Skip to content

Latest commit

 

History

History
90 lines (73 loc) · 3.12 KB

interface-first.md

File metadata and controls

90 lines (73 loc) · 3.12 KB

Interface First

Sketch your module by writing a small interface file that includes only the minimum things you need the module users to know.

This is tedious, yes. But it will lead to better thought through modules that expose only the things that are needed.

For example, if you are building software for a bakery, you will need to keep track of stock for something like loafs of bread. You could expose it as:

module Bakery = {
  type bread = | Baked | Prepared | Ordered;
  type stock = list(bread);
};

But now whoever consumes Bakery.stock can rely on it being a list of bread. Which is perfectly fine if you run a small bakery that doesn't need superb performance or has a large inventory!

Now take a look at what happens when limit this interface:

module Bakery : {
  type bread = | Baked | Prepared | Ordered;
  type stock;
} = {
  type bread = | Baked | Prepared | Ordered;
  type stock = list(bread);
};

Sure there's a little more duplication (the type bread redefinition), but now as a consumer of Bakery.stock I have no clue what stock is under the hood. I simply can't know.

This means that if the bakery starts running faster and needs to switch the implementation from a list to a Hashtbl or perhaps just an Array, it's completely possible to do so without hurting the users.

This means that you'll need a few more functions, but they are usually cheap to write:

module Bakery : {
  type bread = | Baked | Prepared | Ordered;
  type stock;
  let has_in_stock : (bread, stock) => bool;
  let add : (bread, stock) => stock;
} = {
  type bread = | Baked | Prepared | Ordered;
  type stock = list(bread);
  let has_in_stock = List.mem;
  let add = List.cons;
};

Notice how operations also become more semantic all of a sudden! has_in_stock is much more meaningful for your module users than List.mem.

Switching the implementation from a List to a Hashtbl is then a very self-contained change:

module Bakery : {
  type bread = | Baked | Prepared | Ordered;
  type stock;
  let has_in_stock : (bread, stock) => bool;
  let add : (bread, stock) => stock;
} = {
  module Stock = Hashtbl.Make({
    type t = string;
    let equal = String.equal;
    let hash = Hashtbl.hash;
  });
  type bread = | Baked | Prepared | Ordered;
  type stock = Stock.t(bread);
  let has_in_stock = (bread, stock) => Stock.mem(stock, bread);
  let add = (bread, stock) => Stock.add(stock, bread);
};

And you can tell that the interface to the module never changed. This is why I find that starting from an interface made my modules better.

Simple Top-Down Development

This technique is called top-down development and is especially suited to OCaml/Reason because you can design the interface first according to the ideal, semantic way you'd like to use it, then 'fill in the blanks' with an implementation. In fact, to keep the compiler happy while you develop the interface, you can mock up a dummy implementation that will type-check but fail at runtime. The technique is covered in more detail in this blog post (OCaml syntax): https://blog.janestreet.com/simple-top-down-development-in-ocaml/