perf(client): tick-dispatch timing harness + main-thread tick optimizations (late-game p95 −65%) (#4512)

## Problem

Every 100 ms the main thread's worker `onmessage` callback processes a
full game tick (`gameView.update` → `webglBuilder.update` →
`renderer.tick`). At 60 fps this competes with the 16.7 ms frame budget,
and on the Giant World Map it takes several ms — frame drops on low-end
hardware.

## Harness (`npm run perf:client-tick`)

Headless-Chromium harness that times every worker→main `game_update`
dispatch on the main thread, with structured-clone deserialization
measured separately from the handler body (via a
`Worker.prototype.addEventListener` wrapper installed as a page init
script — no product-code changes). It reports windowed distributions,
captures `.cpuprofile` files at chosen ticks, writes raw samples and an
end-of-run screenshot. `AnalyzeCpuProfile.ts` breaks a profile down by
inclusive time under the dispatch subtree.

Init scripts are passed as **strings**: tsx compiles function-form init
scripts with esbuild `keepNames`, whose injected `__name` helper doesn't
exist in-page and silently kills the game worker setup.

## Baseline (Giant World Map, 400 bots, headless)

Dispatch handler ms — cost **grows with game progression**:

| window | mean | p50 | p95 | max |
|---|---|---|---|---|
| tick 506 | 2.22 | 2.20 | 3.40 | 5.00 |
| tick 1506 | 2.60 | 2.00 | 7.00 | 10.40 |
| tick 2000 | 2.67 | 1.90 | **8.70** | **12.70** |

Deserialization is negligible (0.12 ms mean). CPU profiles attributed
the growing tail to the leaderboard's once-per-second refresh: its
Max-troops column calls `config().maxTroops(p)` for **all ~508
players**, and `PlayerView.units()` scanned **every unit in the game**
per call — O(players × units), growing as units accumulate.

## Round 1 — algorithmic fixes

- **GameView**: new `unitsOwnedBy(smallID)` — an active-units-by-owner
index built lazily at most once per tick. `PlayerView.units()` reads its
own units from it: O(own units) instead of O(all units). Also speeds up
unit display, player panel, and buildables queries.
- **NamePass.updateNames**: reads player state directly from the
caller's map by smallID instead of rebuilding three lookup maps per
tick; skips the slot-assignment sweep once every player has a slot.

After (same map, same spawn tile):

| window | mean | p50 | p95 | max |
|---|---|---|---|---|
| tick 506 | 2.12 | 2.00 | 3.10 | 5.20 |
| tick 1506 | 1.86 | 1.80 | 2.90 | 4.30 |
| tick 2000 | 1.74 | 1.60 | **2.40** | **4.70** |

Late-game p95 −65% (8.7 → 2.4 ms), worst dispatch −63% (12.7 → 4.7 ms),
and per-dispatch cost no longer grows with game progression. The
leaderboard disappeared from the dispatch profile entirely.

## Round 2 — allocation churn + time slicing

Aimed at GC pauses and low-end CPUs; measures flat vs round 1 on a fast
machine, as expected:

- **`FrameData.changedTiles`** is now the plain tile-ref array GameView
already builds instead of a per-tile `{ref, state}` object copy — heavy
battle ticks allocated tens of thousands of objects per tick for a
`state` field that was always 0. `TilePair` removed; `TerritoryPass`
buckets refs synchronously, so the live reference is safe.
- **`UnitView.lastPos`** is only re-sliced when a move actually appended
a position — the unconditional `slice(-1)` allocated an identical
1-element array per unit per tick, including for structures that never
move.
- **`NamePass.updateNames`** refreshes slots round-robin, a quarter per
tick — the full per-player diff pass spreads over ~400 ms, under the
existing 500 ms troop-text cadence; positions lerp continuously. Unnamed
slots and snap passes (seeks) are always processed so nothing pops in
late. Dispatch share: 17% → 13%.

Not sliced on purpose: tile ingest and frame upload need a consistent
per-tick snapshot (stale `GameMap` reads would leak into hover queries,
minimap, attack targeting) — a correctness risk not worth ~1 ms while
the worst dispatch already fits in a quarter of the frame budget.

