Skip to content

feat: integrate open feature PRs behind feature flags - #84

Draft
ccruzkauppila wants to merge 193 commits into
mainfrom
ilmo-hackathon-consolidation
Draft

ccruzkauppila wants to merge 193 commits into
mainfrom
ilmo-hackathon-consolidation

Conversation

@ccruzkauppila

Copy link
Copy Markdown

Summary

Integrates the open feature pull requests (#73, #74, #75, #76, #77, #78, #80, #81, #82, #83) and the feat/event-templates branch into one branch, and puts each new feature behind a feature flag that can be switched from the admin UI. With every flag off the app behaves as main plus the unflagged fixes listed below.

Feature flag system

  • Postgres table app_public.feature_flags with states off, admins_only, on, an available marker for features that need configuration, and an audit log.
  • app_public.feature_enabled(key) for SQL gates, a registry in @app/lib/src/featureFlags.ts, isFeatureEnabled on the server, withFeature for worker tasks, useFeature / <Feature> on the client.
  • /admin/features page and the set_feature_flag mutation are restricted to users with the new Keycloak realm role ilmo-developer (users.is_developer). Regular admins cannot see the page or read the flag table.
  • A test checks the registry against the table and against feature_enabled('...') literals in SQL.

Flags

Flag Feature Source
ribbon_text Custom ribbon text on highlighted events #78
confetti Confetti after a successful registration feat/event-templates
typed_time_input Typed time input in the event form #80
event_templates Quota and question templates feat/event-templates
form_preview Registration form preview in the event editor feat/event-templates
keycloak_names Prefill first and last name from Keycloak #73
calendar_invites Calendar invite in the registration email #83
prefill_answers Remember and prefill registration answers (opt-in) #77
registration_reminders Registration reminder emails #83
telegram_reminders Telegram reminders #83
mcp_server MCP server endpoint #75
payments Stripe payments #81

Unflagged: #82 (staged-file linting, docs-only e2e skip), #76, the fixes in #73, #74 (per-event registration positions, bounded pools), the event form fixes and sticky footer from #80, per-user registration rate limiting from #75.

Structural changes

  • Registration email is built in one module with contributors (calendar invite, payment block).
  • External routes (Telegram webhook, MCP, Stripe webhook) are declared in one registry that drives CSRF exemptions and the flag gate.
  • Event form value mapping lives in EventForm/values.ts; registration form initial values in buildRegistrationInitialValues.
  • PAYMENTS_ENABLED and Query.paymentsEnabled are removed in favour of the payments flag.
  • Attribute round-trip test for event_input and the quota and question input types.

Migrations

000005 to 000017. 000005 commits the two migration files that were pending on main. CI now fails on duplicate prefixes in migrations/current/ and on changes to committed migrations. CONTRIBUTING.md describes the workflow.

000008 sets blank registration names to null before adding a stricter name check. Worth checking against production data before deploying.

Test plan

  • pnpm test --ci --runInBand: 115 suites, 1025 tests pass (flags off by default; feature tests switch their own flags on)
  • pnpm lint, pnpm depcheck
  • data/schema.sql reproduces from scratch with all 17 migrations
  • Cypress (Electron, local): admin_features, create_event, update_event, event_form_templates, create_event_registration, payments_register, reminders
  • Remaining Cypress specs in CI
  • Manual check with real Keycloak (ilmo-developer role), Stripe test keys and a Telegram bot
  • Native review of the Finnish and Swedish texts on the Features page

Notes

  • The jest runner now uses a single recycled worker to stay within memory; this is its first run on GitHub's runners.
  • The migration-commit label used by the committed-migrations check has to be created in the repository.

