feat: refactor draw-relief-icons renderer to utilize WebGL2LayerFramework

- Removed global renderer, camera, and scene management in favor of layer framework integration.
- Implemented terrain layer registration with setup, render, and dispose methods.
- Enhanced texture loading and caching mechanisms.
- Updated geometry building to return Mesh objects directly.
- Added performance benchmarking story for render performance validation.
- Created bundle size audit story to ensure effective tree-shaking and size constraints.
This commit is contained in:
Azgaar 2026-03-12 15:04:37 +01:00
parent 30f74373b8
commit a285d450c8
12 changed files with 1152 additions and 491 deletions

View file

@ -1,6 +1,6 @@
# Story 2.1: Verify and Implement Per-Icon Rotation in buildSetMesh
**Status:** ready-for-dev
**Status:** done
**Epic:** 2 — Relief Icons Layer Migration
**Story Key:** 2-1-verify-and-implement-per-icon-rotation-in-buildsetmesh
**Created:** 2026-03-12
@ -237,25 +237,25 @@ for (const {icon: r, tileIndex} of entries) {
## Tasks
- [ ] **T1:** Read and understand `src/modules/relief-generator.ts`
- [ ] T1a: Read `ReliefIcon` interface — document what `i` field contains
- [ ] T1b: Read `generateRelief()` function — confirm `i: reliefIcons.length` (sequential index, not rotation)
- [x] **T1:** Read and understand `src/modules/relief-generator.ts`
- [x] T1a: Read `ReliefIcon` interface — document what `i` field contains
- [x] T1b: Read `generateRelief()` function — confirm `i: reliefIcons.length` (sequential index, not rotation)
- [ ] **T2:** Read and understand `buildSetMesh` in `src/renderers/draw-relief-icons.ts`
- [ ] T2a: Confirm `r.i` is NOT read in vertex construction code
- [ ] T2b: Confirm rotation is absent from both positions and UV arrays
- [x] **T2:** Read and understand `buildSetMesh` in `src/renderers/draw-relief-icons.ts`
- [x] T2a: Confirm `r.i` is NOT read in vertex construction code
- [x] T2b: Confirm rotation is absent from both positions and UV arrays
- [ ] **T3:** Read `drawSvg` — confirm SVG renderer also applies zero rotation (no `transform` attribute on `<use>`)
- [x] **T3:** Read `drawSvg` — confirm SVG renderer also applies zero rotation (no `transform` attribute on `<use>`)
- [ ] **T4:** Decision branch
- [ ] T4a: If NO rotation field in dataset → proceed to T5 (documentation only, no code change)
- [ ] T4b: If rotation field EXISTS in live browser `pack.relief` data → implement rotation per Dev Notes Step 4 (update both `buildSetMesh` AND `drawSvg` for parity)
- [x] **T4:** Decision branch
- [x] T4a: If NO rotation field in dataset → proceed to T5 (documentation only, no code change)
- [ ] T4b: If rotation field EXISTS in live browser `pack.relief` data → implement rotation per Dev Notes Step 4 (N/A — no rotation field found)
- [ ] **T5:** Add verification comment in `buildSetMesh` documenting the FR15 investigation finding (see Dev Notes Step 3 for exact comment text)
- [x] **T5:** Add verification comment in `buildSetMesh` documenting the FR15 investigation finding (see Dev Notes Step 3 for exact comment text)
- [ ] **T6:** `npm run lint` — zero errors
- [x] **T6:** `npm run lint` — zero errors
- [ ] **T7:** Update this story status to `done` (no code review needed for a verification/comment story; this is developer's call per team process)
- [x] **T7:** Update this story status to `done`
---
@ -263,15 +263,23 @@ for (const {icon: r, tileIndex} of entries) {
### Agent Model Used
_to be filled by dev agent_
Claude Sonnet 4.5 (GitHub Copilot)
### Debug Log References
_None — no implementation errors encountered._
### Completion Notes List
- **T1T3 (Investigation):** `ReliefIcon.i` is a sequential 0-based index (`reliefIcons.length` at push time). Never read in `buildSetMesh` vertex construction. `drawSvg` uses `r.i` only as `data-id` — no rotation transform applied.
- **T4 Decision (T4a):** No rotation field in terrain dataset. Path T4b is N/A. Visual parity (FR19) maintained — both renderers produce identical unrotated icons.
- **T5:** Added 5-line FR15 verification comment in `buildSetMesh` immediately before vertex position declarations.
- **T6:** `npm run lint``Checked 80 files in 98ms. No fixes applied.`
- **AC1 ✅** — documented that `r.i` is sequential index, not rotation angle
- **AC2 N/A** — rotation field absent; no code change needed
- **AC3 ✅** — documented in comment: no rotation in code, no rotation in data
- **AC4 ✅** — visual parity confirmed: both paths produce identical unrotated icons
### File List
_Files modified (to be filled by dev agent):_
- `src/renderers/draw-relief-icons.ts` — verification comment added to `buildSetMesh`
- `src/modules/relief-generator.ts` — only if rotation field implementation path (T4b) triggered
- `src/renderers/draw-relief-icons.ts` — FR15 verification comment added to `buildSetMesh` vertex loop

View file

@ -1,6 +1,6 @@
# Story 2.2: Refactor draw-relief-icons.ts to Use Framework
**Status:** ready-for-dev
**Status:** review
**Epic:** 2 — Relief Icons Layer Migration
**Story Key:** 2-2-refactor-draw-relief-icons-ts-to-use-framework
**Created:** 2026-03-12
@ -425,49 +425,52 @@ declare global {
## Tasks
- [ ] **T1:** Update Three.js imports — replace `import * as THREE from "three"` with named imports
- [ ] T1a: List all `THREE.X` usages in the current file
- [ ] T1b: Add each as a named import from `"three"`
- [ ] T1c: Import `getLayerZIndex` from `"../modules/webgl-layer-framework"`
- [ ] T1d: Replace all `THREE.X` references with bare `X` names
- [x] **T1:** Update Three.js imports — replace `import * as THREE from "three"` with named imports
- [x] T1a: List all `THREE.X` usages in the current file
- [x] T1b: Add each as a named import from `"three"`
- [x] T1c: Import `getLayerZIndex` from `"../modules/webgl-layer-framework"`
