TNCSC_Odoo/HANDOFF.md

# Handoff / Resume Notes

Read this first if you're picking this project back up — whether that's
Claude Code in a future session or a human developer. The phase-by-phase
build plan is in `CommunityOS_Implementation_Plan_for_Claude_Code.md`; this
file is the "where things actually stand" companion to it.

## Status at a glance

| Phase | What it is | Status |
|---|---|---|
| 0 | Repo scaffold, Docker dev stack, CI + brand-leak guardrails | ✅ Done, committed |
| 1 | `community_membership` | ✅ Done, committed |
| 2 | `event_qr_ticketing` | ✅ Done, committed |
| 3 | `community_school` | ✅ Done, committed |
| 4 | `community_classifieds` + `community_benefits` | ✅ Done, committed |
| 5 | `community_interac` (Interac e-Transfer payment provider) | ✅ Done, committed |
| 6 | `community_theme_base` + `community_portal` | ✅ Done, committed |
| 7 | `tncsc_deployment` (client config layer) | ✅ Done, committed |
| 8 | TNCSC data migration scripts | ✅ Done, committed (`9a24691`) |
| 9 | Packaging for resale + production go-live | 🟡 9-A in progress (see below); 9-B not started |

Every phase 0–7 commit was verified against a **live** Odoo 19 + Postgres 16
container before being committed — not just written and assumed to work.
Run `git log --oneline` for the full commit list; each commit message
documents what was built and how it was verified, including several real
Odoo 19 API changes that don't match older docs/tutorials (see "Gotchas"
below for the index).

**Git remote is configured**: `origin` -> `https://git.metatroncube.in/admin/TNCSC_Odoo.git`.

## Resuming the dev environment

```bash
cd deploy
docker compose up -d
```

- Odoo: http://localhost:8069, database `communityos_dev`, login `admin` /
  password `admin`.
- This dev database has been used for **all** testing across every phase —
  it has real posted accounting entries in it (from Interac transaction
  confirmations, membership invoices, etc.). That's *why* Phase 7's
  `tncsc_deployment` module can't be installed on it directly (Odoo
  correctly refuses to change a company's currency once journal entries
  exist) — it was instead verified on throwaway fresh databases
  (`tncsc_freshN`, all dropped after verification). A real TNCSC production
  database should be a genuinely fresh install, not this dev one.
- **After any module code change**, the *running* `deploy-odoo-1` container
  has a stale in-memory registry until you either upgrade the module (`-u`)
  and restart the container, or just restart it:
  ```bash
  docker compose restart odoo
  ```
  This bit repeatedly during development — don't skip it when testing.

## Phase 8 — done, committed (`9a24691`)

All three migration scripts exist, are committed, and have been verified
against a **live** Odoo instance (see "How each was tested" below).

- `scripts/_migration_common.py` — shared JSON-RPC helpers (auth, execute_kw,
  CSV normalization, a `MigrationReport` class) for the migration scripts.
- `scripts/migrate_members.py` — idempotent upsert by email, dry-run flag,
  tier lookup by code, report CSV.
- `scripts/migrate_students.py` — idempotent upsert of parent partner (by
  email, reusing a partner created by `migrate_members.py` if one exists)
  + child partner + `community.school.student`, matched by (parent, student
  name) since students don't have their own email in the source data.
  Enrolls into the matching class in the current open term (matched by
  `community.school.level` code) when one exists; otherwise still
  creates/updates the student record and reports why no enrollment
  happened.
- `scripts/migrate_opening_balances.py` — posts one `account.move` in the
  Miscellaneous Operations journal, dated the last fiscal year-end
  (computed from `res.company.fiscalyear_last_day/month`, or `--date`
  override). Refuses to post anything if the CSV doesn't balance
  (`sum(debit) != sum(credit)`) or references an unknown account code —
  checked before any Odoo write happens. Idempotent via a fixed `ref`
  ("Opening Balances Import"): re-running finds the existing posted entry
  and does nothing.

