bimplus-renderer 1.8.38-updatepackages1.4 → 1.8.38

package/KARMA_TO_VITEST_MIGRATION_PLAN.md

@@ -1,235 +0,0 @@
-# Karma to Vitest Migration Plan
-
-Date: 2026-10-08
-Status: Vitest is the local default; Karma packages and configuration have been removed. Jenkins validation remains open.
-
-## Implementation Progress
-
-- `npm test`, Firefox/Chrome, watch, coverage, and modelviewer aliases now invoke Vitest. Removed all Karma packages, QUnit, Karma-only loaders, and the five Karma configuration files; `npm ci` installs 877 packages and confirms Karma/QUnit are absent.
-- Vitest Browser Mode profiles are serial. Renderer and performance sources now use native Vitest imports, hooks, assertions, and awaited promises; the renderer projects still discover the existing regular and performance suites, while normal and long modelviewer DDT remain separate.
-- Removed the QUnit compatibility adapter and its global `describe`/`it` registration. `test/setup/vitestBrowser.setup.js` now contains only browser globals, the DOM fixture lifecycle, and custom Vitest matchers. Polly lifecycle uses Vitest hooks and a suite-local state object; no active QUnit assertion or hook API references remain in test JavaScript.
-- Focused Chrome validation after removing the adapter passed 11 files / 96 tests across color utilities, basic viewport lifecycle, overlay behavior, crop-box interactions, event timing, frame selection, three-content loading, sun calculations, camera math, and performance object-container checks. Updated test sources and helpers pass the full repository ESLint script.
-- Local Firefox and Chrome each pass the 8-test pilot after the clean install: MathExtensions, WebGL2 pixel readback, shader rendering, real viewport initialization/render/disposal, and direct installed `web-ifc` geometry parsing.
-- The existing `Ifc loader basic test` passes in Firefox under the compatibility setup, including ArrayBuffer fixture imports and `/base` WASM paths. The BasicProject modelviewer DDT passes using installed `web-ifc` 0.0.77 and its package-matched WASM.
-- Full Polly-backed runtime parity remains open: focused Polly runs fall through to live `api-dev.bimplus.net/v2/authorize` and receive a network failure/401, consistent with the known recording issue. The converted Polly suites collect under Vitest; refreshed/repaired recordings are needed for full request/replay validation.
-- Firefox Istanbul coverage is configured but still includes only `src/utils/mathExtensions.js`; full source coverage parity and Jenkins artifact integration are not complete.
-- The full normal package-matched DDT checks all 35 files and finds two `numVertices` deltas: `64142279-5316-435f-afea-ca7dab4af9db.ifc` expected 4,233,782, actual 1,026,218; `Quickstart.ifc` expected 769,077, actual 502,101. Karma's historical copied `.75` WASM passed those references; the `.77` package-matched Vitest test intentionally leaves the discrepancy visible. No reference was written or changed.
-- Still pending: full renderer execution with refreshed/repaired Polly recordings, full Chrome and long DDT verification, an owner decision on reviewed `.77` references, full coverage/artifact parity, and actual Jenkins Firefox validation. No Jenkins pipeline file was present in this workspace.
-
-## Objectives
-
-- Replace Karma and QUnit with Vitest without losing existing behavioral checks.
-- Keep Firefox as the required Jenkins browser. Chrome has WebGL/memory problems on that server and must not become the CI default or an automatic fallback.
-- Provide local Google Chrome and Firefox runs, including watch/debug workflows.
-- Preserve the separate modelviewer regression suite that validates the installed `web-ifc` package against real IFC files and stored results.
-- Retain coverage, recorded-network tests, performance checks, and both normal and long DDT datasets.
-- Leave production webpack builds unchanged. Migrating the application build to Vite is outside this work.
