skyr

Type-safe error handling for TypeScript, inspired by Rust's Result type. Method chaining, functional composition with pipe(), automatic async propagation, and optional dependency injection.

Installation

# Deno
deno add jsr:@thefridge/skyr
# Bun
bunx jsr add @thefridge/skyr
# pnpm
pnpm add jsr:@thefridge/skyr
# npm
npx jsr add @thefridge/skyr
# Yarn
yarn add jsr:@thefridge/skyr

Core Concepts

Results Instead of Exceptions

A Result<T, E> is either ok (containing a value of type T) or an err (containing a structured error with a typed code, a message, and an optional cause). This replaces try/catch with values you can inspect, transform, and compose.

import * as R from "@thefridge/skyr";

function validateEmail(email: string) {
	if (!email.includes("@")) {
		return R.err("INVALID_EMAIL", "Email must contain @");
	}
	return R.ok(email);
}

const result = validateEmail("user@example.com");

if (result.isOk()) {
	console.log(result.value); // "user@example.com"
} else {
	console.log(result.code); // "INVALID_EMAIL"
	console.log(result.message); // "Email must contain @"
}

Error codes are string literals tracked by the type system. TypeScript knows exactly which errors a function can produce and autocompletes them for you.

Results are plain objects with a _tag discriminant:

• ok(value) creates { _tag: "Ok", value, ... }
• err(code, message, cause?) creates { _tag: "Err", code, message, cause, ... }

The cause field is undefined when not provided.

Every Result comes with methods for transforming and inspecting it. Type . in your IDE and see what's available.

Letting TypeScript Infer Result Types

There's a subtlety with the example above. Without an explicit return type annotation, TypeScript infers the return type as Result<string, never> | Result<never, "INVALID_EMAIL">, a union of two separate Result types rather than a single unified Result<string, "INVALID_EMAIL">.

You could fix this by adding a return type annotation, but it's generally safer to let TypeScript infer return types whenever possible. Annotations can drift out of sync with the implementation and mask bugs.

Instead, wrap the function with fn():

import * as R from "@thefridge/skyr";

const validateEmail = R.fn((email: string) => {
	if (!email.includes("@")) {
		return R.err("INVALID_EMAIL", "Email must contain @");
	}
	return R.ok(email);
});
// (email: string) => Result<string, "INVALID_EMAIL">

fn() collapses all the Result branches into a single, clean Result<T, E> type. No annotation needed; the ok value type and all possible error codes are inferred automatically.

This is the simplest use of fn(). It also supports generator functions for railway-style programming and dependency injection, covered later.

Type Guards

Results have .isOk() and .isErr() methods that act as type guards:

if (result.isOk()) {
	result.value; // T
} else {
	result.code; // E
	result.message; // string
	result.cause; // unknown | undefined
}

Standalone functions isOk(), isErr(), and isResult() are also available:

R.isOk(result); // narrows to Ok<T>
R.isErr(result); // narrows to Err<E>
R.isResult(value); // checks if any unknown value is a Result

isResult(value) checks whether any unknown value is a Result. A value is considered a Result if it's a non-null object with _tag equal to "Ok" or "Err".

Method Chaining

Every Result has methods for transformation, error handling, and value extraction. Chain them directly, no imports or special syntax needed:

const message = validateEmail("User@Example.com")
	.map((email) => email.toLowerCase().trim())
	.map((email) => `Welcome, ${email}!`)
	.match({
		ok: (greeting) => greeting,
		err: (e) => `Error: ${e.message}`,
	});

console.log(message); // "Welcome, user@example.com!"

Methods skip over errors automatically. If validateEmail returns an error, the .map() calls are never executed and the error flows straight to .match().

Async Propagation

When any step returns a Promise, the result becomes an AsyncResult, a wrapper around Promise<Result> with the same methods. This is called async poison: once async, always async (until you await).

const result = validateEmail("user@example.com") // Result<string, ...>
	.map((email) => fetchUser(email)) // returns Promise → AsyncResult
	.map((user) => user.name); // still async, still chainable
// Type: AsyncResult<string, ...>

const finalResult = await result;
// Type: Result<string, ...> (back to sync)

AsyncResult is PromiseLike, so you can await it to get back a sync Result with all its methods.