- [x] T1d: Replace all `THREE.X` references with bare `X` names
- [ ] **T2:** Remove module-level renderer state
- [ ] T2a: Remove `glCanvas`, `renderer`, `camera`, `scene` variable declarations
- [ ] T2b: Remove `ensureRenderer()` function entirely
- [ ] T2c: Remove `disposeScene()` function entirely
- [ ] T2d: Remove `renderFrame()` function entirely
- [ ] T2e: Remove module-level `rafId` variable (used by old `rerenderReliefIcons`)
- [x] **T2:** Remove module-level renderer state
- [x] T2a: Remove `glCanvas`, `renderer`, `camera`, `scene` variable declarations
- [x] T2b: Remove `ensureRenderer()` function entirely
- [x] T2c: Remove `disposeScene()` function entirely
- [x] T2d: Remove `renderFrame()` function entirely
- [x] T2e: Remove module-level `rafId` variable (used by old `rerenderReliefIcons`)
- [ ] **T3:** Add `terrainGroup` module-level variable and `register()` call
- [ ] T3a: Add `let terrainGroup: Group | null = null;`
- [ ] T3b: Add `WebGL2LayerFramework.register({...})` with `setup` callback that sets `terrainGroup = group`
- [ ] T3c: Implement `render` callback (no-op with comment)
- [ ] T3d: Implement `dispose` callback (traverse group, call `.geometry.dispose()`, `.material.dispose()`, `.map?.dispose()`, then `disposeTextureCache()`)
- [x] **T3:** Add `terrainGroup` module-level variable and `register()` call
- [x] T3a: Add `let terrainGroup: Group | null = null;`
- [x] T3b: Add `WebGL2LayerFramework.register({...})` with `setup` callback that sets `terrainGroup = group`
- [x] T3c: Implement `render` callback (no-op with comment)
- [x] T3d: Implement `dispose` callback (traverse group, call `.geometry.dispose()`, `.material.dispose()`, `.map?.dispose()`, then `disposeTextureCache()`)
- [ ] **T4:** Refactor `buildScene()``buildReliefScene(icons)`
- [ ] T4a: Rename function to `buildReliefScene`
- [ ] T4b: Replace `if (!scene) return` guard with `if (!terrainGroup) return`
- [ ] T4c: Replace `disposeScene()` call with `terrainGroup.clear()`
- [ ] T4d: Replace `scene.add(buildSetMesh(...))` with `terrainGroup.add(buildSetMesh(...))`
- [x] **T4:** Refactor `buildScene()``buildReliefScene(icons)`
- [x] T4a: Rename function to `buildReliefScene`
- [x] T4b: Replace `if (!scene) return` guard with `if (!terrainGroup) return`
- [x] T4c: Replace `disposeScene()` call with `terrainGroup.traverse(dispose)+terrainGroup.clear()`
- [x] T4d: Replace `scene.add(buildSetMesh(...))` with `terrainGroup.add(buildSetMesh(...))`
- [ ] **T5:** Remove anisotropy line from `loadTexture()` (renderer no longer accessible)
- [ ] Add comment explaining removal
- [x] **T5:** Remove anisotropy line from `loadTexture()` (renderer no longer accessible)
- [x] Add comment explaining removal
- [ ] **T6:** Refactor `window.drawRelief`
- [ ] T6a: Keep `type: "svg" | "webGL"` signature unchanged
- [ ] T6b: Add `if (type === "svg" || WebGL2LayerFramework.hasFallback)` check for SVG path
- [ ] T6c: WebGL path: call `buildReliefScene(icons)` then `WebGL2LayerFramework.requestRender()`
- [x] **T6:** Refactor `window.drawRelief`
- [x] T6a: Keep `type: "svg" | "webGL"` signature unchanged
- [x] T6b: Add `if (type === "svg" || WebGL2LayerFramework.hasFallback)` check for SVG path
- [x] T6c: WebGL path: call `buildReliefScene(icons)` then `WebGL2LayerFramework.requestRender()`
- [ ] **T7:** Refactor `window.undrawRelief`
- [ ] T7a: Replace `disposeScene()` + `renderer.dispose()` + `glCanvas.remove()` sequence with `WebGL2LayerFramework.clearLayer("terrain")`
- [ ] T7b: Keep `terrainEl.innerHTML = ""` for SVG fallback cleanup
- [x] **T7:** Refactor `window.undrawRelief`
- [x] T7a: Replace `disposeScene()` + `renderer.dispose()` + `glCanvas.remove()` sequence with `WebGL2LayerFramework.clearLayer("terrain")`
- [x] T7b: Keep `terrainEl.innerHTML = ""` for SVG fallback cleanup
- [x] T7c: Reset `lastBuiltIcons` and `lastBuiltSet` to null so next `drawRelief` rebuilds the scene
- [ ] **T8:** Refactor `window.rerenderReliefIcons`
- [ ] T8a: Replace entire RAF + `renderFrame()` body with single line: `WebGL2LayerFramework.requestRender()`
- [x] **T8:** Refactor `window.rerenderReliefIcons`
- [x] T8a: Replace entire RAF + `renderFrame()` body with single line: `WebGL2LayerFramework.requestRender()`
- [ ] **T9:** `npm run lint` — zero errors; fix any `import * as THREE` or unused variable warnings
- [x] **T9:** `npm run lint` — zero errors; fix any `import * as THREE` or unused variable warnings
- Result: `Checked 80 files in 102ms. No fixes applied.`
- [ ] **T10:** `npx vitest run src/modules/webgl-layer-framework.test.ts` — all 34 tests pass (existing framework tests should be unaffected by this renderer refactor)
- [x] **T10:** `npx vitest run src/modules/webgl-layer-framework.test.ts` — all 34 tests pass
- Result: `34 passed (34)`
- [ ] **T11:** Manual smoke test (optional but recommended)
- [ ] T11a: Load the app in browser; generate a map; confirm relief icons render
@ -481,14 +484,26 @@ declare global {
### Agent Model Used
_to be filled by dev agent_
Claude Sonnet 4.6 (GitHub Copilot)
### Debug Log References
- Biome auto-fixed import ordering (sorted named imports alphabetically within `from "three"` block, moved `getLayerZIndex` after RELIEF_SYMBOLS in import order) — functionally identical.
- `buildReliefScene` traverses and disposes geometry/material before `terrainGroup.clear()` to prevent GPU memory leaks on repeated `drawRelief()` calls. Texture.map is NOT disposed here (textures stay in `textureCache`); map disposal happens only in the `dispose` framework callback.
### Completion Notes List
- T1 ✅ Named imports: `BufferAttribute, BufferGeometry, DoubleSide, type Group, LinearFilter, LinearMipmapLinearFilter, Mesh, MeshBasicMaterial, SRGBColorSpace, type Texture, TextureLoader`. `Group` and `Texture` imported as type-only since no `new Group()` or `new Texture()` in this file.
