frontend module conversion)
ui-state was the smallest, most isolated module: 35 lines, zero legacy-
global dependencies, single concern (localStorage helpers under the
'ped_ui/' namespace). Ideal first migration to validate the strangler-
fig pattern end-to-end.
Changes:
- New: client/ui-state.ts — exports get/set/del + UIState bundle.
Keeps (window as any).UIState = UIState assignment so the 10+ legacy
consumers (sub-nav.js, peGuide.js, app.js, etc.) keep working without
modification. As each consumer is migrated, it will replace the
window.UIState read with an explicit import.
- client/main.ts: import './ui-state' so it gets bundled into main.js.
- scripts/build-client.mjs: explicit ENTRIES list (just 'main.ts' for
now) — was scanning every top-level .ts as an entry, which produced
duplicate ui-state.js bundle.
- public/index.html: <script defer src="/js/ui-state.js"> removed; the
TS bundle (/dist/main.js) now provides window.UIState.
- public/js/ui-state.js: deleted (now lives in client/).
Verification:
- npm run typecheck:client → 0 errors
- npm run build:client → 349-byte main.js bundle (gzipped will be ~200B)
- 46/46 unit tests pass
- The 10 cross-file consumers of UIState.get/set/del still work because
the window assignment lives in the bundle.
Pattern documented for next migrations: each new client/X.ts is
imported as a side-effect from main.ts. The window-bridge stays until
every consumer also moves to ES imports.
ce31a1e343