Functional Composition with pipe()

For those who prefer a functional style, pipe() threads a value through a sequence of functions left-to-right. All methods are also available as standalone operators:

import * as R from "@thefridge/skyr";

const message = R.pipe(
	validateEmail("User@Example.com"),
	R.map((email) => email.toLowerCase().trim()),
	R.map((email) => `Welcome, ${email}!`),
	R.match({
		ok: (greeting) => greeting,
		err: (e) => `Error: ${e.message}`,
	}),
);

Standalone operators accept both Result and Promise<Result> as input and propagate async automatically and work interchangeably with both styles.

Panics: Unexpected Throws

If your code calls something that might throw or reject, or you're unsure whether it could, wrap it with fromThrowable() to convert it into a Result safely. This is the recommended approach for any code you don't fully control.

If a callback passed to map, mapErr, or match throws synchronously without being wrapped, a Panic is thrown, halting execution immediately unless caught:

R.ok(42).map(() => {
	throw new Error("oops");
});
// Throws: Panic("map() callback threw — use fromThrowable() for unsafe code")
//   cause: Error("oops")

Panic extends Error, so you get a full stack trace. Catch it at the top level with instanceof R.Panic if needed.

Async functions (Promises) returned from callbacks are automatically wrapped with fromThrowable implicitly, so rejections become UNKNOWN_ERR results instead of Panics. However, if you use multiple async functions, all their errors will share the same UNKNOWN_ERR code, making it impossible to distinguish between them. Wrapping each one with fromThrowable and a dedicated error code is the recommended approach.

Methods

Every Result has the following methods. AsyncResult has the same methods, but they always return AsyncResult (or Promise for terminal operations).

.map(fn)

Transforms the ok value. Skips if the result is an error.

R.ok(5)
	.map((n) => n * 2)
	.map((n) => `Value: ${n}`);
// Result<string, never> → Ok("Value: 10")

If fn returns a Result, it's automatically flattened (no nested Results). If it returns a Promise, the result becomes an AsyncResult. The Promise is automatically handled like fromThrowable: resolved values become ok, rejected Promises become UNKNOWN_ERR.

.mapErr(fn | handlers)

Transforms or recovers from errors. Has two forms:

Function form - transform all errors:

R.err("NOT_FOUND", "User not found")
	.mapErr((e) => R.err("DEFAULT_ERROR", e.message));
// Result<never, "DEFAULT_ERROR">

Handler object - handle specific error codes with autocomplete:

type AppError = "NOT_FOUND" | "TIMEOUT" | "AUTH_FAILED";

declare function fetchUser(id: string): R.Result<User, AppError>;

const result = fetchUser("123").mapErr({
	NOT_FOUND: () => R.ok(guestUser), // recover with ok()
	TIMEOUT: () => defaultUser, // recover with plain value (same as ok())
	// AUTH_FAILED not listed → passes through unchanged
});
// Result<User, "AUTH_FAILED">

Handlers get autocomplete for the available error codes. Each handler receives the narrowed Err<"CODE"> and can:

• Return ok(value) or a plain value to recover (both treated as success)
• Return err(code, message) to transform the error

Unhandled codes pass through unchanged.

.match({ ok, err })

Pattern match both cases and leave the Result world:

const label = R.ok(42).match({
	ok: (n) => `Got ${n}`,
	err: (e) => `Error: ${e.code}`,
});
// "Got 42"

If either handler returns a Result, the output is a Result. Otherwise it's a plain value. On AsyncResult, .match() returns a Promise.

.inspect(fn) / .inspectErr(fn)

Run side effects (logging, metrics) without changing the Result:

validateEmail("user@example.com")
	.inspect((email) => console.log("Valid:", email))
	.inspectErr((e) => console.error("Failed:", e.code))
	.map((email) => email.toLowerCase());

The callback's return value is ignored; the original Result is always returned unchanged. If the callback throws or the returned Promise rejects, the error is silently swallowed and the original Result passes through. Side effects should never break the pipeline.

