There are two places where 'Illegal token' can be triggered in haskell-indentation:

1. In haskell-indentation-with-starter: the case there seems to be when a lexical structure expect a closing token/keyword and the token is not present:

(defun haskell-indentation-with-starter (parser &optional end where-expr?)
 ...
 (cond 
   ...
      (end (haskell-indentation-parse-error
             "Illegal token: %s" current-token))))))

The expected token at this point is one of closing ')', ']', '}'. I do not see why we could not eat everything up to a closing token, maybe treating layout as taking priority.

1. In haskell-indentation-parse-to-indentations: top level problem.

It seems to be a variation on the same, as top level list is a layout delimited list.

Solution seems to be:

1. React to leading keywords only if, let, import, case etc.
2. React to following keywords only in context of a leading keyword, so then and else only as part of if.
3. Allow partial statements, that is if that is not followed by then