Type Systems for JavaScript

Elm, Flow, and TypeScript

http://bit.ly/js-types

Oliver Zeigermann / @DJCordhose

If you are looking for the most recent and up-to-date version of my comparisons on Typed JavaScript you have just found it :)

Why using type systems?

Anders Hejlsberg@Build2016: Big JavaScript codebases tend to become "read-only".

weak to strong and dynamic to static

https://twitter.com/DJCordhose/status/829242451294552066

http://stateofjs.com

Recently published survey on the state of JavaScript

http://stateofjs.com/2016/flavors/

TypeScript

ease of use and tool support over soundness

• http://www.typescriptlang.org/
• By Microsoft (Anders Hejlsberg)
• Based on ES6 (probably ES7/ES8)
• Adds optional type annotations, visibility, and decorators
• Compiler checks and removes annotations
• 2.x with major changes released recently

Flow

soundness, no runtime exceptions as goal

• http://flowtype.org/
• By Facebook
• Flow is a static type checker, designed to quickly find errors in JavaScript applications
• Not a compiler, but checker
• If present, type annotations can very easily be removed by babel for runtime

Basics

TypeScript

// variables can have type information
let foo: string;
foo = 'yo';
// Error: Type 'number' is not assignable to type 'string'.
foo = 10;

// types can be inferred (return type)
function sayIt(what: string) {
    return `Saying: ${what}`;
}
const said: string = sayIt(obj);

class Sayer {
    what: string; // mandatory

    constructor(what: string) {
        this.what = what;
    }

    // return type if you want to
    sayIt(): string {
        return `Saying: ${this.what}`;
    }
}

Flow

// variables can have type information
let foo: string;
foo = 'yo';
// Error: number: This type is incompatible with string
foo = 10;

// types can be explicit (parameter) or inferred (return type)
function sayIt(what: string) {
    return `Saying: ${what}`;
}
const said: string = sayIt(obj);

class Sayer {
    what: string; // type also mandatory

    constructor(what: string) {
        this.what = what;
    }

    // return type if you want to
    sayIt(): string {
        return `Saying: ${this.what}`;
    }
}

Flow and TypeScript basics are pretty similar

Those basic features help with documentation, refactoring, and IDE support

Nullability

One of my main sources of runtime exceptions when programming Java

Even after many years it is still surprising how many corner cases I miss in complex code

Generic Type information

Types can be parameterized by others

Most common with collection types

let cats: Array<Cat> = []; // can only contain cats
let animals: Array<Animal> = []; // can only contain animals

// nope, no cat
cats.push(10);
           

// nope, no cat
cats.push(new Animal('Fido'));
           

// cool, is a cat
cats.push(new Cat('Purry'));
           

// cool, cat is a sub type of animal
animals.push(new Cat('Purry'));
           

Up to this point this pretty much works in Flow and TypeScript the same way ...

... but wait

TypeScript

let cats: Array<Cat> = []; // can only contain cats
let animals: Array<Animal> = []; // can only contain animals

// error TS2322: Type 'Animal[]' is not assignable to type 'Cat[]'.
//  Type 'Animal' is not assignable to type 'Cat'.
//    Property 'purrFactor' is missing in type 'Animal'.
cats = animals;
           

// wow, works, but is no longer safe
animals = cats;
           

// because those are now all cool
animals.push(new Dog('Brutus'));
animals.push(new Animal('Twinky'));
           

// ouch:
cats.forEach(cat => console.log(`Cat: ${cat.name}`));
// Cat: Purry
// Cat: Brutus
// Cat: Twinky

TypeScript allows for birds and dogs to be cats here :)

Flow

let cats: Array<Cat> = []; // can only contain cats
let animals: Array<Animal> = []; // can only contain animals

// ERROR
// property `purrFactor` of Cat. Property not found in Animal
cats = animals;
           

// same ERROR
animals = cats;
           

Flow does not have have this caveat

The flipside

This code is safe (as we access cats in a readonly fashion)

function logAnimals(animals: Array<Animal>) {
    animals.forEach(animal => console.log(`Animal: ${animal.name}`));
}

logAnimals(cats);