.unwrap() / .unwrapOr(default)

Extract values from Results:

// unwrap() extracts the ok value, or returns undefined on error
const value = R.ok(42).map((n) => n * 2).unwrap();
// number | undefined → 84

const missing = R.err("NOT_FOUND", "gone").unwrap();
// undefined

// Works great with optional chaining
fetchUser("123").unwrap()?.name;

// Or non-null assertion when you know it's Ok
R.ok(42).unwrap()!;

// unwrapOr() extracts the ok value, or returns the default on error
const fallback = R.err("ERROR", "Something went wrong").unwrapOr(0);
// 0

On AsyncResult, .unwrap() returns Promise<T | undefined> and .unwrapOr(default) returns Promise<T | D>.

Converting Throwing Code

fromThrowable(fn | promise, mapper?)

Convert code that throws (or Promises that reject) into Results:

// Wrap a function call
const result = R.fromThrowable(
	() => JSON.parse('{"name": "Alice"}'),
	(err) => R.err("PARSE_ERROR", "Invalid JSON", err),
);
// Result<any, "PARSE_ERROR">

// Wrap a Promise
const response = await R.fromThrowable(
	fetch("https://api.example.com"),
	(err) => R.err("FETCH_ERROR", "Request failed", err),
);
// Result<Response, "FETCH_ERROR">

Without a mapper, errors become "UNKNOWN_ERR".

The function overload calls the function synchronously and catches any thrown error. If you have a Promise, use the Promise overload directly.

wrapThrowable(fn, mapper?)

Like fromThrowable, but returns a reusable wrapper function:

const safeParse = R.wrapThrowable(
	(str: string) => JSON.parse(str),
	(err) => R.err("PARSE_ERROR", "Invalid JSON", err),
);

safeParse('{"valid": true}'); // Ok({valid: true})
safeParse("nope"); // Err("PARSE_ERROR")

Dependency Injection with fn()

fn() is a function builder. You pass it a generator and get back a function you can call just like any other. The difference is that inside the generator you get two superpowers:

1. Dependency requests - yield* R.use(Dep) gives you an implementation of a dependency without worrying about how to acquire it. Think of it like a function parameter, except you don't have to pass it in at the call site. When one fn() function calls another, unmet dependencies propagate up automatically, no prop drilling required. You can also choose to supply some dependencies but not all, and the rest keep propagating.

2. Result unwrapping - yield* someResult extracts the success value from any Result or AsyncResult. If it's an error, the function short-circuits (early return) and the error propagates to the caller, just like the non-generator version (fn(() => ...)).

The return type of an fn() function is Fn<Args, Success, Errors, Deps>:

• Args - the parameter list of your generator (e.g. (email: string) → [string])
• Success - the unwrapped success type
• Errors - a union of all error codes, accumulated from every yield* call
• Deps - a union of all unmet dependencies, accumulated from every yield* R.use() call

The Fn you get back is only callable once all dependencies are injected (via .inject()), at which point the Deps part of the type becomes never.

Declaring Dependencies

const Database = R.dependency<{
	findUser: (email: string) => Promise<User | null>;
}>()("database");

const Logger = R.dependency<{
	info: (msg: string) => void;
}>()("logger");

Convention: Dependencies and functions that still need injection use PascalCase. After injection, use camelCase to signal "ready to call."

Creating Functions

const GetUser = R.fn(function* (email: string) {
	const db = yield* R.use(Database);
	const logger = yield* R.use(Logger);

	logger.info(`Looking up ${email}`);

	const validEmail = yield* validateEmail(email);
	const user = yield* R.fromThrowable(db.findUser(validEmail));

	if (!user) return R.err("NOT_FOUND", "User not found");

	return R.ok(user);
});
// Type: Fn<[string], User, "INVALID_EMAIL" | "NOT_FOUND" | "UNKNOWN_ERR", Database | Logger>

Key points:

• yield* R.use(Database) - acquires a dependency from the DI context
• yield* validateEmail(email) - unwraps a Result; short-circuits on failure
• yield* R.fromThrowable(...) - catches throws/rejections, converting them to a Result, then unwraps it
• Error types accumulate automatically across all yield* calls
• Dependency types accumulate automatically across all yield* R.use() calls

Injecting Dependencies

Use .inject() to provide implementations:

const getUser = GetUser.inject(
	Database.impl({ findUser: async (email) => db.query(email) }),
	Logger.impl({ info: console.log }),
);

// All dependencies satisfied, now callable
const result = await getUser("user@example.com");

If you try to call a function before all dependencies are injected, TypeScript shows an error:

ERROR - Missing dependencies: database, logger. Use inject() first.