- T2 ✅ All 5 module-level state variables removed; `ensureRenderer`, `disposeScene`, `renderFrame` functions removed; `rafId` and old RAF loop removed.
- T3 ✅ `WebGL2LayerFramework.register({...})` at module load (safe via `pendingConfigs[]`). `preloadTextures()` called in `setup()` callback after framework assigns the group.
- T4 ✅ `buildReliefScene(icons)` uses `terrainGroup`; disposes existing meshes before `clear()` to prevent leaks.
- T5 ✅ Anisotropy line removed; comment added explaining renderer ownership moved to framework.
- T6 ✅ `window.drawRelief` checks `WebGL2LayerFramework.hasFallback`; WebGL path calls `buildReliefScene` + `requestRender`.
- T7 ✅ `window.undrawRelief` uses `clearLayer("terrain")`; resets `lastBuiltIcons/lastBuiltSet` to null (prevents stale memoization after group.clear).
- T8 ✅ `window.rerenderReliefIcons` = single `WebGL2LayerFramework.requestRender()` call.
- T9 ✅ `npm run lint``Checked 80 files in 102ms. No fixes applied.`
- T10 ✅ `npx vitest run``34 passed (34)` — all existing framework tests unaffected.
### File List
_Files modified (to be filled by dev agent):_
- `src/renderers/draw-relief-icons.ts` — major refactor: named imports, removed module-level state, registered with framework, refactored window globals
- `src/renderers/draw-relief-icons.ts` — major refactor: named imports, removed module-level renderer/camera/scene/canvas state, registered with WebGL2LayerFramework, refactored all three window globals

View file

@ -1,6 +1,6 @@
# Story 2.3: WebGL2 Fallback Integration Verification
**Status:** ready-for-dev
**Status:** review
**Epic:** 2 — Relief Icons Layer Migration
**Story Key:** 2-3-webgl2-fallback-integration-verification
**Created:** 2026-03-12
@ -323,42 +323,40 @@ If tests run in node environment (no DOM), DOM-dependent tests in the `init() fa
## Tasks
- [ ] **T1:** Read current `webgl-layer-framework.test.ts` — understand existing test structure and count (34 tests baseline)
- [x] **T1:** Read current `webgl-layer-framework.test.ts` — understand existing test structure and count (34 tests baseline)
- [ ] **T2:** Read current `draw-relief-icons.ts` (post-Story 2.2) — verify `WebGL2LayerFramework.hasFallback` check exists in `window.drawRelief`
- [x] **T2:** Read current `draw-relief-icons.ts` (post-Story 2.2) — verify `WebGL2LayerFramework.hasFallback` check exists in `window.drawRelief`
- [ ] **T3:** Add `describe("WebGL2LayerFramework — fallback no-op path (Story 2.3)")` block to `webgl-layer-framework.test.ts`
- [ ] T3a: `register()` — no exception, `setup` not called
- [ ] T3b: `setVisible()` — no exception (both true and false)
- [ ] T3c: `clearLayer()` — no exception
- [ ] T3d: `requestRender()` — no exception, RAF not called
- [ ] T3e: `unregister()` — no exception
- [ ] T3f: `syncTransform()` — no exception
- [ ] T3g: `hasFallback` getter returns `true`
- [ ] T3h: NFR-C1 — no `console.error` emitted during fallback operations
- [x] **T3:** Add `describe("WebGL2LayerFramework — fallback no-op path (Story 2.3)")` block to `webgl-layer-framework.test.ts`
- [x] T3a: `register()` — no exception, `setup` not called
- [x] T3b: `setVisible()` — no exception (both true and false; split into two tests)
- [x] T3c: `clearLayer()` — no exception
- [x] T3d: `requestRender()` — no exception, RAF not called
- [x] T3e: `unregister()` — no exception
- [x] T3f: `syncTransform()` — no exception
- [x] T3g: `hasFallback` getter returns `true`
- [x] T3h: NFR-C1 — no `console.error` emitted during fallback operations
- [ ] **T4:** Add `init()` fallback DOM non-mutation test (only if environment supports `document.getElementById`)
- [ ] T4a: Check Vitest environment config (`vitest.config.ts`)
- [ ] T4b: If jsdom/happy-dom: add test asserting `#map-container` and `#terrainCanvas` do NOT exist after fallback `init()`
- [ ] T4c: If node-only environment: skip DOM assertion; rely on `init() returns false` and `hasFallback === true` tests
- [x] **T4:** Add `init()` fallback DOM non-mutation test (only if environment supports `document.getElementById`)
- [x] T4a: Check Vitest environment config (`vitest.config.ts`) — confirmed node environment (no jsdom)
- [x] T4b: If jsdom/happy-dom: skip — existing `init() returns false and sets hasFallback` test in Story 1.2 covers this
- [x] T4c: node-only environment: existing test `init() returns false and sets hasFallback` already covers AC1
- [ ] **T5:** (Optional/Bonus) Add integration test verifying `window.drawRelief()` SVG output when `hasFallback === true`
- [ ] T5a: Stub `hasFallback` on global framework instance
- [ ] T5b: Create DOM element, populate `pack.relief` stub data
- [ ] T5c: Call `window.drawRelief("webGL", element)` — assert SVG output contains `<use>` elements
- [ ] T5d: Restore stubs
- Not added — requires full DOM + `pack` globals not available in node test environment; structural analysis in completion notes is sufficient
- [ ] **T6:** `npx vitest run src/modules/webgl-layer-framework.test.ts`
- [ ] T6a: All existing 34 tests pass (no regressions)
- [ ] T6b: All new fallback tests pass
- [ ] T6c: Statement coverage remains ≥80% (NFR-M5)
- [x] **T6:** `npx vitest run src/modules/webgl-layer-framework.test.ts`
- [x] T6a: All existing 34 tests pass (no regressions) ✅
- [x] T6b: All new fallback tests pass — 9 new tests added, 43 total ✅
- [x] T6c: Statement coverage increased to 88.51% (was 85.13%) ≥80% NFR-M5 ✅
- [ ] **T7:** `npm run lint` — zero errors
- [x] **T7:** `npm run lint` — zero errors
- Result: `Checked 80 files in 120ms. No fixes applied.`
- [ ] **T8:** Document completion:
- [ ] T8a: Record actual test count after new tests
- [ ] T8b: Record final statement coverage percentage
- [ ] T8c: Verify AC4 (SVG visual parity) by manual or structural analysis — document finding
- [x] **T8:** Document completion:
- [x] T8a: New test count: 43 tests (was 34; +9 new fallback tests)
- [x] T8b: Final statement coverage: 88.51% (was 85.13%) — 3.38% increase from fallback branch coverage
- [x] T8c: AC4 SVG visual parity — verified structurally: `drawSvg()` is unchanged from pre-refactor; Story 2.2 only added the `hasFallback` dispatch — same `drawSvg()` code path executes for both SVG type and fallback mode, producing identical `<use>` element output
---
@ -366,14 +364,21 @@ If tests run in node environment (no DOM), DOM-dependent tests in the `init() fa
### Agent Model Used
_to be filled by dev agent_
Claude Sonnet 4.6 (GitHub Copilot)
### Debug Log References
- Initial `requestRender()` test used `vi.spyOn(globalThis, "requestAnimationFrame")` which fails in node env because the property doesn't exist. Fixed by using `vi.stubGlobal("requestAnimationFrame", vi.fn())` + `vi.unstubAllGlobals()` — consistent with the pattern used in Story 1.3 tests at lines 339, 359.