### How each was tested

Followed the same pattern as `scripts/migrate_classroom.py` (Phase 3):
dry-run first, then a real run, then a second real run to prove
idempotency — against a **fresh** database, not `communityos_dev`. See
"Why a fresh DB, not communityos_dev" below for why.

1. `docker compose up -d`, then created a throwaway DB with both product
   modules and the client layer: `odoo -d tncsc_migration_test -i
   community_membership,tncsc_deployment --stop-after-init` (installing
   `tncsc_deployment` alone pulls in `community_school` and everything else
   as dependencies). Restarted the container afterward.
2. `migrate_members.py`: synthetic CSV at
   `data/raw/members_export_sample.csv` (gitignored, left in place as a
   fixture) with 4 valid rows + 2 deliberately-bad rows (missing email,
   unknown tier code). Dry-run matched real-run output; second real run
   reported `updated` not `created` for all 4; confirmed via RPC that
   exactly one `res.partner` per email existed after both runs and
   `membership_state` was `active`.
3. `migrate_students.py`: needed a `community.school.term` (state=open),
   `community.school.level` (code=`BEG`), and `community.school.class`
   seeded first (the fresh DB has none) — also had to grant the `admin`
   user the School Coordinator group, since a fresh install doesn't put
   admin in it. Synthetic CSV at `data/raw/students_export_sample.csv`
   with 3 valid rows (2 matching the open term's class, 1 with no
   `level_code`) + 1 deliberately-bad row (missing parent). Confirmed via
   RPC after 2 runs: exactly 3 `community.school.student` records, exactly
   2 `community.school.enrollment` records (no duplicates), parent
   partners correctly reused from step 2 instead of duplicated.
4. `migrate_opening_balances.py`: synthetic CSV at
   `data/raw/opening_balances_sample.csv` using real account codes from
   the fresh DB's TNCSC chart of accounts (111100 Cash, 112110 Trade AR,
   322000 Retained Earnings as the balancing line). Verified the
   out-of-balance guard refuses to post (tested with a deliberately
   mismatched debit/credit CSV). Real run posted one balanced, posted
   `account.move`; second run detected the existing entry and did nothing.
   Confirmed via RPC: exactly one move, three correctly-valued lines, AR
   line correctly linked to the migrated Ravi Kumar partner.

Real TNCSC member/student/trial-balance data doesn't exist yet — all three
scripts were intentionally designed against a *documented, assumed* CSV
schema (see each script's docstring) since the real export formats aren't
known yet. Confirm the assumed column names match the real exports before
relying on them, or adjust the scripts to match.

### Why a fresh DB, not `communityos_dev`

`communityos_dev` doesn't have `tncsc_deployment` installed (see the top of
this file — it can't be, since it already has posted journal entries and
Odoo refuses to change a company's currency once those exist). Testing the
Phase 8 scripts there would exercise generic `community_membership`/
`community_school` behavior, not the actual TNCSC-configured target
(TNCSC's real chart of accounts, company currency, groups). Since Phase 8's
entire purpose is rehearsing the real TNCSC data load, a fresh DB with
`tncsc_deployment` installed is the more representative target — and it
avoids adding more synthetic test data to the one long-lived dev DB used
across every other phase. `tncsc_migration_test` was **not** dropped after
this session (unlike the Phase 7 `tncsc_freshN` throwaways) — it's left in
place in case the next session wants to keep testing against it; drop it
before it's mistaken for anything resembling real data.

### To pick this back up

Either wait for real TNCSC export samples to validate the assumed CSV
schemas against, or move on to Phase 9.

