23f32872a583bd008b469fca20f8d63e7522c128
4 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 wmj
|
33a821e61d |
refactor: restructure navigation into 5 grouped hubs, off the 22-item flat strip
Implements the researched IA fix: mobile nav UX consensus caps primary destinations at 4-5 (uxpin.com, fintech/banking 2026 UX research), and this app had grown to 22 tabs in one horizontally-scrolling row — exactly the anti-pattern that research flags for choice paralysis and slower task completion. New structure — 5 bottom-nav hubs, grouped by what the user is actually trying to do, not build order: - Home: Coverage (unchanged, still the landing screen) - Estate: Assets, Faraid, Insurance, Wassiyah, Hibah, Family Waqf, Nominate, Claims (H2) - Giving: Zakat, Sadaqah, Khairat - Family: Tree, Trigger, Mutawalli, Manage (was 'Family', renamed to avoid colliding with the hub's own label), Neighbourhood - Daily: Prayer Times, Qibla, Quran, Locate Each hub reveals its own sub-nav one tap in, instead of every tab competing for space in a single row. Settings moved out of the tab strip entirely into a header gear icon, matching the banking-app pattern of keeping settings out of primary thumb-reach real estate. The bottom nav is now fixed (thumb-zone), sub-nav keeps the old sticky-top position. Cross-component navigation (nav.js requestedTab, used by the Family Tree's 'give sadaqah in memory' link) now resolves a tab label to its owning (hub, sub-tab) pair instead of a flat index — verified live. This touches every E2E suite: a single click on a tab's old selector no longer reaches it (hub, then sub-tab). Added a shared gotoTab(page, label) helper to e2e-auth-helper.cjs encapsulating the two-step navigation, and migrated all ~22 affected test files off direct nav-button selectors — mechanical substitution followed by manual fixes for local clickTab wrappers, template-literal selectors, and active-state assertions that needed to target the new .hub-tab/.subnav structure specifically. Full regression after migration: every suite passes (one isolated Family Tree flake confirmed clean on rerun, unrelated to navigation). |
||
|
wmj
|
5adbad6d53 |
Real SMTP heir notifications — no third-party signup needed
Per explicit direction to use existing VPS/Bitwarden/Gitea infrastructure instead of waiting on a Resend signup. Bitwarden's MCP unlock/list both hung (server-side issue, confirmed via direct CLI retry too — not something to keep retrying), so this used the VPS and Gitea directly. Found real, working infrastructure already in place: a documented but never-deployed falah-ibaas email connector (SMTP wrapper) at /opt/falah-ibaas/connectors/email, backed by a local Postfix relay that Ghost already uses successfully in production on the same VPS (mail__options__host=172.17.0.1:25, no auth needed internally). Built nf-mail-relay: a small HTTP wrapper (Python stdlib, no deps) around that connector, deployed as a Docker Swarm service on the existing Traefik network at https://nfmailrelay.falahos.my, bind-mounting the connector code read-only so it stays in sync with any future updates to it. Shared-secret bearer auth (X-Relay-Secret) — verified a wrong secret gets rejected with 401. notify-heirs Edge Function rewired to call this relay instead of Resend. Two real bugs found and fixed via actual testing, not code review: - The function only took memberId, but a person can belong to multiple families — .maybeSingle() against multiple trigger rows failed closed (correctly, but silently, as "not triggered"). Function and both call sites (db.js notifyHeirs, MutawalliDashboard) now require and pass familyId too, matching the same composite-key fix already applied to the trigger tables themselves. - No CORS/OPTIONS handling: a browser's preflight OPTIONS request has no body, and calling req.json() on it crashed the function before any headers were sent — surfaced in the browser as a generic "Failed to send a request" with no detail. Added an OPTIONS short-circuit and CORS headers on every response path. Verified with a real send to a live inbox through the full chain (browser -> Edge Function -> VPS relay -> Postfix -> SMTP), not just a connectivity check. Also fixed