Simple, strongly-typed, testable, effectful computations.
tl;dr Reader, plus type intersection and subtraction to manipulate the Reader's environment.
coming soon
TODO: Add example
- Communication: see what resources a computation will use just by looking at its type: database, remote service API, logger, etc.
- Safety: only computations whose resource requirements have been fully satisfied can be executed.
- Testability: easily test computations by providing test resources without "magic mocking", overwriting globals, or other error-prone approaches.
Let's start with the types and some basic intuitions about the pieces:
type Env<R, A> = ...
function chain <RA, RB, A, B> (f: (a: A) => Env<RB, B>, e: Env<RA, A>): Env<RA & RB, B>
type Subtract<T, T1> = Omit<T, keyof T1>
function use <RA, RB, A>(rb: RB, e: Env<RA, A>): Env<Subtract<RA, RB>, A>
type None = ...
function runPure <A> (e: Env<None, A>, k: (a: A) => void = () => {}): CancelEnvrepresents computations that must execute in an environment which provides a set of resourcesR.chainaggregates resource requirements by combining twoEnvcomputations to produce a thirdEnvcomputation.useeliminates resources requirements on the environment by providing some or all of the resources.runPureexecutes a computation whose resource requirements have been fully satisfied.
Note the intersection RA & RB in chain and the Subtract in use. They are a key part of what makes Env useful.
An environment is a set of resources. The resources can be anything: app configuration, a logger, a database API, a remote service API.
TODO: Add example
We'll go into more depth on the Env type later. For now, let's keep it simple: Env represents a computation that will produce an A only if executed in an environment that provides a set of resources, R.
For the next few sections, think of Env as being like a function:
type Env<R, A> = (r: R) => AIf Env is like a function, clearly we need to provide an R to execute it. At first glance, then, runPure's type may seem confusing.
function runPure <A> (e: Env<None, A>, ...): ...It can only execute computations that require no resources. We need a way to reduce the requirements of an Env to the empty set so that we can execute it with runPure.
Let's look at the type of use:
type Subtract<T, T1> = Omit<T, keyof T1>
function use <RA, RB, A>(rb: RB, e: Env<RA, A>): Env<Subtract<RA, RB>, A>Notice how the input, e, requires RA, and the output Env requires Subtract<RA, RB>. By providing an RB, use satisfies some or all of e's requirements. To put it another way, by providing RB, use subtracts RB from e's requirements.
When RB contains a subset of the resources required by e, use retur
8000
ns an Env with a smaller set of requirements.
Crucially, when RB contains all the resources of RA (e.g., RB is the same as RA), use returns an Env whose requirements have been fully satisfied, and can be executed with runPure.
What if we have two computations:
// User data access capability
type UserService = {
lookupUserByUsername: (username: string) => Promise<User>
}
// Getting a User requires UserService
function getUserByUsername(username: string): Env<UserService, User> {
// ...
}
// Email sending capability
type EmailService = {
sendMessage: (message: EmailMessage) => Promise<EmailSendStatus>
}
// Sending email requires EmailService
function sendEmail(to: Email, subject: string, body: string): Env<EmailService, EmailSendStatus> {
// ...
}It may seem that we need to call each, then call use twice to satisfy the requirements of each, and then call runPure on each. Instead, we can use chain:
function chain <RA, RB, A, B> (f: (a: A) => Env<RB, B>, e: Env<RA, A>): Env<RA & RB, B>chain allows us to write a new computation that gets a User by username and then sends them an email. Note the intersection RA & RB in the output Env.
In contrast to use, which uses subtraction to eliminate requirements, chain uses an intersection to aggregate requirements.
function sendWelcomeEmailToUsername (username: string): Env<UserService & EmailService, EmailSendStatus> {
return chain(
getUserByName(username),
user => sendEmail(user.email, 'Welcome!', `Hi ${user.displayName}, welcome to the world of composable computations!`
)
}sendWelcomeEmailToUsername returns a computation that requires both the UserService and EmailService. To execute it, we need to provide the resources of both getUserByUsername and sendEmail. And now we have a choice: we provide both in a single use, or we can provide them separately in two use calls, whichever fits out needs best. In either case, we can execute the final computation with a single call to runPure.
Note that the explicit type annotations are not necessary and are included here for clarity/documentation. Typescript will infer them correctly.
const userService: UserService = //...
const emailService: EmailService = //...
// Create a computation that will lookup user 'brian' and send a welcome email
const computation: Env<UserService & EmailService, EmailSendStatus> = sendWelcomeEmailToUsername('brian')
// Provide all the required resources
const pureComputation: Env<{}, EmailSendStatus> = use({ ...userService, ...emailService }, computation)
// Go! Lookup user 'brian' and send a welcome email
runPure(pureComputation)There's a lot going on in the Env type, so let's step through it from left to right.
type Env<R, A> ...Env represents a computation that will produce an A only if executed in an environment that satisfies its set of requirements, R. What does it mean to "satisfy a set of requirements"?
type Env<R, A> = (r: R, ...) => ...Env is a function type. It must be called with an R in order to produce an A. Thus, you "satisfy" its set of requirements R by calling it with an R. If you're familiar with the Reader type (such as Haskell's Reader), you may recognize this as being similar to the ReaderT pattern.
Where does the A go?
type Env<R, A> = (r: R, k: (a: A) => void) => ...The k parameter is a continuation. By using a continuation instead of returning an A, Env supports both synchronous and asynchronous computations. Finally, let's look at the last piece, the return type.
type Env<R, A> = (r: R, k: (a: A) => void) => CancelEnv returns a Cancel, which allows the caller to abort an in-flight asynchronous computation.