## Verification

- `npx tsc --noEmit`, eslint clean; full suite green (1929 tests)
- 6 new GameView tests cover the owner index (grouping, inactive
exclusion, ownership capture, death, type filtering, copy semantics);
changedTiles tests updated to the ref-array contract
- Headless end-of-run screenshots verified after each round: leaderboard
Max-troops values, map names + troop counts + flags all render correctly
(including with name slicing active)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Evan
2026-07-05 13:46:02 -07:00
committed by GitHub
parent 571f58440d
commit 22c873cf55
15 changed files with 893 additions and 114 deletions
+1
View File
@@ -17,6 +17,7 @@
"perf:game": "npx tsx tests/perf/fullgame/FullGamePerf.ts",
"perf:client": "npx tsx tests/perf/client/ClientUpdatePerf.ts",
"perf:client-mem": "npx tsx tests/perf/client/ClientMemoryPerf.ts",
"perf:client-tick": "npx tsx tests/perf/client/ClientTickPerf.ts",
"test:coverage": "vitest run --coverage",
"format": "prettier --ignore-unknown --write .",
"format:map-generator": "cd map-generator && go fmt .",
+4 -2
View File
@@ -8,7 +8,6 @@ import type {
NukeTelegraphData,
PlayerState,
PlayerStatusData,
TilePair,
UnitState,
} from "../types";
@@ -21,7 +20,10 @@ export interface FrameUploadTarget {
tileState: Uint16Array,
trailState: Uint16Array,
): void;
uploadLiveDelta(tileState: Uint16Array, changedTiles: TilePair[]): void;
uploadLiveDelta(
tileState: Uint16Array,
changedTiles: readonly number[],
): void;
uploadLiveTrailDelta(
trailState: Uint16Array,
dirtyRowMin: number,
+4 -2
View File
@@ -25,7 +25,6 @@ import type {
PlayerStatic,
PlayerStatusData,
RendererConfig,
TilePair,
UnitState,
} from "../types";
import type { SpawnCenter } from "./passes/SpawnOverlayPass";
@@ -124,7 +123,10 @@ export class MapRenderer {
// ---- Data upload ----
uploadLiveDelta(tileState: Uint16Array, changedTiles: TilePair[]): void {
uploadLiveDelta(
tileState: Uint16Array,
changedTiles: readonly number[],
): void {
this.renderer?.uploadLiveDelta(tileState, changedTiles);
}
uploadLiveTrailDelta(
+4 -2
View File
@@ -23,7 +23,6 @@ import type {
PlayerStatic,
PlayerStatusData,
RendererConfig,
TilePair,
UnitState,
} from "../types";
import { Camera } from "./Camera";
@@ -653,7 +652,10 @@ export class GPURenderer {
this.trailPass.setLiveRef(trailState);
}
uploadLiveDelta(tileState: Uint16Array, changedTiles: TilePair[]): void {
uploadLiveDelta(
tileState: Uint16Array,
changedTiles: readonly number[],
): void {
this.territoryPass.applyLiveDelta(tileState, changedTiles);
}
-1
View File
@@ -26,6 +26,5 @@ export type {
PlayerState,
PlayerStatic,
RendererConfig,
TilePair,
UnitState,
} from "../types";
+5 -3
View File
@@ -12,7 +12,6 @@
* uploads across render frames.
*/
import type { TilePair } from "../../types";
import type { RenderSettings } from "../RenderSettings";
import { getPaletteSize } from "../utils/ColorUtils";
import { createMapQuad, createProgram, shaderSrc } from "../utils/GlUtils";
@@ -229,11 +228,14 @@ export class TerritoryPass {
* Stable per-ref hash means repeated updates to the same tile stay in
* arrival order in the same bucket — last write wins when drained.
*/
applyLiveDelta(tileState: Uint16Array, changedTiles: TilePair[]): void {
applyLiveDelta(
tileState: Uint16Array,
changedTiles: readonly number[],
): void {
const N = this.nBuckets;
const buckets = this.dripBuckets;
for (let i = 0; i < changedTiles.length; i++) {
const ref = changedTiles[i].ref;
const ref = changedTiles[i];
const b = ((ref * 2654435761) >>> 0) % N;
buckets[b].push(ref, tileState[ref]);
}
+74 -72
View File
@@ -81,7 +81,6 @@ export class NamePass {
// Player management
private playerByID: Map<string, PlayerStatic>;
private smallIDToPlayerID: Map<number, string>;
private slots: Map<string, PlayerSlot> = new Map();
private maxPlayers: number;
private playerColors: Map<string, [number, number, number]> = new Map();
@@ -102,16 +101,15 @@ export class NamePass {
private stringRow: Uint8Array;
private cursorRow: Float32Array;
// Reusable per-tick lookup maps (avoid allocation + GC)
private alivePlayerIDs = new Set<string>();
private troopsByPlayerID = new Map<string, number>();
/** Slots refreshed per updateNames call: index % UPDATE_SLICES === phase. */
private static readonly UPDATE_SLICES = 4;
private slicePhase = 0;
// Hovered player's small ID (0 = no highlight, matches TerritoryPass).
private highlightOwnerID = 0;
// Cursor in world coords — fades names under it (far off-map = no fade).
private mouseWorldX = -1e9;
private mouseWorldY = -1e9;
private playerStateByID = new Map<string, PlayerState>();
constructor(
gl: WebGL2RenderingContext,
@@ -148,10 +146,8 @@ export class NamePass {
// Build player lookups and extract territory colors from palette
this.playerByID = new Map();
this.smallIDToPlayerID = new Map();
for (const p of header.players) {
this.playerByID.set(p.id, p);
this.smallIDToPlayerID.set(p.smallID, p.id);
const off = p.smallID * 4;
this.playerColors.set(p.id, [
paletteData[off],
@@ -215,7 +211,6 @@ export class NamePass {
for (const p of players) {
if (this.playerByID.has(p.id)) continue;
this.playerByID.set(p.id, p);
this.smallIDToPlayerID.set(p.smallID, p.id);
const off = p.smallID * 4;
this.playerColors.set(p.id, [
paletteData[off],
@@ -294,75 +289,83 @@ export class NamePass {
): void {
const now = performance.now() / 1000;
// Build alive set and emoji lookup from smallID → playerID
const alivePlayerIDs = this.alivePlayerIDs;
alivePlayerIDs.clear();
const troopsByPlayerID = this.troopsByPlayerID;
troopsByPlayerID.clear();
const playerStateByID = this.playerStateByID;
playerStateByID.clear();
for (const [, ps] of players) {
const pid = this.smallIDToPlayerID.get(ps.smallID);
if (!pid) continue;
if (ps.isAlive) alivePlayerIDs.add(pid);
troopsByPlayerID.set(pid, ps.troops ?? 0);
playerStateByID.set(pid, ps);
}
// Assign slot indices to players (stable ordering by header index)
