Testing

Testing is treated as a first-class part of the codebase, not an afterthought: unit tests cover the core logic in a node environment, and E2E suites drive a production preview in a headless browser.

Unit tests (Vitest)

npm test           # vitest run
npm run test:watch

Environment: node. One hundred twenty-three test files, 2233 tests:

Suite	Covers
fileFormat	magic-number + extension detection, format matching
cspkg	zip parsing, manifest validation, entry execution (trusted)
sandbox	RPC encode/decode, legacy fallback, full worker round trip via a fake Worker
i18n	locale switching, fallbacks, new sandbox keys
appStore	banners, notifications, perf warnings
wasm	loader retry policy (3×, 1 s) with mocked module
gpuService	compute-service device gating / CPU fallback
gpuCompute	WGSL template generation, buffer packing, CPU integrator, particles fallback
builtinPlugins	contour grid normalization, scatter parsing
sciencePlugins	3-D N-body kernel/pack/momentum conservation + plugin CPU fallback; protein force-directed layout, components, no-data guards
geoPhysicsPlugins	fluid WGSL kernels + fluidStepCPU + mask parsing; wave kernel/stepping/data parsing; double-pendulum physics + initial conditions; GeoJSON parsing
dataPlugins	error-band rows, treemap hierarchy, numeric columns, QQ probit
wgsl	histogram / heatmap / point-cloud GPU kernels: WGSL output, param packing, output sizing, CPU fallback parity
blocks/*	block system end-to-end — registry, compiler (validation/topology/types), executor (incremental cache + invalidation), DataTable ops, geometry, statistics, catalog executors, viz.* render bridge
editor/*	shared IR (validate/round-trip), Blockly JSON ⇄ IR, interpreter, JS/Python codegen, three-mode sync-threeway, block i18n (BKY_*), editor store, Studio API, code samples
stats/*, io/*, plot/*, repro/*	statistical test helpers, NetCDF I/O, plotting helpers, reproducibility
pluginRuntime, pluginBugfixes	plugin lifecycle, marketplace gating, regression tests for fixed plugin defects
structure-sim	truss solver: member forces, utilization, overload collapse

Notable techniques:

• Module-level caches (e.g. the WASM loader) are reset per test with vi.resetModules() + vi.doMock + a fresh dynamic import.
• Worker RPC is tested end-to-end with a FakeWorker that drives the real createPluginWorkerRuntime — no browser required.
• Strict-mode new Function cannot combine a "use strict" directive with default parameter values — the legacy sandbox uses plain parameters and passes undefined explicitly.

E2E suites (Playwright-core)

npm run test:e2e

Each suite boots vite preview on its own port, drives headless Edge (C:/Program Files (x86)/Microsoft/Edge/Application/msedge.exe by default), and asserts real pixels + zero console errors:

Suite	Port	Covers
smoke-test	4173	boot, auto-loaded plugins, reactive params, project restore
verify-ui	4173	layout metrics, theming, canvas visibility
verify-fixes	4177	all example plugins render their sample data correctly
verify-3d	4199	3D point cloud in the host Three.js scene
verify-plugins	4198	3D↔2D surface visibility, contour, scatter, tornado sample
verify-webgpu	4289	real WebGPU path in headless Edge (SwiftShader): numeric harness (GPU vs CPU integrator within ~2e-6) + Particles app integration asserting a wasm-engine toast
verify-block-mode	4173	block editor: mode switch, compile, run, block → code sync
verify-code-mode	4175	Monaco + Pyodide: run a Python program, console, variable panel, plot canvas
verify-ai-samples	4173	AI Training: load all 4 samples (linear / non-linear / logistic / MNIST)
verify-ai-training	4173	AI Trainer: activate, TF.js train, loss curve, model-switch reset, decision boundary, MNIST CNN grid

If Edge is not at the default path, update the EDGE constant in each script. New plugin/feature work should ship with an E2E check — screenshots are written under the temp opencode/shots directory for visual review.

Combined checks

npm run verify       # typecheck + unit tests
npm run build        # wasm + typecheck + production build