The API container used server garbage collection. Server GC makes one heap per
CPU, which is 22 heaps on this host. The two frontend containers sized the Node
heap from host RAM (16 GB), because compose sets no cgroup limit.
Set DOTNET_gcServer=0 and NODE_OPTIONS=--max-old-space-size=2048. This bounds
the memory that `docker compose up` needs.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Restructures into apps/ssp + apps/behandelportal (two Angular projects)
plus libs/shared + libs/beheer (cross-app libraries), replacing WP-61's
separate sibling repo. That split had already produced real drift: a
hand-vendored copy of the backend's OpenAPI doc, a shared/ui+layout tree
forked and silently diverging (7 files), and beheer + the styles.scss
token bridge duplicated byte-for-byte across both repos.
- git mv the SSP's src/app/* into apps/ssp/; fold shared/, beheer/,
environments/, the Storybook docs/*.mdx, and styles.scss into
libs/shared + libs/beheer (all confirmed identical between the two
repos before merging). auth stays deliberately duplicated per
ADR-0002 (actor-specific, expected to diverge) - amended there.
- One generated API client (libs/shared), no more vendored swagger.json.
- .dependency-cruiser split into a base factory + one config per app,
and Storybook into .storybook-ssp/.storybook-behandelportal - both
forced by the @auth/* alias resolving to different directories per app.
- SiteHeaderComponent/ShellComponent gained HEADER_NAV_ITEMS/
HEADER_ADMIN_LINKS/DEBUG_PANEL injection tokens so each app supplies
its own nav/admin-links/dev-panel instead of one being hardcoded.
- CLAUDE.md, ARCHITECTURE.md, dependencies.md, and ADR-0002 updated;
WP-67 backlog entry documents the full decision trail.
npm run ci green (lint, dep:check x2, 360 tests across ssp/
behandelportal/shared/beheer, both localized builds, backend tests,
snippet + api-client drift); both dev servers, both Storybook
instances, and docker compose verified working.
The old sibling repo (/home/eho/repos/behandelportal) is left
untouched, not deleted.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Serve the source locale (nl) at / instead of /nl/: angular.json sourceLocale is now
{ code: 'nl', subPath: '' } → nl output at browser/ root (base href /), en stays /en/.
Rework localeLinks (nl → bare path, en → /en/…) + spec, and serve-i18n.mjs (nl assets at
root, en under /en/, / serves the nl index). And build the docker demo (+ serve:i18n) with
`--configuration development --localize` so isDevMode() stays true and the dev `⚙ state`
panel + role/scenario switchers render in the localized compose demo (they were correctly
gated off in the previous production build). Verified: nl base href /, en /en/, ngDevMode present.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Fix the hardcoded LOCALE_ID: 'nl' — provide it from $localize.locale (the build-time
locale, 'nl'/'en', undefined→'nl' in dev) and register both nl+en locale data, so the en
bundle formats dates/numbers correctly. Make `docker compose up` serve the LOCALIZED build:
the web service now runs `ng build --localize` then serve-i18n.mjs, which gained a PORT env
+ an /api reverse-proxy (API_PROXY_TARGET → the api container) so both /nl/ and /en/ are
served with the language switcher working end-to-end. Drop the now-unused proxy.conf.docker.json
(serve-i18n proxies /api itself); update ARCHITECTURE. `npm start` stays the nl-only HMR loop.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
CI (Gitea Actions, .github/workflows/ci.yml):
- CodeQL off the PR critical path: runs on push-to-main + a weekly cron only
(`if: github.event_name != 'pull_request'`, `schedule: Mondays 03:00 UTC`).
The 2-language 20-min matrix was the slowest thing on every PR; code is still
scanned on main + weekly.
- Cache Playwright browsers (~/.cache/ms-playwright) in the storybook-a11y + e2e
jobs — skips the chromium download on a hit; `install --with-deps` then only does
the fast apt deps check.
- Cache NuGet (~/.nuget/packages, keyed on **/*.csproj — no packages.lock.json) in
the backend / e2e / api-client-drift / codeql-csharp jobs.
- `npm ci --prefer-offline --no-audit --no-fund` in the 4 npm jobs.
Demo (docker-compose.yml, local only — NOT used by CI): web image node:24 →
node:24-slim (~1.1GB → 232MB verified). The container only runs `npm ci && ng serve`
and the native deps ship prebuilt glibc binaries, so slim needs no toolchain —
verified: npm ci clean, ng serve boots, app returns 200.
Note: CI timing/behaviour can only be confirmed on the Gitea runner (not observable
locally). Every change here is independently revertable.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Three fixes so Gitea CI is green and stays that way:
- **ng build --localize**: `messages.en.xlf` (hand-maintained, matched by id) had drifted
~93 units behind the source since WP-24 — `beheer.*`, `brief.canvas.*`, `brief.diff.*`,
`orgTemplate.*`, `brief.besluit.*`, etc. With `i18nMissingTranslation: "error"` that fails
the frontend job. Added English `<target>`s for every missing unit (interpolation `<x/>`
tags preserved); both nl + en bundles build clean.
