The demo code:
- Elm on ellie-app
- TypeScript on TypeScript playground
One of the concepts in functional programming that trips up beginners is Algebraic Data Types (ADT). I am going to go through some examples and relate it to some concepts that would be familiar to Object Oriented Programming (OOP).
For this introduction I will be using TypeScript for the OOP examples and Elm for the functional examples.
Let's start with an example.
type Animal = Cat | Dog
An Animal
can be either a Cat
or a Dog
, but you can't just be an animal. You need to be a type of an animal, the classic "is a" relationship. This is a bit similar to abstract classes in OOP languages.
abstract class Animal {
abstract speak(): string;
}
class Cat extends Animal {
speak(): string {
return "meow";
}
}
class Dog extends Animal {
speak(): string {
return "bark";
}
}
// can't make an abstract Animal
// const myAnimal = new Animal();
const myCat = new Cat();
console.log(myCat.speak());
Now you could make a function that generically takes any type of Animal
as an argument.
function feedAnimal(animal: Animal): string {
return "thanks for the food! " + animal.speak();
}
console.log(feedAnimal(myCat));
Similarly, let's translate this to Elm.
animalSpeak : Animal -> string
animalSpeak animal =
case animal of
Cat ->
"meow"
Dog ->
"bark"
feedAnimal : Animal -> string
feedAnimal animal =
"thanks for the food! " ++ animalSpeak(animal)
Why are they useful?
- Easier refactors. You can catch problems at compile time.
- Can constrain the problem domain by the type system. See Making Impossible States Impossible.
When working on a new feature or a refactor with a strongly typed language I will often start by changing the data structures and then let the compiler tell me where my code fails to compile. When my code compiles again my job is mostly done.
You can do this by leveraging classes and using them as types but it is a lot easier and compact to have a robust type system built into the language. There is usually a lot of boiler code needed to get some better type safety in languages like Java or TypeScript. See this stackoverflow question.
Animal Names
Since there are no classes or member variables in a language like Elm we use records instead.
type alias AnimalRecord =
{ type_ : Animal
, name : String
}
felix = AnimalRecord Cat "Felix"
When I write TypeScript I actually prefer using interfaces instead of defining classes. The reason for this preference is that I like to separate the structure of the data from the operations on that data.
I will, in fact, claim that the difference between a bad programmer and a good one is whether he considers his code or his data structures more important. Bad programmers worry about the code. Good programmers worry about data structures and their relationships.
Linux Torvalds, see https://lwn.net/Articles/193245/
So it would look like this:
interface AnimalRecord {
type_ : Animal;
name : string;
}
const felix : AnimalRecord = {
type_: new Cat(),
name : "Felix",
};
Immutability
One core idea of functional programming is immutability. You can get more immutability in TypeScript by declaring class members or interfaces as immutable.
- for classes https://www.typescriptlang.org/docs/handbook/classes.html#readonly-modifier
- for interfaces https://www.typescriptlang.org/docs/handbook/utility-types.html#readonlyt
An example with error handling
I wrote about this in another post Exceptions considered harmful.
Let's use the Either
type from purify-ts to deal with errors in our code. The TypeScript compiler can help us if we give it the right information.
Below is the type signature for Either (almost). See https://github.com/gigobyte/purify/blob/master/src/Either.ts#L3 Neverthrow's Result type signature is a bit easier to read but has less functionality.
type Either<E, T> = Left<E, T> | Right<T, E>
An Either
can be one of Left
which holds the error or it can be Right
which holds the successful return value.
// might be one of an Error or a result string
function getUsernameFromDB(id : number) : Either<Error, string> {
const username : string;
const error : Error;
try {
username = DB.getUserById(id);
} catch(e) {
error = e;
}
if (error) {
return Left(error); // error values are always Left
} else {
return Right(username); // Right is "right" or the correct value
}
// could use Either.encase instead
// see https://gigobyte.github.io/purify/adts/Either#encase
// return Either.encase(() => DB.getUserByID(id));
}
const result = getUsernameFromDB(1);
// now the compiler forces you to deal with the error
result.caseOf({
Left: error => error.toString(),
Right: value => value,
})
Here are the docs of caseOf
statements for
- Either https://gigobyte.github.io/purify/adts/Either#caseOf
- Maybe https://gigobyte.github.io/purify/adts/Maybe#caseOf
When I first encountered ADT in Elm it looked so foreign. Once I got used to them I wished that every language I used had them. The compiler now becomes your best friend.
You can get fancier with TypeScript that I did in this post to get something closer to Elms ADTs using union types. See https://www.typescriptlang.org/docs/handbook/advanced-types.html#union-types TypeScript also has type guards that aid in handling these types.
Top comments (0)