## Phase 9 — 9-A in progress, 9-B not started

Two sub-parts with very different risk profiles:
- **9-A (productize)**: safe to do autonomously. Done so far — every
  `community_*` module has `static/description/index.html`, a real
  `README.rst` (Usage sections filled in past the Phase-0 placeholder),
  and a `CHANGELOG.rst`; manifests pass `scripts/check_manifests.py`.
  Still open:
  - Every product manifest's `images` key points at
    `static/description/banner.png`, but no module actually has that
    file — the Apps-list/App-Store banner image doesn't exist yet. Needs
    a real designed asset per module, not a placeholder.
  - No `v1.0.0` git tag yet, and no release report — holding off on both
    until the banner-image gap above is resolved and the full
    `--test-enable` suite has actually been run against live Odoo (this
    session had no Docker available, so only the static checks —
    brand-leak grep, manifest completeness — were re-verified, not the
    live module install/tests).
  - No role guides (admin/treasurer/teacher/parent) written yet.
- **9-B (go-live)**: not started. `deploy/odoo.conf` is still dev config
  (`list_db = True`, `proxy_mode = False`, no `workers`) and
  `deploy/nginx-communityos.conf` is an explicit "development skeleton" —
  SSL, DNS cutover, firewall rules, and backups on a **real production
  server** all remain. This needs a human in the loop with actual server
  access/credentials — don't let an agent run this unattended.

## Gotchas discovered this build (Odoo 19 API drift + environment quirks)

This Odoo 19 build (`19.0-20260817`, essentially a nightly) diverges from
older tutorials/docs in several places that cost real debugging time. Each
is documented in detail in the commit message where it was found — this is
just the index so you know to search `git log` for it instead of
re-discovering it:

- `res.groups.category_id` → replaced by `privilege_id` →
  `res.groups.privilege.category_id` (Session 1-A commit).
- Settings pages use a new `<app>/<block>/<setting>` structure inherited
  from `base.res_config_settings_view_form`, not the old raw-div layout
  (Session 1-A commit).
- `ir.cron` dropped `numbercall` entirely (Session 1-B commit).
- `event.event` has no `state` field anymore (replaced by a
  `stage_id`/`event.stage` kanban system) (Phase 2 commit).
- Route `type='json'` should be `type='jsonrpc'` (`'json'` still works but
  is deprecated) (Phase 2 commit).
- Search-view `<group>` elements for "Group By" sections no longer accept a
  `string` attribute, only `name` (Phase 3 / Phase 4 commits).
- `res.groups.users` → renamed to `user_ids` (Phase 4 commit).
- `res.config.settings` fields backed by `config_parameter=` only support
  boolean/integer/float/char/selection/many2one/datetime — **not**
  `Binary`. Using it on a Binary field breaks *every* Settings tab, not
  just the one you're adding, since they share one model (`fix:` commit
  after Phase 6).
- Anonymous visitors hitting an `auth='user'` + `website=True` route in
  this specific build crash with a raw 500 instead of a clean redirect to
  `/web/login` (`Request._serve_db`'s `finally: self.env = None` runs
  before the website error handler tries to build the redirect). Worked
  around by switching those routes to `auth='public'` with a manual
  `if request.env.user._is_public(): return request.redirect(...)` check
  (`fix:` commit after Phase 6 — this is the fix to reach for if you add
  *new* member-only pages later, don't reintroduce `auth='user'` on a
  `website=True` route without it).
- Setting `res.company.country_id` on a brand-new company auto-schedules
  this build's own chart-template installer via a precommit hook, which
  races an explicit `try_loading()` call made in the same install
  transaction and silently reverts it afterwards. `ir.cron` code execution
  is also sandboxed and forbids direct attribute assignment (`STORE_ATTR`)
  — use `.write(...)`. Both fully explained in the Phase 7 commit message,
  fix pattern is in `addons/tncsc_deployment/models/res_company.py`.

## Handing off to a team

1. Repo is already pushed to `origin` (see above) — confirm the team has
   access to `git.metatroncube.in`.
2. Point them at `CommunityOS_Implementation_Plan_for_Claude_Code.md` (the
   plan) and this file (current status).
3. They'll need Docker Desktop (or equivalent) to run `deploy/docker-compose.yml`
   locally — no other local dependencies.
4. If they're using Claude Code to continue: just point it at this repo and
   this file. It doesn't need conversation history — everything load-bearing
   is either in a commit message or in this document.