[OCaml MOOC] week6: MODULES AND DATA ABSTRACTION

this week: programming-in-the-large using the module system of OCaml.

1. STRUCTURING SOFTWARE WITH MODULES

in large project: mangage high number of definitions → abstractions built on top of other abstractions.

  • layers of abstractions: hide information
  • divide program into components
  • identifiers organised to avoid naming conflicts

module as namespace

dot-notation: access module component.

ex. List.length
or first open List then just call length
if open 2 modules having identical identifiers, the last opened module will be used.

to define a module:

module SomeModuleIdentifier = struct  
  (* a seq of definitions *)  
end  
  • module name: start with an upper case
  • to alias a module: module SomeModuleIdentifier = SomeOtherModuleIdentifier
# module Stack = struct  
  type 'a t = 'a list  
  let empty = []  
  let push  x s = x::s  
  let pop = function   
    | [] -> None  
    | x::xs -> Some (x,xs)  
end;;  
              module Stack :  
  sig  
    type 'a t = 'a list  
    val empty : 'a list  
    val push : 'a -> 'a list -> 'a list  
    val pop : 'a list -> ('a * 'a list) option  
  end  
# let s = Stack.empty;;  
val s : 'a list = []  
# let s = Stack.push 1 s;;  
val s : int list = [1]  
# let x,s =   
  match Stack.pop s with   
    | None -> assert false  
    | Some (x,s) -> (x,s);;  
val x : int = 1  
val s : int list = []  
# let r = Stack.pop s;;  
val r : (int * int list) option = None   

hierachical module structure

  • a module can contain other module definitions
  • a signature can also contain module signatures
  • if module B is inside A, use A.B to get its namespace
module Forest = struct  
  type 'a forest = 'a list  
  module Tree = struct  
    type 'a tree =   
        Leaf of 'a   
        | Node of 'a tree forest  
  end   
end;;  
open Forest.Tree;;  
let t = Leaf 42;;   

2. INFORMATION HIDING

a module should come with some user manual ("contract") to indicate to clients:

  1. function preconditions that must be verified
  2. data invariants that must be preserved
  3. definitions that user must not rely on (cause they'll change in the future)

a module signature represents this contract, the type checker will enforce point 2 and 3.

module signatures

  • a module's type is called signature or interface
  • programmer can force a module to have a specific signature

to define a signature:

module type sig   
  (* a seq of declarations of following form:*)  
  val some_identifier: some_type  
  type some_type_identifier = some_type_definition  
  exception SomeException of some_type  
end   

to construct a module with a specific signature:

module M: sig ... end = struct ... end

to name a signature:

module type S = sig ... end

then use this name to annotate module:

module M:S = struct ... end

example: natural numbers

module Naturals: sig  
  (* Invariant: A value of type t is a positive integer *)  
  type t = int  
  val zero: t  
  val succ: t -> t  
  val pred: t -> t  
end = struct  
  type t = int  
  let zero = 0   
  let succ n = if n=max_int then 0 else n+1  
  let pred = function   
 |0 -> 0   
 | n -> n-1  
end ;;   

abstract types

we can use the module normally:

open Naturals;;  
let rec add: t -> t -> t =   
fun x y ->   
if x = zero then y else succ (add (pred x) y);;   

but the invariant can be easily broken:

let i_break_the_abstraction = pred (-1);;   

This don't have compiler error, as the type of pred is int, we can pass any int to it.

⇒ use abstract types that will give no choice to the client but to respect the rule.

in the signature:

module Naturals: sig  
  (* Invariant: A value of type t is a positive integer *)  
  type t (* remove the type value of t in the signature *)  
  val zero: t  
  val succ: t -> t  
  val pred: t -> t  
end   

then calling pred (-1) will cause an error.

→ we have hiddent the definition of the type t
the sig don't publish t's implementation anymore, so the checker ensures clients can't use that fact
t is called an abstract type.

With abstract type, users can't do pattern matching, to allow pattern matching while forbidding the direct application of data constructors, OCaml provides a mechanism called private types. see here.

3. CASE STUDY: A MODULE FOR DICTIONARIES

An example of using abstract types to increase the modularity of programs.

Define a dictionary signature:

module type DictSig = sig  
  type ('key, 'value) t = ('key*'value) list (* internal repr of the dict is exposed@ *)  
  val empty: ('key, 'value) t  
  val add: ('key, 'value) t -> 'key -> 'value -> ('key, 'value) t  
  exception NotFound  
  val lookup: ('key, 'value) t -> 'key -> 'value  
end;;  

module Dict: DictSig = struct  
  type ('key, 'value) t = ('key*'value) list  
  (*......implementation *)  
end;;   

Then a client can use this module:

module ForceArchive = struct   
  let force = Dict.empty  
  let force = Dict.add force "luke" 10  
  let force = Dict.add force "yoda" 100  
  let force_of_luke = Dict.lookup force "luke"  
  let all_jedis = List.map fst force (* here client knows that dict is a list!*)  
end;;   

This is not very good if the internal implemtation of Dict is changed into

For instance, change the implemention into a BST:

type ('key, 'value) t =   
  | Empty  
  | Node of ('key, 'value) t * 'key * 'value * ('key, 'value) t  

→ change the signature of module to abstract type.

4. FUNCTORS

Functors are functions from modules to modules. In other words, a functor is a module parameterized by another module.

Continue the last example, we want to choose a Dict implementation externally. → module functor

To declear a functor, add the Dict module in the parameter

module ForceArchive (Dict: DictSig) = struct   
  let force = Dict.empty  
  let force = Dict.add force "luke" 10  
  let force = Dict.add force "yoda" 100  
  let force_of_luke = Dict.lookup force "luke"  
end;;   

Then we can call the explicit implementation in the client:

module Dict_list: DictSig = struct   
...  
end;;  
module Dict_bst: DictSig = struct  
...  
end;;  

module Client1 = ForceArchieve (Dict_list)  
module Client2 = ForceArchieve (Dict_bst)   
  • a functor is a module waiting for another module
  • syntax:

module SomeModuleIdentifier (SomeModuleIdentifier: SomeSignature) = struct ... end;;

  • to apply a functor to a module: SomeModuleIdentifier (SomeModule)
  • signature of a functor:

functor (ModuleIdentifier: SomeSignature) -> sig ... end

example: Set and Map

They expects a module satisfying the following signature:

module type OrderedType = sig  
  type t  
  val compare: t -> t -> int  
end   

Once a module E has this signature,

  • Set.Make (E) offers over sets of E.t elements
  • Map.Make (E) ....

type parameterization of exception can be done using functor.

In signature declaration:

module type DictSig = sig  
  (* before: type ('key, 'value) t *)  
  type key (* make the key not polymorphic *)  
  type 'value t  
  val empty: 'value t  
  val add : 'value t -> key -> 'value -> 'value t  
  exception NotFound of key (* parameterize exception type in signature *)  
  val lookup: 'value t -> key -> 'value  
end;;   

In the implementation make the module a functor: add the Key module as argument

module Dict(Key: sig   
  type t  
  val compare: t -> t -> int  
end) : DictSig = struct   
  type key = Key.t (* key is the type of the Key module *)   
  ...  
end;;   

(... don't quite get it......)

type constraint: DictSig with key=string

comments powered by Disqus