Generated by Claude Code

claude and others added 30 commits October 17, 2026 06:45
The constrained_name domain and the registration form both rejected any
whitespace in names, to stop people registering with a blank name. That
also blocked real names like "Anna Maria" or "von Wright". Now only
names that are empty or whitespace-only are rejected, the form trims
surrounding spaces before submitting, and the name prefill from the user
profile keeps every word instead of dropping the third one onwards.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0126Kq6TYvUgHFmDYaQL7hp7
CI dumps data/schema.sql from committed migrations, so changes in
migrations/current must not be reflected there until they are committed.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0126Kq6TYvUgHFmDYaQL7hp7
… a global materialized view

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…xhaustion

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…second

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A production-like stack (Postgres with wal_level=replica and 10 slots for
ilmo roles, server pinned to two cores) driven by k6 bots that load the
event page, claim at opening and submit after filling the form. A checker
reports failures, latencies, connection usage and ordering fairness.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The queue promotion trigger read positions without waiting for in-flight
recomputes, so a claim committing between its read and the delete's
recompute could be promoted out of the queue without an email.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Account names are a single field, so a name like "Anna Maria von Wright"
can't be split into first and last name reliably. Leave both fields empty
for names of three or more words instead of guessing.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0126Kq6TYvUgHFmDYaQL7hp7
paymentsEnabled() is the single switch for Stripe payments
(PAYMENTS_ENABLED=1 plus STRIPE_SECRET_KEY and STRIPE_WEBHOOK_SECRET).
Fee helpers work in integer cents: totalChargeCents, guildReceivesCents
(halves round up, computed without float drift), isValidTicketPriceCents
and formatEuro ("18,00 €" in every language). Adds a jest project for
@app/lib with unit tests, and the stripe package (v23) to @app/server
and @app/worker.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RKSTTfRBbDMSCoT7Y56nnd
PaymentsPlugin adds Query.paymentsEnabled (mirroring ssoLoginEnabled) and
SharedLayout_Query fetches it so every page knows whether payment UI
exists. utils/stripe.ts builds the Stripe client on first use and throws
while payments are disabled. The server logs "payments disabled" at
startup when the flag is off, and the setup script prompts for the three
env vars next to the Keycloak ones (.env.ci gets commented placeholders).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RKSTTfRBbDMSCoT7Y56nnd
…th them

Keycloak's ID token carries the standard given_name and family_name
claims, but only the combined name was kept, so the registration form had
to guess where a multi-word first name ends. Store both on the user at
every Keycloak login and prefill the form from them, falling back to the
two-word split for accounts that have never logged in with Keycloak.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0126Kq6TYvUgHFmDYaQL7hp7
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Includes the registration positions table and the pending migrations
from #67 (admin organization access) and #69 (delete omitted quotas and
questions), which production deploys only apply once committed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
pnpm setup only runs committed migrations, so the e2e database lacked
the new users.first_name/last_name columns from migrations/current and
the registration page's query failed. Unit tests already run the current
migrations against their database; do the same for e2e.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0126Kq6TYvUgHFmDYaQL7hp7
…oving

The questions tab renders each row's type and options from a snapshot of
the form values taken when the tab renders. Form.List's move and remove
only re-render the list, so rows kept the type and options of the
question previously at that index. Force a re-render after both, as the
other structural edits in the tab already do.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Record unexpected websocket closes and script exceptions per bot, keep
k6's exit code, surface sampler errors, and have the checker flag runs
with missing bots, no connection samples or a failed k6 as untrustworthy.
Use a neutral default image tag and complete the README tables.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Queue positions are numbered per event, so when the queue head belongs to
another quota the first queued registration of the freed quota has a
position above 1 and got promoted without an email. Pick the lowest queue
position of the quota instead.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…nce view

Add a two-connection test for promotion racing a concurrent claim, and a
guard that compares the maintained positions table with the materialized
view definition of 000002 after inserts, deletes, quota and open quota
size changes and an event delete.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
/mcp is an OAuth-protected Streamable HTTP endpoint. MCP clients sign in with
Prodeko accounts against Keycloak (client ilmo-mcp); each verified access token
maps to an app_private.sessions row of kind mcp, so RLS, ownership and the
existing GraphQL plugins apply exactly as on the web. Tools: list_events,
get_event and register_to_event, which validates answers before claiming a
spot and rolls the claim back if registration fails.