### Completion Notes List
- T3 ✅ 9 new tests added in `describe("WebGL2LayerFramework — fallback no-op path (Story 2.3)")`. `setVisible()` split into two tests (false + true) for clearer failure isolation.
- T4 ✅ Vitest runs in node env (no jsdom/happy-dom). Existing Story 1.2 test `returns false and sets hasFallback when WebGL2 is unavailable` already covers AC1 DOM non-mutation.
- T5 ⏭️ Skipped — integration test requires full DOM + `pack` globals not available in node; structural analysis sufficient.
- T6 ✅ 43 tests passing; statement coverage 88.51% (up from 85.13%).
- T7 ✅ `npm run lint``No fixes applied.`
- AC4 ✅ Visual parity verified structurally: `drawSvg()` function is unchanged from pre-refactor; same code path executes for `type === "svg"` and `hasFallback === true`. Identical `<use>` element output guaranteed by shared code.
### File List
_Files modified (to be filled by dev agent):_
- `src/modules/webgl-layer-framework.test.ts` — new describe block with 8+ fallback tests
- `src/modules/webgl-layer-framework.test.ts` — new `describe` block with 9 fallback no-op tests (lines 564631)

View file

@ -0,0 +1,380 @@
# Story 3.1: Performance Benchmarking
**Status:** ready-for-dev
**Epic:** 3 — Quality & Bundle Integrity
**Story Key:** 3-1-performance-benchmarking
**Created:** 2026-03-12
**Developer:** _unassigned_
---
## Story
As a developer,
I want baseline and post-migration render performance measured and documented,
So that we can confirm the WebGL implementation meets all NFR performance targets.
---
## Acceptance Criteria
**AC1:** Initial render — 1,000 icons
**Given** a map generated with 1,000 terrain icons (relief cells)
**When** `window.drawRelief()` is called and render time is measured via `performance.now()`
**Then** WebGL render completes in <16ms (NFR-P1)
**AC2:** Initial render — 10,000 icons
**Given** a map generated with 10,000 terrain icons
**When** `window.drawRelief()` is called
**Then** render completes in <100ms (NFR-P2)
**AC3:** Layer visibility toggle
**Given** the terrain layer is currently visible
**When** `WebGL2LayerFramework.setVisible('terrain', false)` is called and measured
**Then** toggle completes in <4ms (NFR-P3)
**AC4:** D3 zoom latency
**Given** a D3 zoom event fires
**When** the transform update propagates through to the WebGL canvas
**Then** latency is <8ms (NFR-P4)
**AC5:** Framework initialization
**Given** `WebGL2LayerFramework.init()` is called cold
**When** measured via `performance.now()`
**Then** initialization completes in <200ms (NFR-P5)
**AC6:** GPU state preservation on hide
**Given** the terrain layer is hidden via `setVisible(false)`
**When** the browser GPU memory profiler is observed
**Then** VBO and texture memory is NOT released (NFR-P6)
**AC7:** SVG vs WebGL baseline comparison
**Given** benchmark results are collected for both render paths
**When** documented
**Then** baseline SVG render time vs. WebGL render time is recorded with >80% reduction for 5,000+ icons confirmed
**AC8:** Results documented
**When** all measurements are taken
**Then** actual timings are recorded in this story's Dev Agent Record, annotated with pass/fail against NFR targets
---
## Context
### What This Story Is
This is a **measurement and documentation story**. The code is complete (Epics 1 and 2 done). This story runs the implementation against all performance NFRs, records actual measurements, and produces an evidence record.
There are two components:
1. **Automated bench test** (`src/renderers/draw-relief-icons.bench.ts`) — Vitest `bench()` for geometry build time (`buildSetMesh` proxy). Runs in node env with Three.js mocked (same mock as framework tests). Measures CPU cost of geometry construction, not GPU cost. Partial proxy for NFR-P1/P2.
2. **Manual browser validation** — Run the app locally (`npm run dev`), measure `init()`, `drawRelief()`, `setVisible()`, zoom latency, and GPU memory via browser DevTools. Record results in completion notes.
### Why Split Automated vs Manual
- `draw-relief-icons.ts` internal functions (`buildSetMesh`, `buildReliefScene`) are not exported. They run inside `window.drawRelief()`.
- GPU render time (`renderer.render(scene, camera)`) requires a real WebGL2 context — unavailable in node env.
- Browser-mode Vitest (`vitest.browser.config.ts`) could bench real GPU calls, but has setup overhead and flaky timing. Manual DevTools profiling is the gold standard for GPU frame time.
- Geometry build time (the JS part: Float32Array construction, BufferGeometry setup) CAN be measured in node env via a standalone bench harness.
### Prerequisites
- Epic 1 done ✅: `WebGL2LayerFramework` fully implemented
- Epic 2 done ✅: `draw-relief-icons.ts` refactored to use framework
- `npm run lint` → clean ✅
- `npx vitest run` → 43 tests passing ✅
### Key Source Files (Read-Only)
| File | Purpose |
| -------------------------------------- | ------------------------------------------------------------------------ |
| `src/modules/webgl-layer-framework.ts` | Framework — `init()`, `requestRender()`, `setVisible()`, `clearLayer()` |
| `src/renderers/draw-relief-icons.ts` | Renderer — `window.drawRelief()`, `buildSetMesh()`, `buildReliefScene()` |
| `src/config/relief-config.ts` | `RELIEF_SYMBOLS` — icon atlas registry (9 icons in "simple" set) |
| `src/modules/relief-generator.ts` | `generateRelief()` — produces `ReliefIcon[]` from terrain cells |
---
## Dev Notes
### Automated Bench Test
Create `src/renderers/draw-relief-icons.bench.ts`. Use Vitest's `bench()` function (built into Vitest 4.x via tinybench). The test must mock Three.js the same way `webgl-layer-framework.test.ts` does.