let nextSlotIndex = 0;
for (const p of this.playerByID.values()) {
if (!this.slots.has(p.id)) {
const slot: PlayerSlot = {
index: nextSlotIndex++,
playerID: p.id,
static: p,
srcX: 0,
srcY: 0,
srcScale: 0,
tgtX: 0,
tgtY: 0,
tgtScale: 0,
startTime: now,
alive: false,
nameLen: 0,
troopLen: 0,
lastTroopStr: "",
lastTroopBucket: -1,
flagUrl: p.flag,
flagLayerIdx: -1,
emojiAtlasIdx: -1,
nameHalfWidth: 0,
crown: false,
traitor: false,
disconnected: false,
alliance: false,
allianceReq: false,
target: false,
embargo: false,
nukeActive: false,
nukeTargetsMe: false,
inDoomsdayClock: false,
doomsdayClockDraining: false,
traitorRemainingTicks: 0,
allianceFraction: 0,
allianceRemainingTicks: 0,
};
this.slots.set(p.id, slot);
this.resolveSlotFlag(slot);
} else {
nextSlotIndex = Math.max(
nextSlotIndex,
this.slots.get(p.id)!.index + 1,
);
// Assign slot indices to players (stable ordering by header index).
// Both maps only ever grow, and every assignment pass leaves them the
// same size — equal sizes means every player already has a slot.
if (this.slots.size !== this.playerByID.size) {
let nextSlotIndex = 0;
for (const p of this.playerByID.values()) {
if (!this.slots.has(p.id)) {
const slot: PlayerSlot = {
index: nextSlotIndex++,
playerID: p.id,
static: p,
srcX: 0,
srcY: 0,
srcScale: 0,
tgtX: 0,
tgtY: 0,
tgtScale: 0,
startTime: now,
alive: false,
nameLen: 0,
troopLen: 0,
lastTroopStr: "",
lastTroopBucket: -1,
flagUrl: p.flag,
flagLayerIdx: -1,
emojiAtlasIdx: -1,
nameHalfWidth: 0,
crown: false,
traitor: false,
disconnected: false,
alliance: false,
allianceReq: false,
target: false,
embargo: false,
nukeActive: false,
nukeTargetsMe: false,
inDoomsdayClock: false,
doomsdayClockDraining: false,
traitorRemainingTicks: 0,
allianceFraction: 0,
allianceRemainingTicks: 0,
};
this.slots.set(p.id, slot);
this.resolveSlotFlag(slot);
} else {
nextSlotIndex = Math.max(
nextSlotIndex,
this.slots.get(p.id)!.index + 1,
);
}
}
}
// Round-robin time slicing: each call refreshes 1/UPDATE_SLICES of the
// slots, spreading the per-player diff work across game ticks (full
// refresh every UPDATE_SLICES ticks ≈ 400 ms — under the 500 ms troop
// text cadence, and name positions lerp continuously anyway). Slots
// without a written name (new player, refreshNames reset) and snap
// passes (seek) are always processed so nothing pops in late.
const phase = this.slicePhase;
this.slicePhase = (phase + 1) % NamePass.UPDATE_SLICES;
for (const [playerID, entry] of names) {
const slot = this.slots.get(playerID);
if (!slot) continue;
const alive = alivePlayerIDs.has(playerID);
if (
!snap &&
slot.nameLen !== 0 &&
slot.index % NamePass.UPDATE_SLICES !== phase
) {
continue;
}
// Per-player state straight from the caller's map — smallID is the key.
const ps = players.get(slot.static.smallID);
const alive = ps?.isAlive ?? false;
// Skip dead players already marked dead — no work needed
if (!alive && !slot.alive) continue;
@@ -394,7 +397,7 @@ export class NamePass {
const troopBucket = Math.floor((now + (slot.index % 5) * 0.1) / 0.5);
if (snap || slot.troopLen === 0 || troopBucket !== slot.lastTroopBucket) {
slot.lastTroopBucket = troopBucket;
const troops = troopsByPlayerID.get(playerID) ?? 0;
const troops = ps?.troops ?? 0;
const troopStr = renderTroops(troops);
if (troopStr !== slot.lastTroopStr) {
slot.troopLen = Math.min(troopStr.length, MAX_CHARS);
@@ -433,7 +436,6 @@ export class NamePass {
// Resolve active broadcast emoji for this player
let newEmoji = -1;
const ps = playerStateByID.get(playerID);
if (ps?.outgoingEmojis && ps.outgoingEmojis.length > 0) {
for (const e of ps.outgoingEmojis) {
if (e.recipientID === "AllPlayers") {
+2 -3
View File
@@ -5,7 +5,6 @@ import type {
NukeTelegraphData,
PlayerState,
PlayerStatusData,
TilePair,
UnitState,
} from "./Renderer";
@@ -37,11 +36,11 @@ export interface FrameData {
// ── Upload hints ──────────────────────────────────────────────────────
/**
* Changed tiles this frame for delta uploads.
* Changed tile refs this frame for delta uploads.
* - `null` → no delta info; full upload needed (first tick)
* - array → only these tiles changed (empty = skip upload)
*/
readonly changedTiles: TilePair[] | null;
readonly changedTiles: readonly number[] | null;
readonly railroadDirty: boolean;
readonly revealedRailTiles: number[];
-5
View File
@@ -173,11 +173,6 @@ export interface ConquestFx {
tickAge?: number;
}
export interface TilePair {
ref: number;
state: number;
}
export interface NameEntry {
playerID: string;
x: number;
-1
View File
@@ -21,7 +21,6 @@ export type {
PlayerStatic,
PlayerStatusData,
RendererConfig,
TilePair,
UnitState,
} from "./Renderer";
+47 -14
View File
@@ -34,7 +34,7 @@ import { computePlayerStatus } from "../render/frame/derive/PlayerStatus";
import { buildRelationMatrix } from "../render/frame/derive/RelationMatrix";
import { RailroadCache } from "../render/frame/RailroadCache";
import { TrailManager } from "../render/frame/TrailManager";
import type { FrameData, NameEntry, TilePair } from "../render/types";
import type { FrameData, NameEntry } from "../render/types";
import { STRUCTURE_TYPES } from "../render/types";
import { PlayerView } from "./PlayerView";
import { UnitView } from "./UnitView";
@@ -81,6 +81,14 @@ export class GameView implements GameMap {
private _teams = new Map<number, string>();
private updatedTiles: TileRef[] = [];
private updatedTerrainTiles: TileRef[] = [];
/**
* Active units grouped by owner smallID, built lazily at most once per
* tick. Keeps per-player unit queries (PlayerView.units) at O(own units)
* instead of O(all units) — the leaderboard asks for every player's units
* in one pass, which is O(players × units) without this.
*/
private _unitsByOwner = new Map<number, UnitView[]>();
private _unitsByOwnerStale = true;
// ── FrameData accumulators (renderer-bound state) ─────────────────────
private trailManager!: TrailManager;
@@ -88,7 +96,6 @@ export class GameView implements GameMap {