A dev realm with both clients is available via
`docker compose --profile keycloak up -d`.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Users behind a shared IP (an MCP client, the same wifi) no longer share one
limit, and a fresh SSO session cannot reset it. Anonymous requests are still
limited by IP. The worker now receives the exact key to delete so the key
format lives in one place.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Schedule the window's timer before publishing and catch both synchronous
throws and rejected promises from publish, logging them with the source
topic, so a failing publish cannot leave the window open for good.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Errors without a code, such as pg-pool acquire timeouts on the websocket
path, were only returned to the client. Log them with console.error;
errors that carry a SQLSTATE or application code and errors of the GraphQL
request itself stay unlogged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The worker's start scripts expand it in the shell before .env is loaded.
Pass --max-pool-size in the worker's dev script as in its start script.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Adds current/5-payments-schema.sql (idempotent):

- Types payment_mode, payment_method, payment_status,
  registration_payment_status and event_payment_summary.
- events.payment_mode (default NONE), send_unpaid_email_on_close (forced
  on in AFTER_REGISTRATION mode by a check constraint and by
  create_event/update_event), unpaid_close_email_sent_at,
  last_payment_reminder_sent_at; quotas.price_cents (0 or >= 25);
  registrations.price_cents (price snapshot, null = no payment).
- event_input and the create/update_event_quotas types gain the new
  fields; create_event, update_event, create_event_quotas and
  update_event_quotas (built on current/2-*) write them. Null fields keep
  existing values, and the original statements are kept as they were so
  events without payment settings are handled exactly as before.
- Switching an event from NONE to a paid mode backfills price_cents on
  finished registrations in paid quotas (app_hidden helper, since
  update_event runs as the visitor).
- A before trigger snapshots the quota price when a registration of a paid
  event is finished, so create_registration itself is unchanged and the
  confirmation email still fires once.
- app_public.payments (rows survive registration/event/quota deletion via
  on delete set null, one PAID row per registration via a partial unique
  index, select-only RLS for admins and event organization members, no
  write grants) and app_private.stripe_events for webhook idempotency.
- registrations_payment_status and events_payment_summary computed
  columns, null unless the user may manage the event's payments.
- admin_mark_registration_paid / admin_unmark_registration_paid,
  admin_send_payment_reminders, and refund jobs (payment__refund) queued by
  admin_delete_registration(id, refund default true) and
  delete_registration for paid Stripe payments.

The new composite field breaks positional row() casts, so the existing
update_event_quotas test passes a price, and the adminDeleteRegistration
error snapshots now show the (uuid,boolean) signature. data/schema.graphql
is re-exported; data/schema.sql only holds committed migrations and is
unchanged.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RKSTTfRBbDMSCoT7Y56nnd
The description editor reuses one Slate editor for every language and
only swaps editor.children on switch. The cursor, pending marks and undo
history still pointed into the previous language's document, so a cursor
past the end of the new document crashed the editor and lost the text,
and undo replayed the other language's edits. Clear them on switch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e tokens

Users imported with an explicit realmRoles list do not receive
default-roles-ilmo-dev, which carries offline_access; Claude Code requests that
scope and the code exchange was refused.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The per-event recompute compares statuses before and after and enqueues
the spot-from-queue email for each finished registration that leaves
IN_QUEUE. This covers open quota refills and quota or open quota size
increases, which the delete trigger's lookup could not see, so that
trigger is removed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ccruzkauppila and others added 10 commits November 20, 2026 19:13
…lendar

The event form renders before the page query brings the feature flags,
so the templates spec could pick the scroll time picker and alias four
tabs. It now waits for the preview tab (five tabs) first.

The registration spec checked the event URL in the downloaded .ics text,
where lines longer than 75 octets are folded; with a long event slug
the URL spanned two lines. It now unfolds the text before checking.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
The payment code referred to numbered design boards that are not part of
the repository. The comments now say what the code does without them.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
Migration 000008 replaces the constrained_name check, which accepted an
empty string, with one that requires a non-whitespace character. On a
database that stores a blank first or last name, adding the new check
fails and aborts the migration.