• This works in TypeScript (and it should)
• however, potentially not safe, there is nothing to keep us from writing to cats
• Flow does not allow this, even though it is safe

// Java
void logAnimals(List<? extends Animal> animals) {
    animals.forEach(animal -> System.out.println("Animal: " + animal.name));
    // illegal:
    animals.add(new Animal("Twinky"));
}

Some Type Inference Magic

Consider

class Dog { woof() { } }

const animals = [];
animals.push(new Dog());

animals.forEach((animal: Dog) => animal.woof());

class Cat { meow() { } }
animals.push(new Cat());

Differences in Generic Types

• TypeScript
  • always covariant (more special):
    parametric types are compatible if the type to assign from has a more special type parameter
  • seems most intuitive, allows for obviously wrong code, though
• Flow
  • using generic types you can choose from invariant (exact match), covariant + (more special), and contravariant - (more common)
  • Array in Flow has an invariant parametric type
  • more expressive, harder to master, disallows some correct code
• Elm
  • Generic data structures using type variables
  • all types have to match exactly

Bounds in Generics

Both Flow and TypeScript support upper, not lower bounds

Both Flow and TypeScript support F-Bounded Polymorphism

https://flowtype.org/blog/2015/03/12/Bounded-Polymorphism.html

https://github.com/Microsoft/TypeScript/wiki/What's-new-in-TypeScript#type-parameters-as-constraints

`any` type

can be anything, not specified

can selectively disable type checking

function func(a: any) {
    return a + 5;
}

// cool
let r1: string = func(10);

// cool
let r2: boolean = func('wat');

• flow / TypeScript 2: explicit any supported, but any never inferred
• Elm: does not exist, everything has exact type

Union Types

aka Disjoint Unions aka Tagged Unions aka Algebraic data types

to describe data with weird shapes

depending on some data other data might apply or not

 // a disjoint union type with two cases
type Response = Result | Failure;

type Result = { status: 'done', payload: Object }; // all good, we have the data
type Failure = { status: 'error', code: number}; // error, we get the error code
           

Implementation both in Flow and TypeScript

function callback(response: Response) {
    // works, as this is present in both
    console.log(response.status);
    // does not work,
    // as we do not know if it exists, just yet
    console.log(response.payload); // ERROR
    console.log(response.code); // ERROR
           

    switch (response.status) {
        case 'done':
            // this is the special thing:
            // type system now knows, this is a Result
            console.log(response.payload);
            break;
        case 'error':
            // and this is a Failure
            console.log(response.code);
            break;
    }
}           

Are Flow and TypeScript like Java/C++/C#?

Not really

• Both
  • optionally typed / any
  • built to match common JavaScript programming patterns
  • type systems more expressive
  • type inference
  • control flow based type analysis
• TypeScript
  • semantically compatible with JavaScript
• Flow
  • just a checker
  • not even a language of its own
  • non-nullability as default

Nominal Typing for Flow classes

class Person {
    name: string;
}

class Dog {
    name: string;
}

let dog: Dog = new Dog();

// nope, nominal type compatibility violated
let person: Person = dog; // ERROR: Dog: This type is incompatible with Person

// same problem
let person: Person = { // ERROR: object literal: This type is incompatible with Person
    name: "Olli"
};
            

Structural Typing for TypeScript classes

class Person {
    name: string;
}

class Dog {
    name: string;
}

let dog: Dog = new Dog();

// yes, correct, as structurally compatible
let person: Person = dog;

// same thing, also correct
let person: Person = {
    name: "Olli"
};

            

Structural vs Nominal Typing

• Nominal Typing: types are compatible when their declared types match
• Structural Typing: types are compatible when their structures match
• Java, C#, C++, C all use nominal typing exclusively
• Flow classes are also treated as nominal types
• TypeScript classes are treated as structural types
• Everything else in both Flow and TypeScript uses structural typing
• Elm always uses structural typing with exact matches on Records