-
-## Karma Baseline Evidence (Pre-migration)
-
-| Surface | Current behavior | Migration implication |
-| --- | --- | --- |
-| `package.json` | Main tests start Polly and Karma together; Chrome, Firefox, headless, and watch helpers exist. Karma processes use an 8 GB Node heap limit. | Preserve public workflows and service lifecycle; distinguish Node heap from browser/GPU/WASM memory. |
-| `karma.conf.js` | QUnit main suite, 60-second test timeout, explicit assertion counts, fixture clearing disabled, HTML/LCOV coverage, six `FirefoxWebGL` entries. | Convert assertions and hooks; preserve fixture semantics and measure browser recycling rather than copying the browser list. |
-| `karma-shared-config.js` | Sharding, jQuery 3.4.0, host environment, webpack; concurrency is one. Firefox launcher forces WebGL/WebGL2 and acceleration and disables the GPU sandbox. | Begin serially, retain required setup, and explicitly validate Firefox launch preferences on Jenkins. |
-| `karma-files-config.js` | Broad test loading includes `test/tests`, `test/performance`, helpers, custom assertions, moment-timezone setup, and fixtures. | Inventory all registered tests; explicit Vitest discovery must not accidentally omit performance or nested suites. Helpers/setup are not test entries. |
-| `karma.webpack.js` | Three.js aliases; binary IFC/OBJ imports; image, shader/chunk, and coverage loaders. | Implement equivalent test-only Vite resolution/transforms and asset serving. |
-| `test/helpers/viewportHelper.js` | Requires `qunit-fixture`; reuses a 1000 x 700 `viewport-container`. | Create a compatible DOM fixture before viewport construction and preserve layout. |
-| `test/helpers/pollyHelper.js` | Uses `setupQunit`, binary XHR adapter, REST persister, payload hashes, recording sanitization, and `recordIfMissing: true`. | Replace the framework lifecycle adapter, not the existing recording format or request matching. |
-| `test/helpers/testHelper.js` | Reads `host.env.BIM_PASS`. | Replace Karma host injection with an explicit, minimal test configuration bridge. |
-| `test/setup/momentTimezoneInjector.js` | CommonJS import injects `moment` and `MomentTZ` globals. | Convert test setup to ESM and run before dependent modules. |
-| `test/tests/IfcContentLoaderTests.js` | Real IFC byte imports, WASM initialization, model/layer/geometry checks, unload/disposal. | Keep this browser integration coverage in addition to modelviewer DDT. |
-| `karma-modelviewer-ddt.conf.js` | Separate Firefox suite; test timeout is 6,000,000 ms; no coverage reporter; same shared setup. | Maintain a separately selectable heavy integration project and initially preserve its timeout budget. |
-| `modelviewer-data-driven-tests/tests/loadIfcTests.js` | Loads all files from the DDT server, opens a project, imports IFC, checks diagnostics and stored geometry/error results, then disposes the viewer. | Preserve actual end-to-end `web-ifc` validation and clean up even after failures. |
-| `package.json` DDT scripts | Sibling `../simple-ddt-server`, port 3456; separate normal/long dataset folders and sync commands. | Document/provision the sibling dependency and run only one dataset server on this port at a time. |
-
-The Jenkins job definition and agent display/GPU setup have not been established from the inspected files. Confirm them with the pipeline owner during Phase 0; do not assume a Linux agent or headless WebGL support.
-
-## Target Architecture
-
-Use Vitest Browser Mode with the Playwright provider as the initial candidate. Real renderer tests must run in a real browser, not jsdom/happy-dom, and must not replace WebGL or IFC parsing with mocks.
-
-Use compatible, pinned versions of `vitest`, `@vitest/browser-playwright`, `playwright`, and `@vitest/coverage-istanbul`; validate their Node engine and peer requirements against developer machines and Jenkins before choosing versions. Keep Vitest companion packages aligned. Use an ESM `vitest.config.mjs` so the existing CommonJS webpack/Karma configuration can remain intact during transition.
-
-Define independently selectable projects/profiles:
-
-| Project/profile | Test scope | Browser and execution |
-| --- | --- | --- |
-| Renderer Firefox | Converted regular tests, including Polly and IFC loader checks | Jenkins default; headed Firefox, serial files/tests. |
-| Renderer Chrome | Same regular scope | Local Google Chrome; Playwright `browser: 'chromium'` with launch `channel: 'chrome'`, not merely bundled Chromium. |
-| Modelviewer Firefox | Converted `loadIfcTests.js`; normal or long dataset selected explicitly | Separate serial Jenkins stage; headed Firefox; no routine coverage. |
-| Modelviewer Chrome | Same DDT behavior | Local Chrome debugging/verification with either dataset. |
-| Performance | Existing `test/performance` checks | Separate serial invocation with controlled conditions; preserve current CI coverage until job owners approve any scheduling change. |
-| Optional pure unit | Only tests proven independent of DOM, WebGL, SDK/browser globals, and WASM integration | Node project, added after browser parity; no initial blanket relocation of utility tests. |
-
-Only select one browser profile in a memory-constrained run. Do not launch all projects/browser instances by default. Initially use `fileParallelism: false`, `maxWorkers: 1`, and no concurrent test declarations; also sequence projects/stages so separate invocations cannot overlap.