a stale e2e-trust.cjs assertion using the same instant-isVisible()-after-fixed-wait pattern already fixed elsewhere in this session — real app behavior was correct, only the test's timing assumption was wrong. e2e-per-member.cjs's heir-notification check now asserts an actual "Sent" result via the real relay instead of accepting either Sent or a not-configured failure. Full sweep: e2e-uat 32/32 (stable across 3 runs, one earlier run's failure was a one-off network blip under heavy parallel test load), e2e-fastpath 16/16, e2e-trust 12/12 (stable across 3 runs), e2e-business 10/10, e2e-digital-vehicle 10/10, e2e-property 9/9, e2e-other 4/4, e2e-info 31/31, e2e-per-member 11/11 — 165/165 total. |
||
|
wmj
|
b9a98bd97a |
Restore all 8 pre-auth E2E suites with a sign-in prelude; fix bugs they found
New e2e-auth-helper.cjs: shared signInFreshFamily() prelude — signs in as the confirmed test owner account and creates a uniquely-named family per run, so accumulated data from a previous run's assets/hibah/etc. (now persisted in Supabase, not wiped with the browser context like localStorage was) can never bleed into another run's percentage/coverage assertions. All 8 suites (e2e-uat, e2e-fastpath, e2e-trust, e2e-business, e2e-digital-vehicle, e2e-property, e2e-other, e2e-info) now call it in place of the old anonymous page.goto(BASE). Restoring them surfaced two real product bugs, not just test staleness: 1. WassiyahGenerator.svelte was never migrated to Supabase in the earlier backend work — it still called the old local storage.js load()/save() for assets, bequests, and witnesses, so the one-third meter silently read an empty local cache and always showed 0. Migrated to family-scoped Supabase tables (new nf_wassiyah_bequests, nf_wassiyah_settings, with member-scoped RLS) matching the pattern used for Hibah/Nominations/etc. 2. storage.js's exportAll() did an unguarded JSON.parse on every "nf."-prefixed localStorage key, but family.js stores activeFamilyId as a raw string (not JSON-encoded) — one malformed parse threw and silently aborted the whole export before the file download fired. Made exportAll defensive: falls back to the raw string on a parse failure instead of throwing. The remaining test failures were async-timing gaps inherent to the move from synchronous localStorage reads to async Supabase fetches: several assertions checked <select> option counts or newly-created rows immediately after a fixed short wait, before the async load/refresh had actually landed. Fixed by replacing blind isVisible()/fixed-timeout checks with proper waitFor()/polling in the test helpers (selectByText, corpus-select population, row-creation checks) — not a product bug, but worth fixing since the old timing assumptions no longer hold now that data is live and shared instead of instant and local. Results: e2e-uat 32/32, e2e-fastpath 16/16, e2e-trust 12/12, e2e-business 10/10, e2e-digital-vehicle 10/10, e2e-property 9/9, e2e-other 4/4, e2e-info 31/31 — 124/124. Re-verified e2e-family-agent (12/12) and e2e-smoke-authed (24/24) still pass after the WassiyahGenerator migration. 160/160 total across all ten suites. |
||
|
wmj
|
ff47dc9ed2 |
Cover land parcels via proper trust setup, not generic nomination fields
Nomination Registry's "trust" channel type previously reused the same nominee/institution fields as EPF/Takaful/bank nominations — wrong shape for a trust, which needs a trustee (and successor) holding title, not a beneficiary named on a policy. Coverage math already counted any nomination-type entry as fast-path regardless of shape, so land parcels were silently "covered" without the actual structure a trust requires. Now: selecting "trust" swaps in trustee/successor-trustee/beneficiaries fields, and adds a dedicated trust-setup drafting-aid export (same treatment as the Waqf deed export) explaining what still needs a real trust company/lawyer to execute. e2e-trust.cjs: new suite verifying the full path for a land parcel — exposed -> trust suggestion -> trustee fields -> deed export -> coverage flips to 100% -> death trigger generates its execution packet. 12/12 passing. Re-verified e2e-uat.cjs (32/32) and e2e-fastpath.cjs (16/16) with no regressions. |