/** Long-lived NameEntry map for the renderer's NamePass. */
private _names = new Map<string, NameEntry>();
/** Reusable scratch buffers for per-tick deltas. */
private readonly _changedTilesScratch: TilePair[] = [];
private readonly _trailIdsScratch: number[] = [];
/**
* The single long-lived FrameData object. Fields are mutated in place each
@@ -167,9 +174,9 @@ export class GameView implements GameMap {
this.railroadCache = new RailroadCache(mapW, mapH);
// Long-lived FrameData. Most fields are mutable references to long-lived
// buffers (tileState, trailState, etc.); some (_changedTilesScratch,
// derived arrays) are reused each tick. Properties marked `readonly` on
// FrameData only prevent reassignment, not mutation through the reference.
// buffers (tileState, trailState, etc.); changedTiles points at this
// tick's updatedTiles. Properties marked `readonly` on FrameData only
// prevent reassignment, not mutation through the reference.
// events: fresh arrays we own; cleared and repopulated each tick.
this._frame = {
tick: 0,
@@ -185,7 +192,7 @@ export class GameView implements GameMap {
conquestEvents: [],
bonusEvents: [],
},
changedTiles: this._changedTilesScratch,
changedTiles: null,
railroadDirty: false,
revealedRailTiles: this.railroadCache.revealedRailTiles,
trailDirtyRowMin: 0,
@@ -261,6 +268,9 @@ export class GameView implements GameMap {
}
public update(gu: GameUpdateViewData) {
// Unit set/ownership changes below; rebuild the owner index on demand.
this._unitsByOwnerStale = true;
this.toDelete.forEach((id) => {
this._units.delete(id);
this._unitStates.delete(id);
@@ -437,7 +447,12 @@ export class GameView implements GameMap {
for (const unit of this._units.values()) {
unit._wasUpdated = false;
unit.lastPos = unit.lastPos.slice(-1);
// Only trim when a move appended a position — slicing a ≤1-element
// array would allocate an identical array per unit per tick, and most
// units (structures) never move.
if (unit.lastPos.length > 1) {
unit.lastPos = unit.lastPos.slice(-1);
}
}
gu.updates[GameUpdateType.Unit].forEach((update) => {
let unit = this._units.get(update.id);
@@ -513,12 +528,6 @@ export class GameView implements GameMap {
this._trailIdsScratch,
);
// Changed-tile delta refs (zero-copy: state field unused in live mode).
this._changedTilesScratch.length = 0;
for (let i = 0; i < this.updatedTiles.length; i++) {
this._changedTilesScratch.push({ ref: this.updatedTiles[i], state: 0 });
}
// Names map — rebuilt only when a placement record arrived or a player
// was added (nameData values cannot change between those ticks). Entry
// order is irrelevant for the renderer.
@@ -608,7 +617,9 @@ export class GameView implements GameMap {
f.structuresDirty = true; // force initial structure upload
this._firstPopulate = false;
} else {
f.changedTiles = this._changedTilesScratch;
// Live reference to this tick's changed refs — consumers copy what
// they keep (TerritoryPass buckets them synchronously in the upload).
f.changedTiles = this.updatedTiles;
}
// Reset transient flags for next tick.
@@ -1059,6 +1070,28 @@ export class GameView implements GameMap {
(u) => u.isActive() && types.includes(u.type()),
);
}
/**
* Active units owned by the given player (smallID). The grouping is built
* lazily at most once per tick; the returned array must not be mutated.
*/
unitsOwnedBy(ownerSmallID: number): readonly UnitView[] {
if (this._unitsByOwnerStale) {
this._unitsByOwnerStale = false;
this._unitsByOwner.clear();
for (const u of this._units.values()) {
if (!u.isActive()) continue;
const sid = u.state.ownerID;
const arr = this._unitsByOwner.get(sid);
if (arr === undefined) {
this._unitsByOwner.set(sid, [u]);
} else {
arr.push(u);
}
}
}
return this._unitsByOwner.get(ownerSmallID) ?? [];
}
unit(id: number): UnitView | undefined {
return this._units.get(id);
}
+5 -3
View File
@@ -403,9 +403,11 @@ export class PlayerView {
}
units(...types: UnitType[]): UnitView[] {
return this.game
.units(...types)
.filter((u) => u.owner().smallID() === this.smallID());
const owned = this.game.unitsOwnedBy(this.smallID());
if (types.length === 0) {
return [...owned];
}
return owned.filter((u) => types.includes(u.type()));
}
nameLocation(): NameViewData | undefined {
+89 -6
View File
@@ -663,23 +663,21 @@ describe("GameView.frameData() — renderer contract", () => {
game.update(gu2);
const ct = game.frameData().changedTiles;
expect(ct).not.toBeNull();
expect(ct!.map((t) => t.ref).sort((a, b) => a - b)).toEqual([3, 5, 9]);
expect([...ct!].sort((a, b) => a - b)).toEqual([3, 5, 9]);
});
it("changedTiles scratch array is reused across ticks (no per-tick alloc)", () => {
it("changedTiles holds only the current tick's refs", () => {
const game = makeGameView({ width: 4, height: 4 });
game.update(makeEmptyGu(1)); // first populate (changedTiles = null)
const gu2 = makeEmptyGu(2);
gu2.packedTileUpdates = new Uint32Array([1, 0]);
game.update(gu2);
const ct1 = game.frameData().changedTiles;
expect(game.frameData().changedTiles).toEqual([1]);
const gu3 = makeEmptyGu(3);
gu3.packedTileUpdates = new Uint32Array([2, 0]);
game.update(gu3);
const ct2 = game.frameData().changedTiles;
expect(ct2).toBe(ct1); // same array instance
expect(game.frameData().changedTiles).toEqual([2]);
});
it("frame.units is === gameView.unitStates() (same long-lived map)", () => {
@@ -771,3 +769,88 @@ describe("GameView.frameData() — renderer contract", () => {
expect(relationMatrix[3 * relationSize + 1]).toBe(RELATION_NEUTRAL);
});
});
describe("GameView.unitsOwnedBy — per-tick owner index", () => {
function withUnits(
tick: number,
units: ReturnType<typeof makeUnitUpdate>[],
players: ReturnType<typeof makePlayerUpdate>[] = [],
) {
const gu = makeEmptyGu(tick);
gu.updates[GameUpdateType.Unit] = units;
gu.updates[GameUpdateType.Player] = players;
return gu;
}
it("groups active units by owner smallID", () => {
const game = makeGameView();
game.update(
withUnits(1, [
