@sirexelite/pathsignals
v1.0.3
Published
A lightweight state manager for React that is optimized for **large apps with many subscriptions**:
Downloads
498
Maintainers
Readme
Overview
A lightweight state manager for React that is optimized for large apps with many subscriptions:
- Copy-On-Write (COW) updates via
store.update(draft => { ... }) - Path-based routing: notify only listeners that care about the changed path (not “everyone”)
useStoreautomatically derives subscriptions via read tracking (Proxy on reads)- For complex selectors you can declare deps explicitly via
combine(deps, fn)
Problem
In large React applications with complex global state:
- small state updates cause many unrelated components to rerender
- subscriptions become hard to reason about
- selector-based optimizations are fragile and implicit
- performance issues appear only in production-scale data
Why PathSignals
PathSignals routes updates by state paths, not by global change detection. Components rerender only when the exact parts of state they read change.
Install
npm i @sirexelite/pathsignalsQuick start
Create a store
import { createStore, createActions } from "@sirexelite/pathsignals";
type State = {
count: number;
a: number;
b: { c: number; d: number };
tags: string[];
};
export const store = createStore<State>({
a: 0,
b: { c: 1, d: 2 },
tags: ["x"],
});
export const actions = createActions(store, {
inc(draft, by: number) { draft.count += by; },
reset(draft) { draft.count = 0; },
});Update state (COW draft)
store.update(draft => {
draft.b.c += 1;
draft.a = 10;
});Subscribe (redux-style: any change)
const unsub = store.subscribe(() => {
console.log("anything changed", store.getState());
});
// later:
unsub();Subscribe by path (precise)
const unsub = store.subscribePath(["b", "c"], () => {
console.log("b.c changed");
}, "exact");
// later:
unsub();React usage
useStore with a bound store
import { store, actions } from "./setup";
import type { State } from "./setup";
const useAppStore = createSelectorWithStore(store);
function Counter() {
const c = useAppStore(s => s.b.c);
return <div>{c}</div>;
}
const useAction = createUseActionWithState<State, typeof actions>(store);
const useAct = createUseActionWithState<State, typeof actions>(store);
const useActs = createUseActionsWithState<State, typeof actions>(store);
function Btn() {
const inc = useAct("inc"); // (by: number) => void
const reset = useAct("reset"); // () => void
return (
<>
<button onClick={() => inc(1)}>+1</button>
<button onClick={() => reset()}>reset</button>
</>
);
}
function Toolbar() {
const a = useActs(); // { inc, reset }
return <button onClick={() => a.inc(5)}>+5</button>;
}useStore will:
- run the selector against a read-proxy to collect read paths (deps)
- subscribe to those paths via subscribePath(..., "exact")
- rerender only when the relevant paths change
Conditional selectors (deps change over time)
const v = useAppStore(s => (s.a === 0 ? s.b.c : s.b.d));
return <div>{v}</div>;
}When the condition changes what the selector reads, useStore will recompute deps and resubscribe.
combine(deps, fn) for complex selectors
When a selector is expensive (loops, searching, filtering), or you want explicit control over deps, use combine:
const sum = combine(
[["b", "c"], ["b", "d"]] as const,
(c, d) => c + d
);
// React:
const v = useAppStore(sum);When combine is especially useful
- selector performs loops/filter/search and you want to avoid proxy read-tracking overhead
- you know exactly which paths matter
- you want to avoid “selector may run twice” (read-tracking + real state)
Rules and limitations (important)
1) Only mutate state via store.update
Do not mutate store.getState() directly.
❌ Wrong:
store.getState().b.c = 123; // store won't know, no notifications✅ Correct:
store.update(d => { d.b.c = 123; });2) Selectors must be pure (read-only)
Selectors may run against a read-proxy to collect deps. They must not write.
❌ Wrong:
useAppStore(s => {
s.a = 1; // throws
return s.a;
});✅ Correct:
useAppStore(s => s.a);3) No shared references (same object in multiple branches)
Path-based routing assumes one object belongs to one logical place in the tree. If the same object reference is reachable via multiple paths, change tracking becomes ambiguous.
In dev mode this is detected and throws.
❌ Wrong:
const shared = { x: 1 };
createStore({ a: shared, b: shared }); // dev: error during draft usage✅ Correct:
createStore({ a: { x: 1 }, b: { x: 1 } });4) Arrays policy: primitive arrays only
Arrays in state must contain only primitives:
string | number | boolean | null | undefined | bigint | symbol
✅ Allowed:
tags: ["a", "b"]
ids: [1, 2, 3]❌ Forbidden:
items: [{ id: 1 }, { id: 2 }] // dev: throwsHow to store “array of objects”
Use normalization:
type State = {
users: {
byId: Record<string, User>;
ids: string[];
}
}5) Array mutations are treated as “array path changed”
If you write tags[i] or do push/pop/splice, it is treated as a change of the whole array path ["tags"].
Subscribing to individual array elements is not supported by design.
6) "exact" vs "deep" subscription modes
"exact"— subscribe to a specific path"deep"— subscribe to a path and any descendant changes (useful for tools/logging/dev)
useStore uses "exact".
Recommendation:
- UI components: use
"exact" - tooling/global listeners: use
"deep"
State modeling recommendations
For big collections, normalize
✅ Preferred:
{
messages: {
byId: Record<string, Message>;
ids: string[];
}
}Benefits:
- updating one entity only wakes subscriptions for that byId[id] subtree
- ids changes separately and typically less frequently
Performance notes
- Leaf updates clone only the container chain to root (COW)
- Notifications route by trie paths, not by iterating all subscribers
- In React, normally only the components that read changed paths rerender
Dev vs Prod
createStore(initial, { dev: true }) enables:
- shared reference detection
- array policy checks (no complex elements)
In production you typically want:
createStore(initial, { dev: false })FAQ
Why can a selector run twice?
If you do not use combine, useStore may:
- run selector on a read-proxy to collect deps
- run selector on the real state to compute value
If that matters for a specific selector, use combine(deps, fn).
Why forbid arrays of objects?
Subscribing to array elements causes path explosion and expensive update logic. Primitive arrays + normalized object collections keep routing predictable and fast.
Benchmark
React hook benchmarks (useStore) — update-only
name hz min max mean p75 p99 p995 p999 rme samples
· mount 2000 subscribers (single root) 26.7227 31.7558 58.8020 37.4213 36.1713 58.8020 58.8020 58.8020 ±15.65% 10
· 2000 subscribers mounted -> update 1 leaf (update-only) 1,815.88 0.4863 3.6128 0.5507 0.5278 1.7170 3.6128 3.6128 ±6.73% 182
· teardown 1,154,538.18 0.0007 0.0034 0.0009 0.0009 0.0017 0.0020 0.0025 ±1.18% 1155
Store benchmarks (routing/update only)
name hz min max mean p75 p99 p995 p999 rme samples
· routing: 10k leaf subscriptions -> 1 leaf update (update-only) 308.56 2.9479 5.6728 3.2409 3.1977 5.6728 5.6728 5.6728 ±6.23% 31
· COW: deep leaf write (depth=30) update-only 17,403.35 0.0433 5.5667 0.0575 0.0478 0.1640 0.2429 0.6212 ±11.13% 1741
· arrays: push+pop (stable size) update-only 108,057.73 0.0073 0.4722 0.0093 0.0090 0.0216 0.0269 0.1090 ±1.83% 10806