The migration now first sets blank names (empty or only whitespace) to
null in app_public.registrations, the only table with constrained_name
columns. Both columns are nullable and null already means that no name
was given: claimRegistrationToken creates registrations without names,
and registrations_full_name is null then. The update runs with the
table's user triggers disabled, so it queues no confirmation emails and
keeps updated_at and updated_by.

A new db test builds a scratch database from migrations 000001-000007,
stores registrations with blank names and applies the remaining
migrations with graphile-migrate; it fails on the previous 000008.

The committed migrations 000008-000017 have not been published yet, so
000008 is edited in place and the graphile-migrate hash chain of
000008-000017 (each file's Hash and the next file's Previous) is
recomputed with graphile-migrate's own calculateHash.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
The reminder and payment docs and comments still named files under
migrations/current/ that are now committed migrations:

- 000016 says it builds on committed/000010-add-ribbon-text-to-events.sql
  instead of current/30-event-ribbon-text.sql. 000016 and 000017 are not
  published yet; their Hash and Previous headers are recomputed.
- docs/reminder-release.md becomes docs/reminder-operations.md ("Running
  registration reminders in production"), names 000013 and 000014, and
  describes the production setup and checks without release wording.
- The reminder docs and the Telegram reply task no longer describe
  possible later versions or earlier deployments of the feature.
- docs/keycloak-sso.md: the developer role is for the people who deploy
  and operate ilmo.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
update_event_questions wrote the given prefill on every update, so an
API caller that does not send prefill cleared the stored marking. It now
follows the update rule of update_event: an omitted (null) prefill keeps
the stored value. To clear it, the caller sends the new input value NONE;
a type other than TEXT also clears it, because only TEXT answers can be
prefilled. NONE is never stored (create_event_questions stores it as
null, and a check rejects it in event_questions.prefill), and the enum
comment documents it in the GraphQL schema.

The event form always sends prefill, NONE for a question without one, and
does not offer NONE as a choice. Migration 000017 is not published yet,
so it is edited in place and its Hash header recomputed. data/schema.sql
and data/schema.graphql are regenerated; codegen output is unchanged.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
Payment comments referred to sections of a product requirements document
that is not in the repository. They now state the rule itself.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
ribbon_text, typed_time_input, event_templates and form_preview had no
test for either state. The event form and event card now render on the
server with each flag off and on: the card shows the event's own ribbon
text only with ribbon_text, and the form adds the typed time picker, the
template selectors and the preview tab only with their flags.

The render mocks the modules a node test cannot load (ESM-only packages,
an image import, the Next router, the drag and drop backend). The typed
picker renders the same input as the scroll column picker until the
calendar opens, so the test replaces it with a marker.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
The event form rendered while the page query was still loading, with
every feature read as off, and then swapped in the typed time picker,
the template selectors and the preview tab once the flags arrived. The
form now waits for the flags (a skeleton until then): the flag context
tells whether they have loaded (useFeatureFlagsLoaded).

The templates spec no longer waits for the fifth tab before using the
tabs, since the form now appears with its final tabs.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
With registration_reminders off, the reminder settings page answered
404, so a user could neither remove a reminder nor disconnect a linked
Telegram account, although both always work in the database.

While the flag is off, the page now shows a notice and only the ways
out: the user's reminders with their remove buttons and the Telegram
connection with Disconnect. Preferences, enabling and linking are not
offered. The page stays in this mode after the last item is removed, and
is still not found for a user who has nothing to remove. The reminders
e2e spec removes a reminder with the flag off.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
…eview

mcp_server only serves MCP clients at /mcp, outside any admin's browser
session, so Admins only behaves exactly like Off for it. A registry entry
can now say so with `adminPreview: false` (set for mcp_server). The
Features page then disables the Admins only choice for that flag, says
why next to the control (en, fi, se), and the turn-on confirmation no
longer suggests a preview first.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
@github-actions

Copy link
Copy Markdown

📦 Next.js Bundle Analysis for @app/client

This analysis was generated by the Next.js Bundle Analysis action. 🤖

⚠️ Global Bundle Size Increased

Page Size (compressed)
global 271.57 KB (🟡 +5.56 KB)
Details

The global bundle is the javascript bundle that loads alongside every page. It is in its own category because its impact is much higher - an increase to its size means that every page on your website loads slower, and a decrease means every page loads faster.