At runtime, calling with missing dependencies throws an Error with a message like Missing dependency: "database". Use inject() to provide this dependency.

Injection can be done incrementally:

const withDb = GetUser.inject(Database.impl({/* ... */}));
// Still needs Logger

const getUser = withDb.inject(Logger.impl({/* ... */}));
// Fully callable

The standalone inject() operator works the same way inside pipe():

const getUser = R.pipe(
	GetUser,
	R.inject(
		Database.impl({ findUser: async (email) => db.query(email) }),
		Logger.impl({ info: console.log }),
	),
);

Nested Functions

When one fn() uses another via yield* R.use(ChildFn), the child's dependencies are inherited by the parent:

const CheckPermissions = R.fn(function* (userId: string) {
	const db = yield* R.use(Database);
	// ...
	return R.ok(canAccess);
});

const LoginUser = R.fn(function* (email: string, password: string) {
	const logger = yield* R.use(Logger);

	const user = yield* getUser(email);

	const checkPerms = yield* R.use(CheckPermissions);
	const canAccess = yield* checkPerms(user.id);

	return R.ok(user);
});
// Dependencies: Logger | Database (Database inherited from CheckPermissions)

The two-step pattern, yield* R.use(Fn) then yield* callable(args), separates dependency resolution from execution, keeping the control flow explicit.

API Reference

Constructors

Function	Description
ok(value)	Create an Ok result with methods
err(code, message, cause?)	Create an Err result with methods

Type Guards

Function / Method	Description
.isOk()	Narrow to Ok<T> (method)
.isErr()	Narrow to Err<E> (method)
isOk(result)	Narrow to Ok<T> (standalone)
isErr(result)	Narrow to Err<E> (standalone)
isResult(value)	Check if value has _tag of "Ok" or "Err"

Methods on Result

Method	Description
.map(fn)	Transform ok value; Panics on sync throw
.mapErr(fn)	Transform error; plain values treated as recovery
.mapErr({ CODE: handler })	Handle specific error codes; Panics on sync throw
.match({ ok, err })	Pattern match both cases; Panics on sync throw
.inspect(fn)	Side effect on ok; errors silently swallowed
.inspectErr(fn)	Side effect on error; errors silently swallowed
.unwrap()	Extract ok value or return undefined
.unwrapOr(default)	Extract ok value or return default

AsyncResult

AsyncResult<T, E> wraps a Promise<Result<T, E>> and exposes the same methods as Result. All methods return AsyncResult (async poison), except terminal operations (.match(), .unwrap(), .unwrapOr()) which return Promise. AsyncResult is PromiseLike; await it to get a sync Result.

Standalone Operators (for pipe())

Operator	Description
map(fn)	Transform ok value; Panics on sync throw
mapErr(fn)	Transform error; plain values treated as recovery
mapErr({ CODE: handler })	Handle specific error codes; Panics on sync throw
match({ ok, err })	Pattern match both cases; Panics on sync throw
inspect(fn)	Side effect on ok; errors silently swallowed
inspectErr(fn)	Side effect on error; errors silently swallowed
unwrap	Extract ok value or return undefined
unwrapOr(default)	Extract ok value or return default

Converters

Function	Description
fromThrowable(fn, mapper?)	Convert throwing function to Result
fromThrowable(promise, mapper?)	Convert Promise to async Result
wrapThrowable(fn, mapper?)	Wrap function to return Results

Errors

Type	Description
Panic	Thrown on sync throw in operator callbacks; extends Error

Dependency Injection

Function	Description
dependency<T>()(key)	Declare a dependency type
fn(generator)	Create function with DI and Result unwrapping
fn(func)	Unify Result return type
use(dep)	Acquire dependency inside a generator
use(Fn)	Get contextualized callable for nested Fn
.inject(...impls)	Provide dependency implementations (method)
inject(...impls)	Provide dependency implementations (operator)

License

MIT