makeUnitUpdate({ id: 1, ownerID: 1, unitType: UnitType.City }),
makeUnitUpdate({ id: 2, ownerID: 1, unitType: UnitType.Port }),
makeUnitUpdate({ id: 3, ownerID: 2, unitType: UnitType.City }),
]),
);
expect(game.unitsOwnedBy(1).map((u) => u.id())).toEqual([1, 2]);
expect(game.unitsOwnedBy(2).map((u) => u.id())).toEqual([3]);
expect(game.unitsOwnedBy(99)).toEqual([]);
});
it("excludes inactive units", () => {
const game = makeGameView();
game.update(
withUnits(1, [
makeUnitUpdate({ id: 1, ownerID: 1, isActive: true }),
makeUnitUpdate({ id: 2, ownerID: 1, isActive: false }),
]),
);
expect(game.unitsOwnedBy(1).map((u) => u.id())).toEqual([1]);
});
it("reflects ownership changes on the next tick (capture)", () => {
const game = makeGameView();
game.update(withUnits(1, [makeUnitUpdate({ id: 1, ownerID: 1 })]));
expect(game.unitsOwnedBy(1)).toHaveLength(1);
expect(game.unitsOwnedBy(2)).toHaveLength(0);
game.update(withUnits(2, [makeUnitUpdate({ id: 1, ownerID: 2 })]));
expect(game.unitsOwnedBy(1)).toHaveLength(0);
expect(game.unitsOwnedBy(2)).toHaveLength(1);
});
it("reflects unit death on the tick it happens", () => {
const game = makeGameView();
game.update(withUnits(1, [makeUnitUpdate({ id: 1, ownerID: 1 })]));
expect(game.unitsOwnedBy(1)).toHaveLength(1);
game.update(
withUnits(2, [makeUnitUpdate({ id: 1, ownerID: 1, isActive: false })]),
);
expect(game.unitsOwnedBy(1)).toHaveLength(0);
});
it("PlayerView.units() filters the owner's units by type", () => {
const game = makeGameView();
const gu = withUnits(
1,
[
makeUnitUpdate({ id: 1, ownerID: 1, unitType: UnitType.City }),
makeUnitUpdate({ id: 2, ownerID: 1, unitType: UnitType.Port }),
makeUnitUpdate({ id: 3, ownerID: 1, unitType: UnitType.City }),
makeUnitUpdate({ id: 4, ownerID: 2, unitType: UnitType.City }),
],
[
makePlayerUpdate({ id: "alice", smallID: 1 }),
makePlayerUpdate({ id: "bob", smallID: 2 }),
],
);
game.update(gu);
const alice = game.player("alice");
expect(alice.units(UnitType.City).map((u) => u.id())).toEqual([1, 3]);
expect(alice.units().map((u) => u.id())).toEqual([1, 2, 3]);
// Returned arrays are copies — mutating them must not corrupt the index.
alice.units().pop();
expect(alice.units()).toHaveLength(3);
});
});
+90
View File
@@ -0,0 +1,90 @@
// Inclusive-time breakdown of the tick-dispatch subtree in a .cpuprofile
// captured by ClientTickPerf. Attributes every sampled stack under the
// harness's "wrapped" listener to each enclosing function, so the phase
// split (gameView.update vs uploadFrameData vs renderer.tick) is visible.
// Usage: npx tsx tests/perf/client/AnalyzeCpuProfile.ts <file> [rootName]
import fs from "fs";
interface ProfileNode {
id: number;
callFrame: { functionName: string; url: string; lineNumber: number };
hitCount?: number;
children?: number[];
}
interface Profile {
nodes: ProfileNode[];
startTime: number;
endTime: number;
samples?: number[];
timeDeltas?: number[];
}
const file = process.argv[2];
const rootName = process.argv[3] ?? "wrapped";
const p: Profile = JSON.parse(fs.readFileSync(file, "utf8"));
const byId = new Map(p.nodes.map((n) => [n.id, n]));
const parent = new Map<number, number>();
for (const n of p.nodes) {
for (const c of n.children ?? []) parent.set(c, n.id);
}
// Self micros per node id from samples/timeDeltas.
const selfMicros = new Map<number, number>();
const samples = p.samples ?? [];
const deltas = p.timeDeltas ?? [];
for (let i = 0; i < samples.length; i++) {
selfMicros.set(
samples[i],
(selfMicros.get(samples[i]) ?? 0) + (deltas[i] ?? 0),
);
}
// True when a node's ancestry (inclusive) contains a rootName frame.
const underRoot = (id: number): boolean => {
for (
let cur: number | undefined = id;
cur !== undefined;
cur = parent.get(cur)
) {
if (byId.get(cur)?.callFrame.functionName === rootName) return true;
}
return false;
};
// Attribute each in-subtree sample to every enclosing function (inclusive
// time), stopping at the root frame.
let rootTotal = 0;
const inclusiveByFn = new Map<string, number>();
for (const [id, micros] of selfMicros) {
if (!underRoot(id)) continue;
rootTotal += micros;
const seen = new Set<string>();
for (
let cur: number | undefined = id;
cur !== undefined;
cur = parent.get(cur)
) {
const cf = byId.get(cur)?.callFrame;
if (!cf) break;
const url = cf.url.replace(/^.*\/(src|node_modules)\//, "$1/");
const key = `${cf.functionName || "(anonymous)"} ${url}:${cf.lineNumber + 1}`;
if (!seen.has(key)) {
seen.add(key);
inclusiveByFn.set(key, (inclusiveByFn.get(key) ?? 0) + micros);
}
if (cf.functionName === rootName) break;
}
}
console.log(
`total under "${rootName}": ${(rootTotal / 1000).toFixed(1)} ms of ${((p.endTime - p.startTime) / 1000).toFixed(0)} ms profile`,
);
const rows = [...inclusiveByFn.entries()].sort((a, b) => b[1] - a[1]);
for (const [key, micros] of rows.slice(0, 40)) {
const pct = ((micros / rootTotal) * 100).toFixed(1);
console.log(
`${(micros / 1000).toFixed(1).padStart(9)} ms ${pct.padStart(5)}% ${key}`,
);
}
+568
View File
@@ -0,0 +1,568 @@
/**
* Main-thread tick-processing harness: drives a real singleplayer game in
* headless Chromium and measures how long the worker→main "game_update"
* message dispatch takes on the MAIN thread — the per-tick cost that competes
* with the frame budget (16.7 ms at 60 fps) and causes frame drops on low-end
* hardware when it runs long.
*
* What is measured, per dispatch:
* - deserialization: first access to event.data (structured-clone decode)
* - handler: WorkerClient.handleWorkerMessage → gameView.update →
* webglBuilder.update → renderer.tick
* Both are captured by wrapping Worker.prototype.addEventListener in an init
* script, so no product code changes are needed.
*
* For attribution, CDP sampling profiles (chrome .cpuprofile files) can be
* captured over tick windows and are summarized as top self-time functions.
* Open them in Chrome DevTools → Performance → load profile for flame graphs.
*
* The harness starts its own vite dev server on a private port (default