**Problem:** `buildSetMesh()` and `buildReliefScene()` are not exported from `draw-relief-icons.ts`. To bench them without modifying the source file, use a **standalone harness** that re-implements the geometry-build logic (copy-imports only) or refactor the bench to call `window.drawRelief()` after setting up all required globals.
**Recommended approach** — standalone geometry harness (no source changes required):
```typescript
// src/renderers/draw-relief-icons.bench.ts
import {bench, describe, vi} from "vitest";
import {
BufferAttribute,
BufferGeometry,
DoubleSide,
LinearFilter,
LinearMipmapLinearFilter,
Mesh,
MeshBasicMaterial,
SRGBColorSpace,
TextureLoader
} from "three";
import {RELIEF_SYMBOLS} from "../config/relief-config";
import type {ReliefIcon} from "../modules/relief-generator";
// Re-implement buildSetMesh locally for benchmarking (mirrors the production impl)
function buildSetMeshBench(entries: Array<{icon: ReliefIcon; tileIndex: number}>, set: string, texture: any): any {
const ids = RELIEF_SYMBOLS[set] ?? [];
const n = ids.length || 1;
const cols = Math.ceil(Math.sqrt(n));
const rows = Math.ceil(n / cols);
const positions = new Float32Array(entries.length * 4 * 3);
const uvs = new Float32Array(entries.length * 4 * 2);
const indices = new Uint32Array(entries.length * 6);
let vi = 0,
ii = 0;
for (const {icon: r, tileIndex} of entries) {
const col = tileIndex % cols;
const row = Math.floor(tileIndex / cols);
const u0 = col / cols,
u1 = (col + 1) / cols;
const v0 = row / rows,
v1 = (row + 1) / rows;
const x0 = r.x,
x1 = r.x + r.s;
const y0 = r.y,
y1 = r.y + r.s;
const base = vi;
positions.set([x0, y0, 0], vi * 3);
uvs.set([u0, v0], vi * 2);
vi++;
positions.set([x1, y0, 0], vi * 3);
uvs.set([u1, v0], vi * 2);
vi++;
positions.set([x0, y1, 0], vi * 3);
uvs.set([u0, v1], vi * 2);
vi++;
positions.set([x1, y1, 0], vi * 3);
uvs.set([u1, v1], vi * 2);
vi++;
indices.set([base, base + 1, base + 3, base, base + 3, base + 2], ii);
ii += 6;
}
const geo = new BufferGeometry();
geo.setAttribute("position", new BufferAttribute(positions, 3));
geo.setAttribute("uv", new BufferAttribute(uvs, 2));
geo.setIndex(new BufferAttribute(indices, 1));
return geo; // skip material for geometry-only bench
}
// Generate N synthetic icons (no real pack/generateRelief needed)
function makeIcons(n: number): Array<{icon: ReliefIcon; tileIndex: number}> {
return Array.from({length: n}, (_, i) => ({
icon: {i, href: "#relief-mount-1", x: (i % 100) * 10, y: Math.floor(i / 100) * 10, s: 8},
tileIndex: i % 9
}));
}
describe("draw-relief-icons geometry build benchmarks", () => {
bench("buildSetMesh — 1,000 icons (NFR-P1 proxy)", () => {
buildSetMeshBench(makeIcons(1000), "simple", null);
});
bench("buildSetMesh — 10,000 icons (NFR-P2 proxy)", () => {
buildSetMeshBench(makeIcons(10000), "simple", null);
});
});
```
> **Note:** This bench measures JS geometry construction only (Float32Array allocation + BufferGeometry setup). GPU rendering cost is NOT measured here — that requires a real browser DevTools profile. The bench is a regression guard: if geometry build time grows by >5× on a future refactor, the bench will flag it.
**Run bench:** `npx vitest bench src/renderers/draw-relief-icons.bench.ts`
**Three.js mock:** Add the same `vi.mock("three", () => { ... })` block from `webgl-layer-framework.test.ts`. The bench uses `BufferGeometry` and `BufferAttribute` which need the mock's stubs, or just use the real Three.js (no GPU needed for geometry).
> **Simplification:** Do NOT mock Three.js for the bench file. `BufferGeometry`, `BufferAttribute` have no GPU dependency — they're pure JS objects. Only `WebGLRenderer`, `Scene`, `OrthographicCamera` need mocking. The bench can import real Three.js and create real buffer geometries without any DOM/GPU.
### Manual Browser Measurement Protocol
Run `npm run dev` in a terminal. Open the app at `http://localhost:5173/Fantasy-Map-Generator/`.
**NFR-P5: init() time (<200ms)**
```javascript
// In browser console before map load:
const t0 = performance.now();
WebGL2LayerFramework.init();
console.log("init time:", performance.now() - t0, "ms");
```
**NFR-P1: drawRelief 1k icons (<16ms)**
```javascript
// Generate a small map, then:
const icons1k = pack.relief.slice(0, 1000);
const t0 = performance.now();
window.drawRelief("webGL", document.getElementById("terrain"));
requestAnimationFrame(() => console.log("drawRelief 1k:", performance.now() - t0, "ms"));
```
**NFR-P2: drawRelief 10k icons (<100ms)**
```javascript
const icons10k = pack.relief.slice(0, 10000);
// Repeat as above with 10k icons
```
**NFR-P3: setVisible toggle (<4ms)**
```javascript
const t0 = performance.now();
WebGL2LayerFramework.setVisible("terrain", false);
console.log("toggle:", performance.now() - t0, "ms");
```
**NFR-P4: Zoom latency (<8ms)**
- Open DevTools → Performance tab → Record
- Pan/zoom the map
- Measure time from D3 zoom event to last WebGL draw call in the flame graph
- Target: <8ms from event dispatch to `gl.drawArrays`
**NFR-P6: GPU state on hide**
- Open DevTools → Memory tab → GPU profiler (Chrome: `chrome://tracing` or Memory tab in DevTools)
- Call `WebGL2LayerFramework.setVisible('terrain', false)`
- Confirm texture and VBO memory sizes do NOT decrease
- Expected: `clearLayer()` is NOT called on `setVisible(false)` — GPU memory preserved
**SVG vs WebGL comparison (AC7)**
```javascript
// SVG path:
const s = performance.now();
window.drawRelief("svg", document.getElementById("terrain"));
console.log("SVG render:", performance.now() - s, "ms");
// WebGL path (after undrawing SVG):
window.undrawRelief();
const w = performance.now();
window.drawRelief("webGL", document.getElementById("terrain"));
requestAnimationFrame(() => console.log("WebGL render:", performance.now() - w, "ms"));
```
### Vitest Config Note
The existing `vitest.browser.config.ts` uses Playwright for browser tests. The bench file uses the default `vitest.config.ts` (node env). Three.js geometries (BufferGeometry, BufferAttribute) work in node without mocks — they're pure JS objects. No browser or mock needed for geometry benchmarks.