-
-### Browser and Memory Policy
-
-- Set `browser.headless: false` explicitly for required WebGL runs, including CI, because repository comments describe headless failures/stalls. Do not rely on CI-dependent defaults.
-- For Linux Jenkins, use an approved display/Xvfb setup if needed; on Windows use a provisioned interactive desktop/session. A virtual display alone does not guarantee WebGL2 or sufficient GPU memory.
-- Playwright requires its supported Firefox build; stock system Firefox is not a drop-in replacement. Install/cache the browser revision matched to the Playwright version and revalidate after upgrades.
-- Map the existing Firefox preferences through Playwright `launchOptions.firefoxUserPrefs`: `webgl.force-enabled`, `webgl.enable-webgl2`, and `layers.acceleration.force-enabled`. Treat `security.sandbox.gpu.level: 0` as a security-sensitive compatibility setting: retain only on the controlled test agent if validation proves it necessary and the owner approves it.
-- If Playwright Firefox fails the Jenkins WebGL pilot, test Vitest's WebdriverIO provider against the approved system Firefox/geckodriver setup. Do not switch Jenkins to Chrome. Keep Karma Firefox as the temporary gate until a viable provider is proven.
-- Browser Mode isolates files, not individual tests within a file. Keep file isolation initially; explicitly dispose viewers, renderers, WebGL contexts where owned, WASM models/APIs, workers, listeners, and timers in awaited teardown.
-- Measure process/GPU/WASM memory over repeated files and large IFC loads. A single serial browser can still accumulate resources; if necessary, use bounded sequential batches with browser-process restarts and correct report merging.
-- The existing six launcher entries and sharding are not evidence of six concurrent browsers: Karma limits concurrency to one. Do not mechanically reproduce them as parallel Vitest instances.
-- Keep Chrome GC exposure only if useful for existing `GC.run()`. Firefox cleanup must not depend on a global `gc`; raising Node's heap limit does not fix Chrome's browser/WebGL memory issue.
-- Headless Chrome/Firefox commands can remain experimental options, but are not required Jenkins gates until separately proven stable on the target hardware.
-
-## Compatibility Work
-
-### Test Discovery and QUnit Conversion
-
-1. Record the actual Karma module/test names, counts, skips/todos, and performance cases. Use runner output rather than treating every helper JavaScript file as a test.
-2. Preserve paths where practical. Use explicit includes or a migration allowlist for converted files; Vitest's default `*.test.*` discovery will not find existing `*Tests.js` files or `loadIfcTests.js` automatically.
-3. Convert `QUnit.module` to `describe`, tests to `test`, and hooks to `beforeAll`/`afterAll`/`beforeEach`/`afterEach`. Audit nested hooks, suite state, callbacks, and QUnit `this` usage; Vitest does not provide QUnit's context automatically.
-4. Replace `assert.async()` with awaited promises or async tests. Convert `assert.expect()` to `expect.assertions()` only after accounting for assertions in hooks: QUnit currently counts hook assertions as well. Preserve failure detection instead of blindly copying counts.
-5. Audit loose QUnit `equal` separately from `strictEqual`; do not silently introduce or remove coercion. Preserve deep equality, rejection, exception, skip/todo, and diagnostic-message semantics.
-6. Port float/vector assertions with `expect.extend`, preserving strict epsilon comparisons and the current `1e-7` default. `toBeCloseTo` is not an exact semantic replacement.
-7. Remove QUnit runtime dependencies from migrated helpers. Avoid a broad QUnit compatibility shim that leaves two runners controlling lifecycle.
-8. During coexistence, use mutually exclusive Karma and Vitest discovery lists so converted files are not loaded by Karma's current broad glob and unconverted QUnit files are not loaded by Vitest. Compare combined inventories for omissions/duplicates.
-
-### DOM, Globals, Assets, and Workers
-
-- Supply `qunit-fixture` and the expected viewport container through browser setup initially; changing fixture names is optional follow-up, not required migration work. Preserve fixture reuse within a suite rather than clearing it before renderer disposal.