* 9017) so results always come from THIS checkout, even when another
* working copy is serving port 9000.
*
* One-time browser setup (installs playwright + chromium libs, no sudo):
* bash .claude/skills/run-openfront/setup.sh
*
* Usage:
* npm run perf:client-tick -- --map "Giant World Map" --ticks 2000 \
* --window 250 --profile-at 500,1500
*
* Flags:
* --map <name> GameMapType value (default "Giant World Map")
* --bots <n> bot count (default 400, the solo-modal default)
* --difficulty <d> Easy|Medium|Hard|Impossible (default modal default)
* --ticks <n> run until this game tick (default 2000)
* --window <n> report stats every n ticks (default 250)
* --profile-at <list> comma-separated ticks to start a CPU profile
* --profile-window <n> ticks each CPU profile spans (default 100)
* --spawn <x,y> fixed human spawn tile (default: auto-pick)
* --port <n> vite dev-server port (default 9017)
* --raf-interval <ms> rAF throttle; SwiftShader frames cost seconds of
* CPU, so an unthrottled frame loop starves the sim
* (default 3000)
* --out-dir <dir> output dir (default tests/perf/output)
*
* Headless caveats: rendering uses SwiftShader and the rAF loop is throttled,
* so this measures the tick-dispatch path, not draw calls. GL upload calls
* issued inside the dispatch go through SwiftShader's command buffer and cost
* differently than on real GPUs. Solo games are RNG-driven, so numbers vary
* a few percent run-to-run; compare trends, not microseconds.
*/
import { ChildProcess, spawn as spawnProcess } from "child_process";
import fs from "fs";
import path from "path";
interface Options {
map: string;
bots: number;
difficulty: string | undefined;
ticks: number;
window: number;
profileAt: number[];
profileWindow: number;
spawn: { x: number; y: number } | null;
port: number;
rafIntervalMs: number;
outDir: string;
}
/** One worker→main game-update dispatch, as recorded by the init script. */
interface TickSample {
/** Game tick of the (last) update in the dispatch. */
tick: number;
/** ms spent deserializing event.data (structured-clone decode). */
deserMs: number;
/** ms spent in the message handler (gameView/webglBuilder/renderer). */
handlerMs: number;
/** Updates in the dispatch (>1 for game_update_batch catch-up). */
updates: number;
}
interface WindowStats {
label: string;
fromTick: number;
toTick: number;
dispatches: number;
updates: number;
meanMs: number;
p50Ms: number;
p95Ms: number;
maxMs: number;
meanDeserMs: number;
maxDeserMs: number;
/** Mean handler ms per game tick (normalizes batched dispatches). */
meanPerUpdateMs: number;
}
function parseArgs(): Options {
const opts: Options = {
map: "Giant World Map",
bots: 400,
difficulty: undefined,
ticks: 2000,
window: 250,
profileAt: [],
profileWindow: 100,
spawn: null,
port: 9017,
rafIntervalMs: 3000,
outDir: path.join("tests", "perf", "output"),
};
const argv = process.argv.slice(2);
for (let i = 0; i < argv.length; i++) {
const next = () => argv[++i];
switch (argv[i]) {
case "--map":
opts.map = next();
break;
case "--bots":
opts.bots = parseInt(next(), 10);
break;
case "--difficulty":
opts.difficulty = next();
break;
case "--ticks":
opts.ticks = parseInt(next(), 10);
break;
case "--window":
opts.window = parseInt(next(), 10);
break;
case "--profile-at":
opts.profileAt = next()
.split(",")
.map((s) => parseInt(s.trim(), 10))
.filter((n) => Number.isFinite(n));
break;
case "--profile-window":
opts.profileWindow = parseInt(next(), 10);
break;
case "--spawn": {
const [x, y] = next().split(",").map(Number);
opts.spawn = { x, y };
break;
}
case "--port":
opts.port = parseInt(next(), 10);
break;
case "--raf-interval":
opts.rafIntervalMs = parseInt(next(), 10);
break;
case "--out-dir":
opts.outDir = next();
break;
default:
throw new Error(`unknown flag: ${argv[i]}`);
}
}
return opts;
}
// ---------- dev server ----------
async function startViteServer(port: number): Promise<ChildProcess> {
// --strictPort makes vite exit instead of silently picking another port —
// that also guards against measuring a different checkout's server.
const child = spawnProcess(
"npx",
["vite", "--port", String(port), "--strictPort"],
{
env: { ...process.env, SKIP_BROWSER_OPEN: "true" },
stdio: ["ignore", "pipe", "pipe"],
detached: true, // own process group, so cleanup kills vite's children
},
);
let output = "";
child.stdout?.on("data", (d: Buffer) => (output += d.toString()));
child.stderr?.on("data", (d: Buffer) => (output += d.toString()));
const deadline = Date.now() + 60_000;
while (Date.now() < deadline) {
if (child.exitCode !== null) {
throw new Error(
`vite exited with code ${child.exitCode} (port ${port} busy?)\n${output}`,
);
}
try {
const res = await fetch(`http://localhost:${port}/`);
if (res.ok) return child;
} catch {
// not up yet
}
await new Promise((r) => setTimeout(r, 500));
}
throw new Error(`vite did not become ready on port ${port}\n${output}`);
}
function stopViteServer(child: ChildProcess): void {
if (child.pid !== undefined && child.exitCode === null) {
try {
process.kill(-child.pid, "SIGTERM"); // whole process group
} catch {
// already gone
}
}
}
// ---------- stats ----------
function quantile(sorted: number[], q: number): number {
if (sorted.length === 0) return 0;
const idx = Math.min(sorted.length - 1, Math.ceil(q * sorted.length) - 1);
return sorted[Math.max(0, idx)];
}
function summarizeWindow(label: string, samples: TickSample[]): WindowStats {
const handler = samples.map((s) => s.handlerMs).sort((a, b) => a - b);
const deser = samples.map((s) => s.deserMs);
const updates = samples.reduce((acc, s) => acc + s.updates, 0);
const totalHandler = handler.reduce((acc, v) => acc + v, 0);
return {
label,
fromTick: samples.length > 0 ? samples[0].tick : 0,
toTick: samples.length > 0 ? samples[samples.length - 1].tick : 0,
dispatches: samples.length,
updates,
meanMs: samples.length > 0 ? totalHandler / samples.length : 0,
p50Ms: quantile(handler, 0.5),
