moar articles
Follow @nicferrier

Using Polymorphism as a Lisp refactoring tool

Sometimes I get stuck with a function that I want to change the contract of and I can't because Lisp doesn't have an expressive enough way of doing it. Because I practice iterative development this happens to me a surprising amount of the time. It can be hell to fix.

EmacsLisp can do quite flexible argument definitions already, through the 'cl library extension defun*.

The trouble is that it's hard to add optional information to an existing function contract.

Say you have a function like this:

(defun test1 (arg1 &optional arg2)
   (list arg1 (or arg2 default-value)))

That's great. But now say you don't want to that contract, but just add the ability to add a third optional parameter. The existing optional parameter should still be optional and you should be able to specify either of them or neither of them.

It starts to sound like you need keyword args. But this wouldn't work:

(defun* test1 (arg1 &optional arg2 &key (arg3 arg3-default))
  (list arg1 (or arg2 default-value) arg3))

Because an &key arg cannot be used after an &optional argument.

Switching to keyword args would definitely work. If you can refactor all your code to support keyword args maybe that's the right thing to do.

But maybe you don't want to change that contract. There is still a way to go ahead: pattern matching the args use pcase.

Here's the solution:

(defun test1 (&rest args)
  (pcase args
   (`(,arg1 :arg3 ,arg3 . ,rest)
    (destructuring-bind (arg1 &optional arg2) (cons arg1 rest)
      (list arg1 (or arg2 default-value) (or arg3 arg3-default))))
   (`(,arg1 . ,rest)
    (destructuring-bind (arg1 &optional arg2) (cons arg1 rest)
      (list arg1 (or arg2 default-value) arg3-default)))))

(let ((default-value 6)
      (arg3-default 19))
  (test1 20 :arg3 11 7)) => (20 7 11)

I admit, it looks a bit wierd having the pattern match drop directly into the destructuring-bind but that's the best way of processing the args. And the fact that the function body is duplicated in both these patterns is unusual, often I've found I want variation in the function body.

Still, it makes me wonder if some sort of combined definition thing of pattern matching and arguments would be useful or whether just using patterns more is the right thing. Certainly it seems like a defun-pcase would be very useful.

Nota Bene - I am not really into ML or Haskell and therefore have never really exercised the pattern thing.