-- Configure a stable browser viewport large enough for the 1000 x 700 fixture. Verify mouse offsets, canvas dimensions, pixel reads, screenshots, and font/image readiness in real Firefox and Chrome.
-- Supply jQuery explicitly only where needed, replacing Karma's injected global. Initialize moment-timezone before importing dependent tests; verify sun/timezone-sensitive cases.
-- Replace `host.env` with an allowlisted configuration bridge. Never expose all `process.env`, use a public `VITE_*` secret, or bake credentials into persistent bundles/reports. CI replay should not require live credentials; explicit local recording may receive only the necessary ephemeral credential.
-- Recreate the exact Three.js aliases from `karma.webpack.js` with ESM-safe resolution.
-- Replace shader-loader with a Vite-compatible transform that preserves the existing shader chunk/include semantics; raw text imports alone may not be equivalent. Validate by compiling/rendering a shader in the pilot.
-- Replace IFC/OBJ ArrayBuffer imports with an explicit binary fixture helper or transform. Preserve the byte contract expected by `new Uint8Array(...)`; a Vite asset URL is not an ArrayBuffer. Ensure large IFC fixtures are fetched rather than embedded in huge generated modules.
-- Serve existing fixture URLs (`/data/`, and any remaining `/base/...` paths) through explicit test-only mappings or migrate call sites deliberately. Preserve image/font/binary MIME types and byte contents.
-- Serve `/web-ifc.wasm` and `/web-ifc-mt.wasm` from the installed package's matching binaries. Validate that responses are WASM bytes, not an HTML fallback. If tests keep `src/wasm`, run the existing `copy-wasm-node-modules` preparation and verify those copies match the locked `web-ifc` version.
-- Audit generated worker/bin paths and Vite module/worker loading before converting worker tests. Reuse existing worker build preparation when required; do not assume Karma exclusions eliminate runtime worker dependencies. Compile Emscripten artifacts only where the chosen test scope actually needs them.
-- Confirm whether the tested IFC path uses single-threaded or multi-threaded WASM. Configure COOP/COEP/SharedArrayBuffer support only when needed and verify it does not block cross-origin DDT/Polly/assets.
-
-### Recorded-Network Tests
-
-- Keep Polly, the binary XHR adapter, REST persister, existing HAR files, payload hashing, SDK cache behavior, binary response conversion, and secret sanitization.
-- Replace `setupQunit` with explicit Vitest hooks: create and attach Polly before each test, configure it, then await flush/stop in teardown even when the test fails.
-- Preserve existing recording IDs independently of Vitest's display names. Validate a representative GLB replay against the current HAR without re-recording; renamed suites must not silently generate a new corpus.
-- Keep `polly listen` for relevant scopes, with a readiness probe before browser execution and reliable cleanup afterward. Check origin/CORS and prevent asset/WASM requests being accidentally captured.
-- Make CI replay-only (`recordIfMissing: false`) with no unintended live-network fallback. Recording/update is an explicit local action; it must retain sanitization and require review of changed HAR files.
-
-## Modelviewer / web-ifc Regression Contract
-
-This is an end-to-end compatibility gate, not a mocked loader test or a module-import smoke test. It must exercise `ProjectViewer.openProject`, `loadIfcFile`, real WASM parsing and geometry creation, `setupProjectView`, diagnostics, and disposal.
-
-1. Keep normal and long datasets and the existing sync workflows separate. Provision the sibling DDT server, its dependencies, IFC files, and reference results explicitly on Jenkins; missing/empty data must fail the stage.
-2. Start the server on port 3456 with the selected folder; verify `/testdata` readiness before launching Vitest. Do not start normal and long servers together. Validate cross-origin access from the Vitest browser server.
-3. Initially preserve the serial per-file loop to avoid concurrent WASM heaps and changed reference semantics. Per-file Vitest cases can be introduced later with deterministic enumeration and identical baseline keys.
-4. Replace `QUnit.config.current.testName + "_0"` with an explicit stable legacy key, `Ifc loader basic test_0`. Verify a read against existing stored references before changing naming; do not derive persistence keys from new Vitest suite titles.
-5. Preserve comparisons of layer/object/type/submesh/vertex/face counts, rounded bounding-box dimensions, nonnegative per-object counts, and accepted IFC loader errors. Preserve the order: compare accepted errors independently, then compare geometry results without mutating the stored reference object.