p95Ms: quantile(handler, 0.95),
maxMs: handler.length > 0 ? handler[handler.length - 1] : 0,
meanDeserMs:
samples.length > 0
? deser.reduce((acc, v) => acc + v, 0) / samples.length
: 0,
maxDeserMs: deser.length > 0 ? Math.max(...deser) : 0,
meanPerUpdateMs: updates > 0 ? totalHandler / updates : 0,
};
}
function printReport(windows: WindowStats[], opts: Options): void {
console.log(
`\n=== Main-thread tick dispatch (map=${opts.map}, bots=${opts.bots}) ===`,
);
console.log(
"handler ms per worker game-update dispatch (frame budget at 60 fps: 16.7 ms)",
);
console.log(
`${"window".padEnd(14)} ${"disp".padStart(5)} ${"upd".padStart(5)} ${"mean".padStart(7)} ${"p50".padStart(7)} ${"p95".padStart(7)} ${"max".padStart(7)} ${"deser".padStart(7)} ${"ms/upd".padStart(7)}`,
);
for (const w of windows) {
console.log(
`${w.label.padEnd(14)} ${String(w.dispatches).padStart(5)} ${String(w.updates).padStart(5)} ${w.meanMs.toFixed(2).padStart(7)} ${w.p50Ms.toFixed(2).padStart(7)} ${w.p95Ms.toFixed(2).padStart(7)} ${w.maxMs.toFixed(2).padStart(7)} ${w.meanDeserMs.toFixed(2).padStart(7)} ${w.meanPerUpdateMs.toFixed(2).padStart(7)}`,
);
}
}
// ---------- CPU profile summarization ----------
interface CpuProfileNode {
id: number;
callFrame: {
functionName: string;
url: string;
lineNumber: number;
};
hitCount?: number;
children?: number[];
}
interface CpuProfile {
nodes: CpuProfileNode[];
startTime: number;
endTime: number;
samples?: number[];
timeDeltas?: number[];
}
/** Print the top-N functions by self time from a V8 sampling profile. */
function summarizeProfile(profile: CpuProfile, top: number): void {
// Self time per node = sum of timeDeltas for samples attributed to it.
const selfMicros = new Map<number, number>();
const samples = profile.samples ?? [];
const deltas = profile.timeDeltas ?? [];
for (let i = 0; i < samples.length; i++) {
const nodeId = samples[i];
selfMicros.set(nodeId, (selfMicros.get(nodeId) ?? 0) + (deltas[i] ?? 0));
}
const byFunction = new Map<string, number>();
for (const node of profile.nodes) {
const micros = selfMicros.get(node.id) ?? 0;
if (micros === 0) continue;
const { functionName, url, lineNumber } = node.callFrame;
const shortUrl = url.replace(/^.*\/(src|node_modules)\//, "$1/");
const key = `${functionName || "(anonymous)"} ${shortUrl}:${lineNumber + 1}`;
byFunction.set(key, (byFunction.get(key) ?? 0) + micros);
}
const totalMicros = profile.endTime - profile.startTime;
const rows = [...byFunction.entries()].sort((a, b) => b[1] - a[1]);
console.log(
` top self-time functions (of ${(totalMicros / 1000).toFixed(0)} ms profiled):`,
);
for (const [key, micros] of rows.slice(0, top)) {
const pct = ((micros / totalMicros) * 100).toFixed(1);
console.log(
` ${(micros / 1000).toFixed(1).padStart(8)} ms ${pct.padStart(5)}% ${key}`,
);
}
}
// ---------- main ----------
async function main(): Promise<void> {
const opts = parseArgs();
fs.mkdirSync(opts.outDir, { recursive: true });
console.log(`starting vite on port ${opts.port}`);
const vite = await startViteServer(opts.port);
// The skill driver reads OPENFRONT_URL at import time, so set it before
// the dynamic imports below.
process.env.OPENFRONT_URL = `http://localhost:${opts.port}`;
const { launch, gotoHome, openSoloModal } =
// @ts-expect-error untyped .mjs skill module
await import("../../../.claude/skills/run-openfront/driver.mjs");
const {
startSoloGame,
waitForGameReady,
spawn,
waitForSpawnPhaseEnd,
waitForTick,
gameState,
} =
// @ts-expect-error untyped .mjs skill module
await import("../../../.claude/skills/run-openfront/game.mjs");
let browser: { close(): Promise<void> } | null = null;
try {
console.log("launching headless chromium…");
const launched = await launch({
rafIntervalMs: opts.rafIntervalMs,
// Headless Chromium throttles backgrounded/occluded pages, which
// starves the singleplayer turn loop on top of SwiftShader's cost.
args: [
"--disable-background-timer-throttling",
"--disable-backgrounding-occluded-windows",
"--disable-renderer-backgrounding",
],
});
browser = launched.browser;
const page = launched.page;
// Surface in-page failures in the harness log — a broken init script or
// GL fault otherwise shows up only as an opaque ready-timeout. Blocked
// external fetches (ads, auth, cosmetics) are expected noise; skip them.
page.on("console", (msg: { type(): string; text(): string }) => {
const text = msg.text();
if (
msg.type() === "error" &&
!/Failed to load resource|Failed to fetch|ramp\./.test(text)
) {
console.log(`CONSOLE[error]: ${text}`);
}
});
// Headless Chromium only has SwiftShader, and the WebGL gate (#4324)
// refuses software renderers by matching the unmasked renderer string.
// Spoof the string so the gate passes; rendering still runs on
// SwiftShader (hence the rAF throttle).
await page.addInitScript(() => {
const orig = WebGL2RenderingContext.prototype.getParameter;
WebGL2RenderingContext.prototype.getParameter = function (p: number) {
const v = orig.call(this, p);
return typeof v === "string"
? v.replace(/swiftshader|llvmpipe|software/gi, "PerfHarnessGPU")
: v;
};
});
// Time every worker→main "message" dispatch. The first event.data access
// pays the structured-clone deserialization, so it is timed separately
// from the handler body. Only game updates are recorded.
//
// MUST stay a string, not a function: tsx compiles this file with
// esbuild's keepNames, which rewrites named function expressions to
// `__name(fn, "name")` — and the `__name` helper doesn't exist in the
// page, so a function-form script throws ReferenceError inside
// Worker.addEventListener and silently kills the game worker setup.
// (Member-expression assignments like the GL spoof above escape
// keepNames, which is why that one can stay a function.)
await page.addInitScript(`(() => {
// Init scripts also run in dedicated workers, where window is
// undefined — only the page context is measured.
if (typeof window === "undefined") return;