Any third party scripts you have added directly to your app using the <script> tag are not accounted for in this analysis

If you want further insight into what is behind the changes, give @next/bundle-analyzer a try!

New Pages Added

The following pages were added to the bundle from the code in this PR:

Page Size (compressed) First Load % of Budget (450 KB)
/admin/features 374.64 KB 646.2 KB 143.60%
/pay/[updateToken] 248.58 KB 520.14 KB 115.59%
/settings/reminders 303.41 KB 574.98 KB 127.77%

Twenty-six Pages Changed Size

The following pages changed size from the code in this PR compared to its base branch:

Page Size (compressed) First Load % of Budget (450 KB)
/ 347.97 KB 619.54 KB 137.68% (🟡 +0.82%)
/404 227.34 KB 498.91 KB 110.87% (🟡 +1.26%)
/_error 232.27 KB 503.84 KB 111.96% (🟡 +1.26%)
/admin/emails 257.4 KB 528.96 KB 117.55% (🟡 +1.49%)
/admin/event-category/create 287.28 KB 558.84 KB 124.19% (🟡 +1.02%)
/admin/event-category/list 340.16 KB 611.73 KB 135.94% (🟡 +1.30%)
/admin/event-category/update/[id] 287.45 KB 559.02 KB 124.23% (🟡 +1.02%)
/admin/event/create 546.3 KB 817.87 KB 181.75% (🟡 +5.58%)
/admin/event/list 347.29 KB 618.85 KB 137.52% (🟡 +1.13%)
/admin/event/update/[id] 546.43 KB 818 KB 181.78% (🟡 +5.50%)
/admin/organization/[slug] 314.91 KB 586.48 KB 130.33% (🟡 +0.93%)
/admin/organization/create 263.28 KB 534.85 KB 118.86% (🟡 +0.83%)
/admin/users/list 342.88 KB 614.45 KB 136.54% (🟡 +1.85%)
/event/[slug] 402.75 KB 674.32 KB 149.85% (🟡 +7.95%)
/event/[slug]/register/[quotaId] 309.3 KB 580.87 KB 129.08% (🟡 +4.87%)
/forgot 244.75 KB 516.32 KB 114.74% (🟡 +1.17%)
/invitations/accept 228.81 KB 500.38 KB 111.20% (🟡 +1.23%)
/login 258.43 KB 530 KB 117.78% (🟡 +1.18%)
/reset 250.8 KB 522.37 KB 116.08% (🟡 +1.15%)
/settings/accounts 314.31 KB 585.88 KB 130.19% (🟡 +1.24%)
/settings/delete 248.74 KB 520.31 KB 115.62% (🟡 +1.23%)
/settings/emails 292.98 KB 564.54 KB 125.45% (🟡 +1.37%)
/settings/profile 256.83 KB 528.4 KB 117.42% (🟡 +1.80%)
/settings/security 261.44 KB 533.01 KB 118.45% (🟡 +1.32%)
/update-registration/[updateToken] 270.21 KB 541.78 KB 120.39% (🟡 +4.81%)
/verify 227.75 KB 499.31 KB 110.96% (🟡 +1.24%)
Details

Only the gzipped size is provided here based on an expert tip.

First Load is the size of the global bundle plus the bundle for the individual page. If a user were to show up to your website and land on a given page, the first load size represents the amount of javascript that user would need to download. If next/link is used, subsequent page loads would only need to download that page's bundle (the number in the "Size" column), since the global bundle has already been downloaded.