-6. Require existing reference results in CI. The current test posts a result when none exists; replace that with a CI failure and a separately authorized local baseline-create/update mode. Do not approve parser regressions by automatically overwriting baselines after a package update.
-7. On IFC import failure, report filename, installed `web-ifc` version, WASM source/version, and original exception. Fail explicitly; do not hide the error and continue operating on a null model.
-8. Wrap each file's viewer/viewport/WASM lifecycle in `try/finally` with awaited disposal. Ensure failed comparisons still release resources before processing another file or ending the test.
-9. Preserve the existing 6,000,000 ms DDT test timeout initially; configure hook/teardown and Jenkins stage timeouts separately. Measure normal and long runs before tightening budgets. Regular tests retain the current 60,000 ms test budget initially.
-10. Make DDT client failures settle their promises predictably. In particular, `getTestDataList` currently returns without rejecting on non-200 responses, and reference-result 404 handling should terminate cleanly. Fix these locally during DDT conversion so failures do not become very long timeouts or blocking dialogs.
-
-Run the normal modelviewer Firefox gate whenever `web-ifc`, WASM assets, IFC loader code, or viewer integration changes. Keep the long dataset available for scheduled/pre-release runs and dependency upgrades; confirm existing Jenkins scheduling before changing it. Local Chrome must support both datasets too.
-
-## Coverage and Jenkins Reporting
-
-- Use `@vitest/coverage-istanbul` for required Firefox coverage. Vitest's V8 provider cannot collect coverage from Firefox.
-- Define source inclusion/exclusion explicitly, preserving intended vendor/generated/worker exclusions. Do not instrument helpers or dependencies as production code, and avoid double instrumentation with webpack's coverage loader.
-- Preserve HTML and LCOV artifacts under an agreed coverage path; add Vitest JUnit results for Jenkins. Use distinct output locations for main, DDT, performance, and browser profiles to avoid overwrites.
-- Compare covered source files and uncovered files as well as totals. Different instrumentation can move percentages; investigate differences before adopting new thresholds. DDT coverage is optional and separate from the regular suite, matching its current reporting intent.
-- If browser recycling/shards are needed, merge raw compatible coverage data before reporting, and aggregate results without lost/duplicated tests. Never concatenate LCOV blindly or let the last batch overwrite earlier results.
-- Archive failed-test screenshots/logs and, where supported and useful, Playwright traces. Record browser/provider versions, WebGL2 availability, stage duration, and memory observations. Do not leak credentials or model-sensitive content into public artifacts.
-- Set finite startup, test, teardown, and overall stage budgets. Fail on browser launch errors, missing datasets/assets, unhandled rejections, missing recordings, and empty test selection; do not mask instability with broad retries.
-
-## Proposed Command Contract
-
-The listed local commands are implemented. The established test command names are aliases to Vitest; no Karma fallback runner remains in this checkout.
-
-| Intended command | Responsibility |
-| --- | --- |
-| `npm test` / `npm run test-firefox-headless` | One-shot regular suite in headless Firefox; default suitable for Jenkins. |
-| `npm run test-firefox-headed` / `npm run test-chrome-headed` | One-shot renderer suite with a visible browser; `test-firefox` / `test-chrome` are aliases. |
-| `npm run test-chrome-headless` | One-shot same suite in headless installed Google Chrome. |
-| `npm run test-firefox-watch` / `npm run test-chrome-watch` | Headed Firefox / Chrome watch mode; dependent servers remain available for the session. |
-| `npm run test-firefox-coverage` / `npm run test-chrome-coverage` | Headless regular suite with Istanbul HTML/LCOV and JUnit artifacts. Coverage inclusion is currently limited to the pilot utility. |
-| `npm run test-vitest-modelviewer-firefox` | Normal dataset, Firefox, no automatic reference writes. |
-| `npm run test-vitest-modelviewer-firefox-long` | Long dataset, Firefox, separately scheduled. |
-| `npm run test-vitest-modelviewer-chrome` / `npm run test-vitest-modelviewer-chrome-long` | Explicit local Chrome DDT variants. |
-| `npm run test-modelviewer-ddt-watch` / `npm run test-modelviewer-ddt-long-watch` | Firefox DDT watch workflows. |
-| `npm run test-vitest-modelviewer-chrome-watch` | Local Chrome DDT watch workflow. |
-| `npm test` | Includes the existing performance tests in the serial Firefox renderer project. |
-| `npm run sync-modelviewer-ddt` / `npm run sync-modelviewer-ddt-long` | Preserve dataset synchronization; not automatic baseline approval. |
-
-Use cross-platform service orchestration with readiness checks and awaited shutdown. `npm-run-all -p -r` alone starts services concurrently but does not establish readiness. Restrict Polly/DDT startup to projects that need them, propagate the test exit code, and clean up children/ports on success, failure, interruption, and watch exit.