const samples = [];
window.__tickPerf = {
take() {
return samples.splice(0, samples.length);
},
};
const origAdd = Worker.prototype.addEventListener;
Worker.prototype.addEventListener = function (type, listener, options) {
if (type !== "message" || typeof listener !== "function") {
return origAdd.call(this, type, listener, options);
}
const wrapped = function (event) {
const t0 = performance.now();
const d = event.data; // first access → structured-clone decode
const t1 = performance.now();
const isTick = d && d.type === "game_update";
const isBatch = d && d.type === "game_update_batch";
if (!isTick && !isBatch) {
return listener.call(this, event);
}
try {
return listener.call(this, event);
} finally {
const t2 = performance.now();
const updates = isBatch ? (d.gameUpdates ?? []) : [d.gameUpdate];
const last = updates[updates.length - 1];
samples.push({
tick: last ? last.tick : -1,
deserMs: t1 - t0,
handlerMs: t2 - t1,
updates: updates.length,
});
// Bounded so a stalled harness can't grow the page's heap.
if (samples.length > 200000) samples.splice(0, 100000);
}
};
return origAdd.call(this, type, wrapped, options);
};
})();`);
// Solo games are fully local: block external requests (ad scripts) and
// all websockets — vite's HMR socket times out under heavy throttling
// and force-reloads the page mid-game.
await page.route("**/*", (route: any) => {
const host = new URL(route.request().url()).hostname;
return host === "localhost" || host === "127.0.0.1"
? route.continue()
: route.abort();
});
await page.routeWebSocket("**", () => {});
// CDP session against the page = the main thread only (the core sim
// worker is a separate target and not included in CPU profiles).
const cdp = await page.context().newCDPSession(page);
await cdp.send("Profiler.enable");
await cdp.send("Profiler.setSamplingInterval", { interval: 200 });
const takeSamples = async (): Promise<TickSample[]> =>
(await page.evaluate(() => (window as any).__tickPerf.take())) ?? [];
console.log("loading home page + solo modal…");
await gotoHome(page);
await openSoloModal(page);
console.log(
`starting solo game: map=${opts.map}, bots=${opts.bots}` +
(opts.difficulty !== undefined
? `, difficulty=${opts.difficulty}`
: ""),
);
try {
await startSoloGame(page, {
map: opts.map,
bots: opts.bots,
...(opts.difficulty !== undefined
? { difficulty: opts.difficulty }
: {}),
});
} catch {
// Giant maps can exceed the skill's 180 s ready timeout on a cold
// headless start — give the load one more window before giving up.
console.log("game not ready after 180 s, waiting another 300 s…");
await waitForGameReady(page, 300_000);
}
console.log("spawning…");
const tile = await spawn(page, opts.spawn);
console.log(`spawned at (${tile.x},${tile.y})`);
await waitForSpawnPhaseEnd(page, 120_000);
const spawnState = await gameState(page);
const spawnedTick: number = spawnState?.ticks ?? 0;
// Spawn-phase dispatches are not representative; drop them.
await takeSamples();
const startWall = Date.now();
const windows: WindowStats[] = [];
const allSamples: TickSample[] = [];
const pendingProfiles = [...new Set(opts.profileAt)].sort((a, b) => a - b);
const captureProfile = async (startTick: number): Promise<void> => {
const endTick = startTick + opts.profileWindow;
console.log(`[profile] capturing ticks ${startTick}${endTick}`);
await cdp.send("Profiler.start");
await waitForTick(page, endTick, opts.profileWindow * 2000 + 120_000);
const { profile } = await cdp.send("Profiler.stop");
const file = path.join(opts.outDir, `client-tick${startTick}.cpuprofile`);
fs.writeFileSync(file, JSON.stringify(profile));
console.log(`[profile] ${file}`);
summarizeProfile(profile as CpuProfile, 25);
};
const targets: number[] = [];
for (let t = spawnedTick + opts.window; t < opts.ticks; t += opts.window) {
targets.push(t);
}
if (opts.ticks > spawnedTick) targets.push(opts.ticks);
for (const target of targets) {
// A due profile splits the window: profile spans real ticks inside it.
while (pendingProfiles.length > 0 && pendingProfiles[0] < target) {
const at = pendingProfiles.shift()!;
const state = await gameState(page);
const current: number = state?.ticks ?? 0;
if (at > current) {
// Generous timeout: headless sim speed varies wildly with map size
// and bot count (0.510 ticks/s).
await waitForTick(page, at, (at - current) * 2000 + 120_000);
}
await captureProfile(Math.max(at, current));
}
const state = await gameState(page);
const current: number = state?.ticks ?? 0;
if (target > current) {
await waitForTick(page, target, (target - current) * 2000 + 120_000);
}
const samples = await takeSamples();
allSamples.push(...samples);
const w = summarizeWindow(`tick ${target}`, samples);
windows.push(w);
console.log(
`[window] ${w.label}: ${w.dispatches} dispatches, mean ${w.meanMs.toFixed(2)} ms, p95 ${w.p95Ms.toFixed(2)} ms, max ${w.maxMs.toFixed(2)} ms`,
);
}
// Any profiles requested at/beyond the final tick.
while (pendingProfiles.length > 0) {
await captureProfile(pendingProfiles.shift()!);
}
const samplesFile = path.join(opts.outDir, "client-tick-samples.json");
fs.writeFileSync(samplesFile, JSON.stringify(allSamples));
console.log(`[samples] ${samplesFile} (${allSamples.length} dispatches)`);
// End-of-run screenshot — a cheap rendering sanity check (a black map
// means the GL pipeline broke even if no pageerror surfaced).
const shotPath = path.join(opts.outDir, "client-tick-final.png");
await page.screenshot({ path: shotPath });
console.log(`[screenshot] ${shotPath}`);
printReport(windows, opts);
printReport([summarizeWindow("overall", allSamples)], {
...opts,
map: `${opts.map} (all windows)`,
});
const finalState = await gameState(page);
console.log(
`\nfinal state: ${JSON.stringify(finalState)}\ntotal wall time ${((Date.now() - startWall) / 1000 / 60).toFixed(1)} min`,
);
} finally {
await browser?.close();
stopViteServer(vite);
}
}
main().catch((err) => {
console.error(err);
process.exit(1);
});