Step 4 of docs/calendar-auth-migration.md, and the close of
DEFERRED_SECURITY.md item 1: no calendar route reads sessionId/sessionKey from
the query string any more, so a live credential no longer travels through
access logs, browser history and Referer headers.
The four write routes sit behind requireAppAccess('calendar'), which also
narrows who may edit from "any activated @nachklang.art account" to an
explicit per-user permission. They answer 401 signed out and 403 without the
permission, where they previously answered 403 for both.
The three read routes cannot use the middleware: one URL serves an anonymous
visitor, an iCal subscription holding a shared password, and a signed-in
editor who should see drafts. They resolve the session optionally instead, and
a signed-in user without the calendar permission is treated as anonymous
rather than refused - so they keep the public calendar access anyone has.
That public calendar staying anonymous is load-bearing: nachklang.art reads it
to show the next upcoming event. It is now pinned at both the password-table
and the route level, and so is the rule that a shared password can never be
used to write.
credentials.service.ts loses its session half and becomes the password table
it always wanted to be. The shared passwords survive only for iCal clients,
which cannot send a cookie.
Writes record the author as an admin user id and no longer have a legacy int
to write, which is what migration 003 makes room for.
/calendar/users/* is left in place: nothing calls it and a session it mints
opens nothing, but they are still live password-accepting endpoints, so
removing them belongs with the rest of the legacy path in step 5.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Steps 1 and 3 of docs/calendar-auth-migration.md. The calendar is the last
module still authenticating against its own users/sessions tables; this is
the groundwork that lets step 4 swap it for the shared admin identity.
An event now records its creator twice: created_by_id, the legacy INT into
the calendar database's own users table, and created_by_user_id, the admin
module's VARCHAR(36) id. The two live in different databases, so there is no
foreign key and no join - a cross-schema reference would tie the schemas'
lifecycles together, and the name is instead resolved through one lookup per
result set against the admin database.
The creator is only ever rendered as a name; nothing authorises on it. That
is what makes the planned account backfill unnecessary - dropped by decision -
and what makes the read degrade rather than fail: an admin id that no longer
resolves falls back, and an unreachable admin database costs a name rather
than the response. The public calendar is read anonymously by nachklang.art
and has never depended on the admin database being up.
Since there is no backfill, step 5 dropping the legacy users table would have
erased the authorship of every pre-cutover event. Migration 002 brings that
part of step 5 forward and snapshots the names onto the events themselves, so
the data is safe well before the table holding it goes away.
The same SELECT and row mapper existed in four copies; collapsed to one of
each first, so the dual read is written once rather than four times.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A production password containing '#' was truncated after the dotenv 8 -> 16
bump (unquoted '#' now starts a comment), causing MariaDB access denied.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Prep PR for the admin auth module (docs/plan-admin-auth.md step 1).
better-auth 1.7 ships ESM only, so the API moves off CommonJS:
- "type": "module", module nodenext, target ES2024, .js suffixes on all
relative imports, require('mariadb'|'cors') replaced by imports, and
export= packages (winston, app-root-path, bcrypt) consumed via default
imports. The logger now uses appRoot.path explicitly.
- TypeScript 5.9, @types/node 26, tslint removed. Node 26 pinned via
engines and .nvmrc (Plesk runs 26).
- Jest 28 + ts-jest replaced by vitest 5. Eight test files depend on
hoisted module mocks with static imports and resetModules + require,
which Jest's ESM mode does not support; vitest keeps them nearly
verbatim. Coverage via @vitest/coverage-v8 (lcov), Sonar generic report
via vitest-sonar-reporter, so sonar-project.properties is unchanged.
vitest.config.ts sets FEEDBACK_IP_SALT so the suite passes without a
local .env.
- dotenv 8 -> 16 and axios 0.24 -> 1.x: their old typings are not
resolvable under nodenext.
- autoCommit: false dropped from the pool configs; it is not a mariadb
connector option and was silently ignored.
tsc clean, 96/96 tests green, compiled app boots and serves /, /docs and
CORS under Node ESM.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Paddy