- **docker npm noise**: the web container's `npm ci` now runs `--no-fund --loglevel=error`,
so `docker compose up` boots without npm 11's deprecation / allow-scripts / funding
warnings (verified in the running container: 0 warning lines). The underlying Angular
webpack-builder deprecation is upstream; migrating to @angular/build is out of scope.
- **local CI check**: `scripts/ci-local.sh` + `npm run ci` mirror the CI jobs (lint,
format:check, check:tokens, test, build --localize, audit, backend format+test,
api-client drift; `--full` adds storybook-a11y) so a red build is caught before pushing.
Opt-in pre-push hook at scripts/githooks/pre-push. Documented in CLAUDE.md.
Verified: `npm run ci` passes end-to-end; storybook-a11y 169/169 (the local parallel-worker
timeouts were the docker stack starving chromium, not a regression).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds LetterHtml.Render, a pure composer mirroring the FE letter canvas'
class vocabulary, behind two ExcludeFromDescription()'d endpoints
(GET /brief/preview, GET /admin/org-template/{subOrgId}/preview).
Auto-resolvable placeholders pull from seed/case data; unresolved
manual ones render as "[NOG IN TE VULLEN: label]". A sent brief
archives its composed HTML (BriefEntity.ArchivedHtml) so a later
org-template republish never changes it. FE gets a hand-written fetch
adapter (text/html, not JSON) and a "Voorbeeld" button that opens the
preview in a new tab.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Applications, documents (+ audit log) and the brief move off static in-memory
Dictionaries onto a real SQLite file via EF Core, so demo data survives a
process restart or `docker compose restart api` for the first time. The three
stores (ApplicationStore/DocumentStore/BriefStore) keep their exact public
signatures and static-class shape — no DI, no async ripple into Program.cs's
minimal-API handlers — each method just opens a short-lived AppDbContext via
Db.Create() under the same lock it already had. Opaque nested shapes (a
wizard's draft snapshot, a brief's sections/placeholders/status) are stored as
JSON text columns rather than redesigned into relational tables, matching the
existing "don't interpret it" posture.
Found two things the WP's own text got wrong, corrected in
docs/backlog/WP-22-durable-persistence.md's Deviations section: SeedData never
seeded these three stores (only the read-only BRP/DUO-mimicking GETs, which
stay in-memory) so there's no seed step; and no new docker-compose volume is
needed since the existing bind mount already covers the SQLite file — verified
against this environment's real podman-backed compose stack, not just by
reading the file.
Also: pinned SQLitePCLRaw.bundle_e_sqlite3 to 3.0.3 (EF Core Sqlite's own
transitive default bundles a pre-3.50.2 SQLite with a known high-severity
memory-corruption advisory); found and fixed a real xUnit test race where
concurrent test-class hosts stomped a shared static connection-string field,
fixed by disabling cross-class test parallelization rather than adding DI the
stores don't otherwise need.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
One-time prettier --write so the new format:check CI gate starts green.
.prettierignore excludes generated (api-client.ts, documentation.json),
vendored (public/cibg-huisstijl), and backend (dotnet format owns it).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Move the authoritative business rules off the frontend into a real backend,
realising the BFF-lite + decision-DTO design (ADR-0001) that until now lived
only in static mock JSON.
Backend (backend/):
- ASP.NET Core (.NET 10) minimal API, contract-first, Swagger UI at /swagger.
- DDD Domain/ rules layer: profession derivation + applicable policy questions
(DiplomaRules), herregistratie eligibility + reason (HerregistratieRule),
scholing threshold (IntakePolicy), submit rejections + reference generation
(SubmissionRules). In-memory seeded data, ProblemDetails (RFC 7807) errors.
- 27 xUnit tests: rule units + endpoint integration incl. BRP no-address and
DUO not-found fallbacks and 422 submit paths.
Frontend (only infrastructure/ + contracts/ change, as the architecture promised):
- NSwag-generated typed client (api-client.ts), routed through Angular HttpClient
via a small fetch adapter so the ?scenario= interceptor still applies.
- GET adapters use resource({ loader: client.x }); submit commands call the client
and map ProblemDetails -> err. The hardcoded uren==0 / manual-diploma rules are
deleted (now server-side). Domain, stores, UI and format validators unchanged.
- Deleted the now-dead public/mock/*.json.
Tooling/docs:
- npm start proxies /api -> backend; npm run gen:api regenerates the client;
docker compose up runs both (bind mounts use :z for SELinux/Fedora).
- backend/README.md walkthrough: adding a policy question is a one-file backend
change, no FE change, no client regen. Updated CLAUDE.md + ARCHITECTURE.md.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
ehoandClaude Opus 5