-
-## Implementation Phases and Gates
-
-### Phase 0: Establish the Baseline
-
-- Confirm Jenkins OS, display/session, Firefox launch preferences, available memory/GPU, Node version, job commands, artifact consumers, and external DDT server provisioning.
-- Run current Karma Firefox regular, normal DDT, long DDT, and performance checks in their approved environments; record inventory, results, known failures, durations, and coverage. Run representative local Chrome checks too.
-- Freeze comparison inputs: lockfile/package versions, WASM binaries, DDT datasets/references, HAR files, viewport dimensions, and relevant renderer settings. Keep unrelated library upgrades out of the migration.
-
-Exit gate: an agreed inventory and reproducible baseline; known pre-existing failures documented rather than attributed to Vitest.
-
-### Phase 1: Prove the Browser and Asset Path
-
-- Add pinned development packages, the minimal test-only Vite configuration, Firefox/Chrome profiles, fixtures, aliases, WASM/binary serving, and required shader transformation.
-- Convert a small pure utility case, a real viewport/shader case, an IFC loader case, a recorded binary-XHR case, and the modelviewer path using at least one existing IFC/reference.
-- Run the pilot on the actual Jenkins Firefox agent and local installed Chrome. Verify WebGL2 creation, rendering, WASM initialization, binary replay, legacy DDT key lookup, failure propagation, and cleanup through repeated runs.
-- Collect a small Firefox Istanbul coverage report and confirm source mapping and JUnit output.
-
-Exit gate: the pilot passes in both required environments without Chrome fallback on Jenkins, regenerated references/recordings, or sustained memory growth. If Firefox fails, investigate the provider/display path before proceeding.
-
-### Phase 2: Convert Shared Setup and Regular Suites
-
-- Port custom assertions, fixture/global setup, host configuration, Polly lifecycle, and asynchronous teardown.
-- Convert tests in bounded groups: utilities/settings, renderer/viewport, selection/clipping/overlays, markups/interactors/view controls, loaders/workers, and recorded SDK integration. Keep inventory/discovery exclusions current throughout.
-- Run each converted group on Firefox and representative Chrome cases; compare checks and coverage against Karma before the next group.
-- Keep performance assertions and timings intact initially; separate their command/reporting without silently dropping them from required gates.
-
-Exit gate: every baseline test is accounted for as migrated, intentionally retained temporarily, or explicitly approved for removal; full regular Firefox coverage and local Chrome runs pass.
-
-### Phase 3: Complete Modelviewer DDT
-
-- Implement stable reference keys, explicit baseline-update policy, server readiness, deterministic error handling, per-file failure cleanup, and matching package/WASM preparation.
-- Run the full normal dataset in Jenkins Firefox and local Chrome. Run the long dataset serially in Firefox and verify the local Chrome command can execute it.
-- Compare all existing reference fields without generating new references; measure largest-model memory and teardown behavior.
-
-Exit gate: both datasets remain runnable, required reference checks pass, missing references fail CI, and IFC parser/WASM errors produce useful failures rather than hangs.
-
-### Phase 4: Jenkins Cutover and Removal
-
-- Run equivalent Karma and Vitest jobs sequentially during an agreed overlap window to avoid compounding Jenkins memory pressure.
-- Switch public scripts and Jenkins stages only after suite inventory, Firefox stability, DDT parity, performance coverage, artifacts, and local Chrome/watch gates pass.
-- Remove Karma configuration/dependencies, QUnit, framework-only adapters, and obsolete helper scripts once no remaining consumer needs them. Audit webpack loaders individually; production builds may still need them.
-- Update README prerequisites and test instructions, including browser installation, headed Jenkins display requirements, DDT server/dataset preparation, WASM refresh, replay/record policy, and debugging.
-- Re-run the production webpack build as a regression gate after dependency cleanup. Do not commit or change branches as part of implementing this plan unless requested.