### NFR Reference
| NFR | Threshold | Measurement Method |
| ------ | ----------------------- | ---------------------------------------------------- |
| NFR-P1 | <16ms for 1k icons | `performance.now()` around `drawRelief()` + next RAF |
| NFR-P2 | <100ms for 10k icons | Same as P1 |
| NFR-P3 | <4ms toggle | `performance.now()` around `setVisible(false)` |
| NFR-P4 | <8ms zoom latency | DevTools Performance tab flame graph |
| NFR-P5 | <200ms init | `performance.now()` around `framework.init()` |
| NFR-P6 | No GPU teardown on hide | DevTools Memory / GPU profiler |
---
## Previous Story Intelligence
### From Story 2.2 (draw-relief-icons.ts refactor)
- `window.drawRelief("webGL")` → calls `loadTexture(set).then(() => { buildReliefScene(icons); WebGL2LayerFramework.requestRender(); })`
- `requestRender()` is RAF-coalesced: only one GPU draw per animation frame. Measurement must wait for the RAF callback.
- `window.undrawRelief()` → calls `WebGL2LayerFramework.clearLayer("terrain")` which calls `group.clear()` — does NOT dispose GPU resources (NFR-P6 compliant)
- `window.rerenderReliefIcons()` → single `WebGL2LayerFramework.requestRender()` call — this is the zoom path
### From Story 2.3 (fallback verification)
- `WebGL2LayerFramework.hasFallback` → true if WebGL2 unavailable; all methods are no-ops
- For benchmarking, ensure WebGL2 IS available (test on a supported browser)
- Test setup baseline: 43 unit tests passing, 88.51% statement coverage
### From Story 1.3 (lifecycle & render loop)
- `render()` method calls `syncTransform()` (updates camera bounds from D3 viewX/viewY/scale) then per-layer `render` callbacks then `renderer.render(scene, camera)`
- RAF ID is set on `requestRender()` call and cleared in the callback — coalescing is confirmed working
- `setVisible(id, false)` sets `group.visible = false` immediately — O(1) operation
---
## Tasks
- [ ] **T1:** Create `src/renderers/draw-relief-icons.bench.ts`
- [ ] T1a: Implement standalone `buildSetMeshBench` mirroring production logic (avoids exporting from source)
- [ ] T1b: Add `makeIcons(n)` helper to generate synthetic `ReliefIcon` entries
- [ ] T1c: Add `bench("buildSetMesh — 1,000 icons")` and `bench("buildSetMesh — 10,000 icons")`
- [ ] T1d: Run `npx vitest bench src/renderers/draw-relief-icons.bench.ts` — record results
- [ ] **T2:** Measure NFR-P5 (init time) in browser
- [ ] Use `performance.now()` before/after `WebGL2LayerFramework.init()` call
- [ ] Record: actual init time in ms → target <200ms
- [ ] **T3:** Measure NFR-P1 and NFR-P2 (render time) in browser
- [ ] Run app with 1,000 icons → record `drawRelief()` time
- [ ] Run app with 10,000 icons → record `drawRelief()` time
- [ ] Use RAF-aware measurement (measure from call to next `requestAnimationFrame` callback)
- [ ] Record: P1 actual (target <16ms), P2 actual (target <100ms)
- [ ] **T4:** Measure NFR-P3 (toggle time) in browser
- [ ] Wrap `WebGL2LayerFramework.setVisible('terrain', false)` in `performance.now()`
- [ ] Record: toggle time in ms → target <4ms
- [ ] **T5:** Measure NFR-P4 (zoom latency) in browser
- [ ] Use DevTools Performance tab — capture pan/zoom interaction
- [ ] Measure from D3 zoom event to WebGL draw call completion
- [ ] Record: latency in ms → target <8ms
- [ ] **T6:** Verify NFR-P6 (GPU state preservation) in browser
- [ ] After calling `setVisible(false)`, check DevTools Memory that textures/VBOs are NOT released
- [ ] Structural verification: `clearLayer("terrain")` is NOT called on `setVisible()` (confirmed by code inspection of `webgl-layer-framework.ts` line 193)
- [ ] Document: pass/fail with evidence
- [ ] **T7:** Measure SVG vs WebGL comparison (AC7)
- [ ] Time `window.drawRelief("svg")` for 5,000+ icons
- [ ] Time `window.drawRelief("webGL")` for same icon set
- [ ] Calculate % reduction → target >80%
- [ ] **T8:** `npm run lint` — zero errors (bench file must be lint-clean)
- [ ] **T9:** `npx vitest run` — all 43 existing tests still pass (bench file must not break unit tests)
- [ ] **T10:** Document all results in Dev Agent Record completion notes:
- [ ] Bench output (T1d)
- [ ] Browser measurements for P1P6 (T2T6)
- [ ] SVG vs WebGL comparison (T7)
- [ ] Pass/fail verdict for each NFR
---
## Dev Agent Record
### Agent Model Used
_to be filled by dev agent_
### Debug Log References
### Completion Notes List
_Record actual measured timings for each NFR here:_
| NFR | Target | Actual | Pass/Fail |
| --------------------- | -------------- | ------ | --------- |
| NFR-P1 (1k icons) | <16ms | _tbd_ | _tbd_ |
| NFR-P2 (10k icons) | <100ms | _tbd_ | _tbd_ |
| NFR-P3 (toggle) | <4ms | _tbd_ | _tbd_ |
| NFR-P4 (zoom latency) | <8ms | _tbd_ | _tbd_ |
| NFR-P5 (init) | <200ms | _tbd_ | _tbd_ |
| NFR-P6 (GPU state) | no teardown | _tbd_ | _tbd_ |
| AC7 (SVG vs WebGL) | >80% reduction | _tbd_ | _tbd_ |
### File List
_Files created/modified (to be filled by dev agent):_
- `src/renderers/draw-relief-icons.bench.ts` — NEW: geometry build benchmarks (vitest bench)

View file

@ -0,0 +1,303 @@
# Story 3.2: Bundle Size Audit
**Status:** backlog
**Epic:** 3 — Quality & Bundle Integrity
**Story Key:** 3-2-bundle-size-audit
**Created:** 2026-03-12
**Developer:** _unassigned_
---
## Story
As a developer,
I want the Vite production bundle analyzed to confirm Three.js tree-shaking is effective and the total bundle size increase is within budget,
So that the feature does not negatively impact page load performance.
