit tab a few times to see a few different indentation possibilities.

For example, imagine the following is open in a haskell-mode buffer, where ! represents the point:

foo :: Int -> String
foo 0 = f 4 ++ s
  where f 4 = "hello" ++ 
!

If you ask haskell-mode to indent for you, where should it indent to? There are four basic options:

1. You want to finish off the expression you were writing in the last line. Haskell-mode indents to be underneath the " character at the beginning of "hello":

   where f 4 = "hello" ++
               !
   

   This is debatably a bad choice as you'd probably want to indent a bit further in to make it clear that you were carrying on an expression, but the layout rule would accept something like the following:

   where f 4 = "hello" ++
               "world"
   

2. You want to add a second equation for f. Haskell-mode will indent to line up with the first argument, and fill in the f in the equation:

   where f 4 = "hello" ++
         f !
   

   This is an unlikely choice as the expression in the previous line isn't complete, but haskell-mode isn't smart enough to know that. (If f had been something without arguments, like where f = "hello", then it's impossible to have more than one equation and haskell-mode won't offer this indentation level.)

3. You want to add a second binding to the where-block. Haskell-mode indents to line up with the f:

   where f 4 = "hello" ++
         !
   

4. You want to start an entirely new top-level binding. Haskell-mode indents to the first column:

   foo :: Int -> String
   foo 0 = f 4 ++ s
     where f 4 = "hello" ++
   !
   

These four locations can be reached by repeatedly pressing TAB. This is what's known as the tab-cycle. The innermost location is offered first, then cycling progresses outwards. Although this may seem like an inefficient system (and it is indeed a shame that Haskell's design didn't result in an unambiguous indentation system), you do quickly get used to the tab-cycle and indenting Haskell code.

Notes:

Do not use indent-region

Using indent-region is generally a bad idea on Haskell code, because it would need to know which of the tab-cycle stops you wish to choose for each line. The innermost one is chosen in each case, which often results in unusable code. Moral: just don't use indent-region with haskell-mode.

Using rectangular region commands

Emacs has a set of commands which operate on the region as if it were rectangular. This turns out to be extremely useful when dealing with whitespace sensitive languages.

C-x r o is "Open Rectangle". It will shift any text within the rectangle to the right side. Also see:

C-x r t is "String Rectangle". It will shift any text within the rectangle over to the right, and insert a given string prefixing all the lines in the region. If comment-region didn't already exist, you could use this instead, for example.

C-x r d is "Delete Rectangle". It will delete the contents of the rectangle and move anything on the right over.

C-x r r is "Copy Rectangle to Register". It will prompt you for a register number so it can save it for later.

C-x r g is "Insert register". This will insert the contents of the given register, overwriting whatever happens to be within the target rectangle. (So make room)

C-x r k is "Kill rectangle". Delete rectangle and save contents for:

C-x r y is "Yank rectangle". This will insert the contents of the last killed rectangle.

As with all Emacs modifier combos, you can type C-x r C-h to find out what keys are bound beginning with the C-x r prefix.

Aligning code

Emacs22 has a neat tool called: align-regexp. Select a region you want to align text within, M-x align-regexp, and type a regexp representing the alignment delimiter.

For example, I often line up my Haddock comments:

f :: a -- ^ does a
  -> Foo b -- ^ and b
  -> c -- ^ to c

Select the region, and let the regexp be: --

f :: a     -- ^ does a
  -> Foo b -- ^ and b
  -> c     -- ^ to c

Of course, this works for just about anything. Personally, I've globally bound it to C-x a r:

(global-set-key (kbd "C-x a r") 'align-regexp).