GeezerVison

This library is in early alpha and not yet ready for production use.

Typed helpers to compose React’s server cache with Effect in a type-safe, ergonomic way.

pnpm add @mcrovero/effect-react-cache effect react

React exposes a low-level cache primitive to memoize async work by argument tuple during a React Server Component render. This library wraps an Effect-returning function with React’s cache so you can:

• Deduplicate concurrent calls: share the same pending promise across callers
• Memoize by arguments: same args → same result without re-running the effect
• Keep Effect ergonomics: preserve R requirements and typed errors

import { Effect } from "effect"
import { reactCache } from "@mcrovero/effect-react-cache/ReactCache"

// 1) Wrap an Effect-returning function
const fetchUser = (id: string) =>
  Effect.gen(function* () {
    yield* Effect.sleep(200)
    return { id, name: "Alice" as const }
  })

const cachedFetchUser = reactCache(fetchUser)

// 2) Use it like any other Effect
// React memoization only happens when this Effect is executed
// from an active React server render.
await Effect.runPromise(cachedFetchUser("u-1"))

import { Effect } from "effect"
import { reactCache } from "@mcrovero/effect-react-cache/ReactCache"

const getUser = (id: string) =>
  Effect.gen(function* () {
    yield* Effect.sleep(100)
    return { id, name: "Alice" as const }
  })

export const cachedGetUser = reactCache(getUser)

// When executed inside the same React server render:
// same args → computed once, then memoized
await Effect.runPromise(cachedGetUser("42"))
await Effect.runPromise(cachedGetUser("42")) // reuses cached promise

import { Effect } from "effect"
import { reactCache } from "@mcrovero/effect-react-cache/ReactCache"

export const cachedNoArgs = reactCache(() =>
  Effect.gen(function* () {
    yield* Effect.sleep(100)
    return { ok: true as const }
  })
)

import { Context, Effect } from "effect"
import { reactCache } from "@mcrovero/effect-react-cache/ReactCache"

class Random extends Context.Tag("MyRandomService")<Random, { readonly next: Effect.Effect<number> }>() {}

export const cachedWithRequirements = reactCache(() =>
  Effect.gen(function* () {
    const random = yield* Random
    const n = yield* random.next
    return n
  })
)

// Inside the same React server render, the first call for a given args tuple
// determines the cached value
await Effect.runPromise(cachedWithRequirements().pipe(Effect.provideService(Random, { next: Effect.succeed(111) })))

// Subsequent calls with the same args reuse the first result,
// even if a different Context is provided!
await Effect.runPromise(cachedWithRequirements().pipe(Effect.provideService(Random, { next: Effect.succeed(222) })))

declare const reactCache: <F extends (...args: Array<any>) => Effect.Effect<any, any, any>>(
  effect: F
) => (...args: Parameters<F>) => ReturnType<F>

• Input: an Effect-returning function
• Output: a function with the same signature, whose evaluation is cached by argument tuple using React’s cache

• Internally uses react/cache to memoize by the argument tuple.
• For each unique args tuple, the first evaluation creates a single promise that is reused by all subsequent calls (including concurrent calls).
• The Effect context (R) is captured at call time, but for a given args tuple the first completed Exit is reused for the lifetime of the current React request/render cache.

• First call wins: for the same args tuple, the first call’s context and outcome (success or failure) are cached. Later calls with a different context still reuse that result.
• Errors are cached: if the first call fails, the rejection is reused for subsequent calls with the same args tuple.
• Concurrency is deduplicated: concurrent calls with the same args share the same pending promise.

• Do: cache pure/idempotent computations that return plain data.
• Do: include discriminators (locale, tenant, user) in the argument tuple when results depend on them.
• Don't: pass effects that require Scope or create live resources (DB/client handles, file handles, sockets). Acquire resources outside and provide them, or use a Layer.
• Don't: rely on per-call timeouts/cancellation or different Context for the same args. The first call determines the cached outcome and context.

• No scoped resources: Effects requiring Scope are rejected at the type level. React's cache evaluates once and reuses the result, so any scoped resource would be finalized immediately after creation, breaking later callers.
• First call wins: For a given args tuple, the first call's context and outcome (success or failure) are cached and reused.
• Context sensitivity: If results depend on request context (logger level, locale, tracer span, etc.), include those discriminators in the arguments or avoid caching.
• Streams/Channels: Don't cache effects that return live Stream/Channel handles tied to resources.

When running tests outside a React server render, you may want to mock react’s cache to ensure deterministic, in-memory memoization. React’s default non-server build treats cache as a passthrough, so plain Effect.runPromise(...) calls will not memoize on their own. A simple primitive-oriented mock looks like this:

import { vi } from "vitest"

vi.mock("react", () => {
  return {
    cache: <F extends (...args: Array<any>) => any>(fn: F) => {
      const memo = new Map<string, ReturnType<F>>()
      return ((...args: Array<any>) => {
        const key = JSON.stringify(args)
        if (!memo.has(key)) {
          memo.set(key, fn(...args))
        }
        return memo.get(key) as ReturnType<F>
      }) as F
    }
  }
})

See test/ReactCache.test.ts for a more faithful identity-based mock and examples covering caching, argument sensitivity, context provisioning, and concurrency.

• The cache is keyed by the argument tuple using React’s semantics. Prefer primitives or stable object identities as arguments.
• Since the first outcome is cached, design your effects such that this is acceptable for your use case. For context-sensitive computations, include discriminators in the argument list.
• This library is designed for React Server Components. Outside a React server render, react’s default cache implementation is effectively a passthrough.

You can use this library together with @mcrovero/effect-nextjs to deduplicate Effect-based functions within the same Next.js server render across pages, layouts, and server components.