-
-Exit gate: no active Karma/QUnit path remains, Jenkins uses Firefox, local Chrome is available, and documentation matches the working commands.
-
-## Acceptance Checklist
-
-- [ ] Jenkins executes real WebGL2 renderer tests in Firefox without launching Chrome.
-- [ ] Local Google Chrome and Firefox one-shot/watch workflows work.
-- [ ] Regular, recorded-network, worker/asset, performance, and modelviewer checks have inventory parity.
-- [ ] Firefox Istanbul coverage produces valid HTML/LCOV; Jenkins receives complete JUnit results.
-- [ ] Normal and long modelviewer datasets exercise real installed `web-ifc` with matching WASM.
-- [ ] Legacy reference/HAR identities are preserved; CI cannot create or update baselines/recordings automatically.
-- [ ] Empty/missing datasets, WASM failures, missing recordings/references, and browser startup failures fail promptly.
-- [ ] Repeated runs and large IFC inputs complete without unacceptable memory growth or leaked service/browser processes.
-- [ ] Production webpack build remains functional after runner dependency removal.
-
-## Rollback and Open Decisions
-
-Karma scripts, dependencies, and configuration have been removed from this checkout. A rollback would require restoring the prior runner files and dependencies from an external revision; no Git operation was performed. Do not re-record HAR data or regenerate DDT references as a rollback step.
-
-Phase 0/1 must resolve the Jenkins display/GPU environment, Playwright Firefox versus system-Firefox provider viability, supported Node/package versions, long-DDT scheduling, and whether bounded browser restarts are necessary. These are validation gates, not reasons to weaken the Firefox requirement.
-
-## References
-
-- Vitest Browser Mode: https://vitest.dev/guide/browser/
-- Playwright provider configuration: https://vitest.dev/config/browser/playwright
-- Vitest test projects: https://vitest.dev/guide/projects
-- Vitest coverage providers: https://vitest.dev/guide/coverage
-- Playwright browser installation and supported builds: https://playwright.dev/docs/browsers
-
-Use documentation matching the selected pinned release when implementing; configuration APIs and browser revisions can change.

package/eslint.config.mjs

@@ -1,43 +0,0 @@
-import globals from "globals";
-import js from "@eslint/js";
-
-export default [
-  js.configs.recommended,
-
-  {
-    ignores: ["recordings/*", "src/compiledWorkers/*"],
-  },
-
-  {
-    files: ["src/**/*", "test/**/*", "modelviewer-data-driven-tests/tests/**/*", "modelviewer-data-driven-tests/simple-ddt-client/**/*"],
-    languageOptions: {
-      ecmaVersion: 2022,
-      sourceType: "module",
-      globals: {
-        ...globals.browser,
-        $: "readonly",
-        jQuery: "readonly",
-        __VITEST_HOST_ENV__: "readonly",
-        __VITEST_DDT_FILE__: "readonly",
-        define: "readonly",
-        require: "readonly",
-        module: "readonly",
-        __dirname: "readonly",
-      },
-    },
-    rules: {
-      semi: ["warn", "always"],
-      "no-useless-assignment": "off",
-      "preserve-caught-error": "off",
-    },
-  },
-
-  {
-    files: ["vitest.config.mjs"],
-    languageOptions: {
-      globals: {
-        ...globals.node,
-      },
-    },
-  },
-];

package/openAllFiles.ps1

@@ -1,17 +0,0 @@
-# security warning about running scripts, you need to change the execution policy
-# to allow running scripts. You can do this by running the following command in PowerShell:
-# Set-ExecutionPolicy RemoteSigned -Scope CurrentUser
-
-# Define the path to the projects directory
-$projectsPath = ".\src"
-
-# Get all .ts, .less, and .html files, excluding node_modules and dist directories
-$files = Get-ChildItem -Path $projectsPath -Recurse -Include *.js, *.less, *.html | Where-Object {
-    $_.FullName -notmatch '\\node_modules\\' -and $_.FullName -notmatch '\\dist\\'
-}
-
-# Open each file in Visual Studio Code
-$files | ForEach-Object { code $_.FullName }
-
-# Output the count of opened files
-$files.Count

package/stats.json

@@ -1,4 +0,0 @@
-[diffend] Oversized file quarantined before diffing.
-name: package/stats.json
-size: 38561098 bytes
-sha256: af44beeaea980931278104708261b44bdbe0e059592d76831925bbd1e3046127