Structural Typing for both TypeScript and Flow

interface NamedObject {
    name: string;
}

// this is fine as nominal typing only applies to Flow classes
let namedObject: NamedObject = dog;

// same thing, also fine
let namedObject: NamedObject = {
    name: "Olli"
};

// not fine in either, missing name
let namedObject: NamedObject = {
    firstName: "Olli"
};
            

Classes in TypeScript

TypeScript has special support for classes

Similar features can be found in Java/C++/C#

• abstract classes and methods
• special shortcut constructors (combined definition of fields and initialization)
• public, private, protected
• decorators (aka Attributes in C# and Annotations in Java)
• readonly properties (TypeScript 2) like in C#

Flow does not feature those or any other syntactic sugar, as it is a checker only

Integration with raw JavaScript

3rd Party Libraries in Flow

• Core Declarations come with Flow Checker, includes React
• Other external declarations are optional
• External Flow Type Definitions
• some libraries even come with added flow type declarations (e.g. immutable.js)
• Declarations can be added to flow config files
• If there are no existing declaration files: still works, but less powerful

TypeScript Declaration files

• Much larger base for external declarations DefinitelyTyped
• TypeScript 1.x
  • needed External Type Declarations for 3rd party libraries
  • turned out to be a major PITA
• TypeScript 2.x
  • like Flow
  • install external type declarations transparently using npm (e.g.npm install @types/react --save)

IDE Support

Visual Studio Code

• https://code.visualstudio.com/
• Excellent TypeScript support
• Directly uses Language Service of TypeScript Compiler
• Written in TypeScript
• Basic Elm support via Plugin

Atom / Nuclide

• https://atom.io/
• https://nuclide.io/ (Atom Package)
• Probably best Flow support

IntelliJ IDEA / Webstorm

Starting from 2016.3

• Flow: integrated checking and display of messages
• Even better TypeScript support: uses Language Service of TypeScript Compiler
• Basic Elm support via Plugin

Barrier to entry for JavaScript people

• Flow: low
  • fits seamlessly in standard JavaScript build chain
  • can be enabled file by file
  • weak mode only shows clear errors in order not to overwhelm people
• TypeScript: medium
  • different build chain (tsc compiler)
  • can include plain JavaScript, but entry point must be TypeScript
• Elm: high
  • different syntax, semantics, and tool chain

Should you use a type checker?

• don't be fooled: checkers do not make your programs error free
• there seems to be little or no impact on productivity
• initial effort to introduce a checker is low, though
• but a type system is a complex thing, it comes at a cost

My biased recommendation

• your project does not live for long: no
• your project is really simple: no
• there is a chance you will need to refactor the thing: yes
• your system is very important or even crucial for the success of your company: yes
• people enter or leave your team frequently: yes
• you have substantial amount of algorithmic code: yes

Where do they excel?

• TypeScript: supporting people from Java and C# land
  • more complete IDE support
  • language server
  • large set of 3rd party declaration files
• Flow: providing typings for idiomatic JavaScript
  • easy to get started even with existing project
  • more powerful and flexible generics
  • nominal typing for classes
• Elm: functional language deliberately different from JavaScript
  • simplicity of type system (no JavaScript legacy)
  • always completely typed (no any)
  • everything immutable and constant always and everywhere
  • complete package (also great orientation for beginners)

General observations / Trivia

• If it is Angular it will be TypeScript
• If people come from a JavaScript background and use React, they are more likely to choose flow
• TypeScript has better tooling support
• TypeScript 2.x has caught up with so many features previously only available in flow, so people are likely to choose TypeScript 2.x when they can start from scratch

Special thanks for giving feedback and helping with this presentation

• Daniel Rosenwasser: @drosenwasser (from the TypeScript team)
• Avik Chaudhuri: @__avik (from the Flow team)
• Richard Feldman: @rtfeldman and Evan Czaplicki: @czaplic (Elm people)

Thank you!

Questions / Discussion

Oliver Zeigermann / @DJCordhose

http://djcordhose.github.io/flow-vs-typescript/elm-flow-typescript.html