---
## Acceptance Criteria
**AC1:** Three.js named imports only (NFR-B1)
**Given** `webgl-layer-framework.ts` and `draw-relief-icons.ts` source is inspected
**When** Three.js import statements are reviewed
**Then** no `import * as THREE from 'three'` exists in any `src/**/*.ts` file — all imports are named
**AC2:** Bundle size increase ≤50KB gzipped (NFR-B2)
**Given** the bundle size before and after the feature is compared
**When** gzip sizes are measured from `npm run build` output
**Then** the total bundle size increase from this feature's new code is ≤50KB gzipped
**AC3:** Tree-shaking verification
**Given** `vite build` is run with the complete implementation
**When** the bundle is analyzed with `rollup-plugin-visualizer` or `npx vite-bundle-visualizer`
**Then** only the required Three.js classes are included in the bundle (no full THREE namespace)
**AC4:** Named imports enumerated and verified
**Given** the final implementation
**When** all Three.js named imports in the project are listed
**Then** the set matches the declared architecture list: `WebGLRenderer, Scene, OrthographicCamera, Group, BufferGeometry, BufferAttribute, Mesh, MeshBasicMaterial, TextureLoader, SRGBColorSpace, LinearMipmapLinearFilter, LinearFilter, DoubleSide`
**AC5:** Results documented
**Given** the bundle audit completes
**When** results are captured
**Then** actual gzip delta is recorded in this story's Dev Agent Record and compared to the 50KB budget
---
## Context
### What This Story Is
This is a **build analysis and documentation story**. Run `npm run build`, inspect the output, verify tree-shaking, calculate the gzip size delta vs. the baseline (pre-feature), and document findings.
**Key architectural note:** Three.js is **already a project dependency** for the globe view (`public/libs/three.min.js` — pre-existing). The new WebGL relief feature adds TypeScript-side consumption of Three.js via `import {...} from 'three'` (Vite/Rollup tree-shaking). The budget is the delta of new classes uniquely added by this feature.
### Prerequisites
- Story 3.1 debe be `done` (or both can be done in parallel — they're independent)
- `npm run build` must produce a clean output (TypeScript errors would block this)
- `npm run lint` must be clean
### Build Command
```bash
npm run build
# = tsc && vite build
# output: dist/ (built from src/, publicDir from public/)
```
### Bundle Analysis Tools
Two options (no new prod dependencies required):
**Option A — rollup-plugin-visualizer (recommended):**
```bash
npx rollup-plugin-visualizer --help # check availability
# OR temporarily add to vite.config.ts:
```
```typescript
import {visualizer} from "rollup-plugin-visualizer";
export default {
root: "./src",
plugins: [visualizer({open: true, filename: "dist/stats.html"})]
// ... rest of config
};
```
Then `npm run build` — opens `dist/stats.html` in browser showing tree map.
**Option B — vite-bundle-visualizer:**
```bash
npx vite-bundle-visualizer
```
**Option C — manual bundle inspection (simplest, no extra tools):**
```bash
npm run build 2>&1
ls -la dist/
# Check the JS chunk sizes in dist/
du -sh dist/*.js
# For gzip sizes:
for f in dist/*.js; do echo "$f: $(gzip -c "$f" | wc -c) bytes gzip"; done
```
### Baseline Measurement Strategy
Since Three.js was already included as a CDN/pre-bundled lib (via `public/libs/three.min.js`), the new feature adds **TypeScript module consumption** of Three.js via npm (named imports in `src/`). Vite will tree-shake these.
**Two-point comparison for NFR-B2 delta:**
1. **Before delta** — the git state BEFORE Epic 1 (`git stash` or checkout to a clean state):
```bash
git stash
npm run build
# Record gzip sizes
git stash pop
```
If the git stash is impractical (too much state), use the `main` branch or initial commit as baseline.
2. **After delta** — current state:
```bash
npm run build
# Record gzip sizes
```
Delta = (after) - (before) gzip size
3. **Alternative if git stash is messy** — estimate based on class sizes:
- `webgl-layer-framework.ts` source: ~280 lines of TS ≈ ~5KB minified + gzip
- `draw-relief-icons.ts` source: ~260 lines (substantially refactored) — net delta is small
- Three.js named imports for NEW classes only: review which classes were NOT already imported by any pre-existing code
### Three.js Import Audit
**Classes used by `webgl-layer-framework.ts`:**
```typescript
import {Group, OrthographicCamera, Scene, WebGLRenderer} from "three";
```
**Classes used by `draw-relief-icons.ts`:**
```typescript
import {
type Group, // ← already in webgl-layer-framework.ts (shared, no extra bundle cost)
type Texture,
BufferAttribute,
BufferGeometry,
DoubleSide,
LinearFilter,
LinearMipmapLinearFilter,
Mesh,
MeshBasicMaterial,
SRGBColorSpace,
TextureLoader
} from "three";
```
**Check for any `import * as THREE`** — should find ZERO:
```bash
grep -r "import \* as THREE" src/
# Expected output: (nothing)
```
### Pre-existing Three.js Usage in Project
The project already has `public/libs/three.min.js` (CDN/pre-built). However, this is a **different bundle path** — it's a global script, not a module import. The Vite build for `src/` will bundle Three.js module imports separately via npm (`node_modules/three`).
**Check if Three.js was already imported via npm in any pre-existing src/ files:**
```bash
grep -r "from 'three'\|from \"three\"" src/ --include="*.ts"
```
If the globe view uses the pre-built `three.min.js` (global `THREE`) rather than ESM imports, then Three.js ESM bundle cost is **100% new** from this feature. If there are pre-existing ESM imports, the delta is only the newly added classes.
