Sat, 24 May 2008

Objects vs Modules.

Although I've been using Ocaml for a several years now, I've not yet been in a situation where I've needed to write an Ocaml class to define a C++/Java/Python/Smalltalk/OO style object. I've found that most of the problems I encountered could be easily solved using functional code and that Ocaml's objects didn't provide an obviously better solution. Until now (or so I thought).

The problem was one of moving around the filesystem keeping track of the old directories so they were easy to return to. The obvious model for this was the pushd and popd built-ins in command shells like GNU Bash. This functionality can be easily wrapped up in an Ocaml object as in the following example and demo code (which needs to be linked to the Unix module): 


  class dirstack = object
      val mutable stack = []

      method push dirname =
          (* Find the current working directory. *)
          let cwd = Unix.getcwd () in
          (* Change to the new directory. *)
          Unix.chdir dirname ;
          (* If successful, push old cwd onto the stack. *)
          stack <- cwd :: stack

      method pop () =
          match stack with
          |    [] -> failwith "Directory stack is empty."
          |    head :: tail ->
                  Unix.chdir head

  	end

  let () =
      print_endline (Unix.getcwd ()) ;
      let dstack = new dirstack in
      dstack#push "/tmp" ;
      print_endline (Unix.getcwd ()) ;
      dstack#push "/bin" ;
      print_endline (Unix.getcwd ()) ;
      dstack#pop () ;
      print_endline (Unix.getcwd ()) ;
      dstack#pop () ;
      print_endline (Unix.getcwd ())

However, there are some problems with the above code. Firstly, if the push and pop methods need to be used throughout the program, the dstack object needs to be made more widely accessible using one of the following three methods:

  1. Being placed in the global scope.
  2. Being made into a Singleton objecct.
  3. Being passed around as a parameter to whatever function may need it.

Yuck! Yuck! Double yuck! Suddenly, this object oriented solution didn't look like such a great idea.

Then it struck me. This object can be easily transformed into an Ocaml module like this: 


  module Dirstack = struct
      let stack = ref []

      let push dirname =
          (* Find the current working directory. *)
          let cwd = Unix.getcwd () in
          (* Change to the new directory. *)
          Unix.chdir dirname ;
          (* If successful, push old cwd onto the stack. *)
          stack := cwd :: !stack

      let pop () =
          match !stack with
          |    [] -> failwith "Directory stack is empty."
          |    head :: tail ->
                  stack := tail ;
                  Unix.chdir head

     end

  let () =
      print_endline (Unix.getcwd ()) ;
      Dirstack.push "/tmp" ;
      print_endline (Unix.getcwd ()) ;
      Dirstack.push "/bin" ;
      print_endline (Unix.getcwd ()) ;
      Dirstack.pop () ;
      print_endline (Unix.getcwd ()) ;
      Dirstack.pop () ;
      print_endline (Unix.getcwd ())

This solution using a module is much better than the one using an object. The Dirstack module itself is globally accessible and is already a singleton while the stack used to hold past directories is implemented as a list whose scope is limited to the module itself. (Furthermore, if Dirstack is implemented in its own file instead of using a module defined within a larger file, then the stack variable can be hidden completely by not listing it in the Dirstack interface file.)

So while I'm pleased with this solution, it does mean that I'll have to continue my hunt for a problem where an object provides a better solution than any other feature of the Ocaml language. This is particularly ironic because when choosing between two strict statically typed languages, Haskell and Ocaml, I chose Ocaml because I thought I needed objects. However, I stuck with Ocaml because of its pragmatism.

Posted at: 07:45 | Category: CodeHacking/Ocaml | Permalink