Files

T

Antigravity Agent 504fd5fb42 [G1-Sync] Manual knowledge update

2026-05-10 22:08:15 +09:00

6.7 KiB

Raw Blame History

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

title

JavaScript Optimization Patterns

매 한 줄

"매 hot path 의 type stability + allocation 최소화". V8/JSC/SpiderMonkey 모두 매 inline cache + hidden class 의 의존 — 매 monomorphic call site 가 매 fastest. 매 micro-optimization 보다 매 algorithmic / batching 이 매 win 큼, 매 그러나 hot loop 에서는 매 GC pressure 의 의식 필수.

매 핵심

매 V8 의 fast path

Hidden class (Map): object shape — 매 동일 property 순서 의 같은 hidden class.
Inline cache (IC): call site 별 type 기록 — monomorphic > polymorphic > megamorphic.
TurboFan / Maglev: hot function 의 optimizing compile, type feedback 기반.
Sparkplug (2021+): 매 baseline JIT, 매 quick startup.

매 GC pressure

Young generation (Scavenger): 매 short-lived alloc — 매 cheap.
Old generation (Mark-Compact): 매 promoted object — 매 stop-the-world 의 risk.
Strategy: 매 hot loop 에서 alloc 의 회피 (object pooling, typed arrays).

매 응용

Animation / game loop 의 60+ FPS 유지.
매 large data transformation (millions of records).
Server-side hot path (Node, Bun).

💻 패턴

Object shape stability

// X — hidden class 가 다름 (assignment 순서)
function A() { this.x = 1; this.y = 2; }
function B() { this.y = 2; this.x = 1; }
const arr = [new A(), new B()]; // 매 polymorphic IC

// O — 매 동일 shape
class Point { constructor(x, y) { this.x = x; this.y = y; } }
const arr2 = [new Point(1, 2), new Point(3, 4)];

Monomorphic call site

function add(a, b) { return a + b; } // V8 의 type feedback 기록

// X — megamorphic (string + number + array)
add(1, 2); add('a', 'b'); add([], []);

// O — monomorphic (always number)
for (let i = 0; i < 1e6; i++) add(i, i + 1);

Typed arrays (no boxing)

// X — 매 Array of number — boxed double, GC pressure
const heights = new Array(1_000_000);
for (let i = 0; i < heights.length; i++) heights[i] = Math.random();

// O — Float64Array — flat, no boxing
const heights2 = new Float64Array(1_000_000);
for (let i = 0; i < heights2.length; i++) heights2[i] = Math.random();

Object pooling (hot loop)

class Vec3Pool {
  #pool = [];
  acquire(x = 0, y = 0, z = 0) {
    const v = this.#pool.pop() ?? { x: 0, y: 0, z: 0 };
    v.x = x; v.y = y; v.z = z;
    return v;
  }
  release(v) { this.#pool.push(v); }
}

const pool = new Vec3Pool();
function frame() {
  const tmp = pool.acquire(1, 2, 3);
  // ... compute ...
  pool.release(tmp);
}

Loop hoisting

// X — length 매 iter 마다 access
for (let i = 0; i < items.length; i++) { ... }

// O — modern V8 의 자동 hoist 하지만 매 explicit 가 안전
const n = items.length;
for (let i = 0; i < n; i++) { ... }

// O+ — for-of with iterator (매 modern, V8 fast path)
for (const it of items) { ... }

Map vs object lookup

// 매 string key 의 dynamic — Map 의 매 fast & predictable
const cache = new Map();
cache.set(key, value);
cache.get(key);

// 매 known small fixed keys — object 매 inline cache 의 winning
const config = { mode: 'fast', retries: 3 };

Avoid `arguments`, use rest

// X — arguments 의 deopt (non-array, function-bound)
function sum() {
  let s = 0;
  for (let i = 0; i < arguments.length; i++) s += arguments[i];
  return s;
}

// O
function sum(...nums) {
  let s = 0;
  for (const n of nums) s += n;
  return s;
}

Batching DOM / state updates

// X — 매 update 의 layout thrash
items.forEach(it => container.appendChild(make(it)));

// O — DocumentFragment batch
const frag = document.createDocumentFragment();
items.forEach(it => frag.appendChild(make(it)));
container.appendChild(frag);

Web Workers (off-main-thread)

// main.js
const worker = new Worker('worker.js', { type: 'module' });
worker.postMessage({ data: bigArray }, [bigArray.buffer]); // 매 transfer, zero-copy

// worker.js
self.onmessage = (e) => {
  const result = heavy(e.data.data);
  self.postMessage(result);
};

Structured cloning vs transferable

// 매 1MB+ data — transferable 의 50-100x faster
const buf = new ArrayBuffer(1_000_000);
worker.postMessage(buf, [buf]); // 매 ownership transfer

Lazy evaluation (generator)

function* range(from, to) {
  for (let i = from; i < to; i++) yield i;
}
function* map(it, f) { for (const v of it) yield f(v); }
function* filter(it, p) { for (const v of it) if (p(v)) yield v; }

// 매 1B element 의 first 10 — alloc 의 X
const result = [];
for (const v of filter(map(range(0, 1e9), x => x * 2), x => x % 3 === 0)) {
  result.push(v);
  if (result.length === 10) break;
}

매 결정 기준

상황	Approach
Hot loop alloc	typed array + object pool
Large data transform	streaming / generator
Heavy compute	Web Worker + transferable
DOM batch	DocumentFragment / requestAnimationFrame
Type stability 의심	DevTools Profiler "ICs" tab
매 micro-bench 결과 의 의심	always profile in production-like build

기본값: 매 algorithmic win 우선, 매 hot path 만 micro-optimize.

🔗 Graph

부모: JavaScript · V8
변형: Hermes-Optimization · JSC
응용: Web-Workers · Animation-Performance
Adjacent: Garbage-Collection · Hidden-Classes

🤖 LLM 활용

언제: 매 measurable bottleneck 진단 후, hot loop 작성, large data transform. 언제 X: 매 cold path / one-shot script (premature optimization), 매 readability cost > perf gain.

❌ 안티패턴

Premature micro-opt: 매 readability 희생, 매 measure 없이 추측.
delete obj.prop: hidden class transition 유발, IC deopt.
Hot loop 의 closure alloc: arr.map(x => x*2) 매 frame — 매 함수 hoist.
try/catch in hot loop: 매 V8 의 optimization 일부 비활성 (개선되었지만 여전히 cost).
string concat in loop: += 매 hot loop — 매 array.join.

🧪 검증 / 중복

Verified (V8 blog, Chrome DevTools Performance docs, Bun benchmark methodology).
신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — V8 hot-path optimization 패턴 정리

6.7 KiB Raw Blame History