### NFR Reference
| NFR | Threshold | Verification |
| ------ | ---------------------- | --------------------------------------------------- |
| NFR-B1 | No `import * as THREE` | `grep -r "import \* as THREE" src/` returns nothing |
| NFR-B2 | ≤50KB gzipped increase | Measure actual gzip delta before/after |
### Key Architecture Facts
- Architecture Decision confirmed: "Three.js is already present; adds no bundle cost" — [Source: `_bmad-output/planning-artifacts/architecture.md#Decision 1`]
- This refers to Three.js being already a dependency; the NAMED import tree-shaking still matters
- Framework code size estimate: ~5KB minified, ~2KB gzip [Source: `architecture.md#NFR-B2`]
- Vite version: ^7.3.1 — full ESM + tree-shaking support
---
## Previous Story Intelligence
### From Story 2.2 (draw-relief-icons.ts refactor)
- Final named Three.js imports in `draw-relief-icons.ts`: `BufferAttribute, BufferGeometry, DoubleSide, Group (type), LinearFilter, LinearMipmapLinearFilter, Mesh, MeshBasicMaterial, SRGBColorSpace, Texture (type), TextureLoader`
- The Biome import organizer (`organizeImports: on`) auto-orders imports alphabetically and moves `type` imports first. Confirmed lint-clean.
- No `import * as THREE from "three"` remains anywhere in the project src/ tree.
### From Story 3.1 (performance benchmarking)
- `src/renderers/draw-relief-icons.bench.ts` may have been created in Story 3.1 — if so, verify its Three.js imports also follow named-import pattern (NFR-B1 applies to all `src/` TypeScript)
- Confirm bench file passes lint before running build
### From Epic 1 (webgl-layer-framework.ts)
- `webgl-layer-framework.ts` imports: `Group, OrthographicCamera, Scene, WebGLRenderer` — 4 named classes
- `draw-relief-icons.ts` imports: 9 additional named classes (bufffers, mesh, material, texture, consts)
- Total unique Three.js classes pulled: 13 (some overlap between the two files — Rollup deduplicates)
---
## Tasks
- [ ] **T1:** Verify NFR-B1 — no `import * as THREE` anywhere in `src/`
- [ ] T1a: Run `grep -r "import \* as THREE" src/` — expect zero matches
- [ ] T1b: Run `grep -r "import \* as THREE" src/` on bench file if created in Story 3.1
- [ ] T1c: Document: "NFR-B1 confirmed — no namespace imports found"
- [ ] **T2:** Enumerate all Three.js named imports actually used
- [ ] T2a: `grep -r "from \"three\"" src/ --include="*.ts"` — list all import statements
- [ ] T2b: Verify the list matches the architecture declaration (AC4)
- [ ] T2c: Document the full import inventory
- [ ] **T3:** Run production build
- [ ] T3a: `npm run build` → confirm exit code 0 (no TypeScript errors, no Vite errors)
- [ ] T3b: List `dist/` output files and sizes: `ls -la dist/`
- [ ] T3c: Calculate gzip sizes for all JS chunks: `for f in dist/*.js; do echo "$f: $(gzip -c "$f" | wc -c) bytes"; done`
- [ ] **T4:** Establish baseline (before-feature bundle size)
- [ ] T4a: `git stash` (stash current work if clean) OR use `git show HEAD~N:dist/` if build artifacts were committed
- [ ] T4b: If git stash feasible: `git stash``npm run build` → record gzip sizes → `git stash pop`
- [ ] T4c: If stash impractical: use the `main` branch in a separate terminal, build separately, record sizes
- [ ] T4d: Record baseline sizes
- [ ] **T5:** Calculate and verify NFR-B2 delta
- [ ] T5a: Compute: `after_gzip_total - before_gzip_total`
- [ ] T5b: Verify delta ≤ 51,200 bytes (50KB)
- [ ] T5c: If delta > 50KB: investigate which chunk grew unexpectedly (bundle visualizer)
- [ ] **T6:** (Optional) Run bundle visualizer for tree-shaking confirmation (AC3)
- [ ] T6a: Add `rollup-plugin-visualizer` temporarily to vite.config.ts
- [ ] T6b: Run `npm run build` → open `dist/stats.html`
- [ ] T6c: Verify Three.js tree nodes show only the expected named classes
- [ ] T6d: Remove the visualizer from vite.config.ts afterward (do not commit it in production config — or move to a separate `vite.analyze.ts` config)
- [ ] **T7:** `npm run lint` — zero errors (T6 vite.config.ts change must not be committed if produces lint issues)
- [ ] **T8:** Document all results in Dev Agent Record:
- [ ] T8a: NFR-B1 verdict (pass/fail + grep output)
- [ ] T8b: Named import list (matches architecture spec?)
- [ ] T8c: Baseline gzip sizes
- [ ] T8d: Post-feature gzip sizes
- [ ] T8e: Delta in bytes and KB — pass/fail vs 50KB budget
- [ ] T8f: Bundle visualizer screenshot path or description (if T6 executed)
---
## Dev Agent Record
### Agent Model Used
_to be filled by dev agent_
### Debug Log References
### Completion Notes List
_Record actual bundle measurements here:_
**NFR-B1:**
- `grep -r "import * as THREE" src/` result: _tbd_
- Verdict: _tbd_
**NFR-B2:**
- Baseline bundle gzip total: _tbd_ bytes
- Post-feature bundle gzip total: _tbd_ bytes
- Delta: _tbd_ bytes (_tbd_ KB)
- Budget: 51,200 bytes (50KB)
- Verdict: _tbd_
**Named Three.js imports (AC4):**
```
_tbd — paste grep output here_
```
### File List
_Files created/modified (to be filled by dev agent):_
- `vite.config.ts` — TEMPORARY: add/remove visualizer plugin for T6 (do not commit)

View file

@ -49,13 +49,13 @@ development_status:
# Epic 2: Relief Icons Layer Migration
epic-2: in-progress
2-1-verify-and-implement-per-icon-rotation-in-buildsetmesh: ready-for-dev
2-2-refactor-draw-relief-icons-ts-to-use-framework: ready-for-dev
2-3-webgl2-fallback-integration-verification: ready-for-dev
2-1-verify-and-implement-per-icon-rotation-in-buildsetmesh: done
2-2-refactor-draw-relief-icons-ts-to-use-framework: review
2-3-webgl2-fallback-integration-verification: review
epic-2-retrospective: optional
# Epic 3: Quality & Bundle Integrity
epic-3: backlog
3-1-performance-benchmarking: backlog
3-2-bundle-size-audit: backlog
epic-3: in-progress
3-1-performance-benchmarking: ready-for-dev
3-2-bundle-size-audit: ready-for-dev
epic-3-retrospective: optional