Any third party scripts you have added directly to your app using the <script> tag are not accounted for in this analysis

The "Budget %" column shows what percentage of your performance budget the First Load total takes up. For example, if your budget was 100kb, and a given page's first load size was 10kb, it would be 10% of your budget. You can also see how much this has increased or decreased compared to the base branch of your PR. If this percentage has increased by 20% or more, there will be a red status indicator applied, indicating that special attention should be given to this. If you see "+/- <0.01%" it means that there was a change in bundle size, but it is a trivial enough amount that it can be ignored.

ccruzkauppila and others added 3 commits November 20, 2026 20:40
The event page now loads the event with a query instead of a
subscription, so the event can arrive before the shared query. The page
then read currentUser from undefined data and crashed when opened
directly. It now shows the spinner until both queries have data.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
…tten

The update page hands the form a new initialValues object whenever the
normalized cache writes the event again, for example when another query
on the page loads the same event. The form then reset its fields and
dropped edits made right after it opened, such as a payment mode
choice. It now resets only when the initial values actually change.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
The event page now loads the event over HTTP, so a tampered csrfToken
fails on opening the page rather than on the registration page. The
spec checks the error there, then refreshes and continues to the
registration form as before.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
@github-actions

Copy link
Copy Markdown

📦 Next.js Bundle Analysis for @app/client

This analysis was generated by the Next.js Bundle Analysis action. 🤖

⚠️ Global Bundle Size Increased

Page Size (compressed)
global 271.57 KB (🟡 +5.56 KB)
Details

The global bundle is the javascript bundle that loads alongside every page. It is in its own category because its impact is much higher - an increase to its size means that every page on your website loads slower, and a decrease means every page loads faster.

Any third party scripts you have added directly to your app using the <script> tag are not accounted for in this analysis

If you want further insight into what is behind the changes, give @next/bundle-analyzer a try!

New Pages Added

The following pages were added to the bundle from the code in this PR:

Page Size (compressed) First Load % of Budget (450 KB)
/admin/features 374.64 KB 646.2 KB 143.60%
/pay/[updateToken] 248.58 KB 520.14 KB 115.59%
/settings/reminders 303.41 KB 574.98 KB 127.77%

Twenty-six Pages Changed Size

The following pages changed size from the code in this PR compared to its base branch:

Page Size (compressed) First Load % of Budget (450 KB)
/ 347.97 KB 619.54 KB 137.68% (🟡 +0.82%)
/404 227.34 KB 498.91 KB 110.87% (🟡 +1.26%)
/_error 232.27 KB 503.84 KB 111.96% (🟡 +1.26%)
/admin/emails 257.4 KB 528.96 KB 117.55% (🟡 +1.49%)
/admin/event-category/create 287.28 KB 558.84 KB 124.19% (🟡 +1.02%)
/admin/event-category/list 340.16 KB 611.73 KB 135.94% (🟡 +1.30%)
/admin/event-category/update/[id] 287.45 KB 559.02 KB 124.23% (🟡 +1.02%)
/admin/event/create 546.33 KB 817.89 KB 181.75% (🟡 +5.59%)
/admin/event/list 347.29 KB 618.85 KB 137.52% (🟡 +1.13%)
/admin/event/update/[id] 546.45 KB 818.02 KB 181.78% (🟡 +5.51%)
/admin/organization/[slug] 314.91 KB 586.48 KB 130.33% (🟡 +0.93%)
/admin/organization/create 263.28 KB 534.85 KB 118.86% (🟡 +0.83%)
/admin/users/list 342.88 KB 614.45 KB 136.54% (🟡 +1.85%)
/event/[slug] 402.79 KB 674.35 KB 149.86% (🟡 +7.95%)
/event/[slug]/register/[quotaId] 309.3 KB 580.87 KB 129.08% (🟡 +4.87%)
/forgot 244.75 KB 516.32 KB 114.74% (🟡 +1.17%)
/invitations/accept 228.81 KB 500.38 KB 111.20% (🟡 +1.23%)
/login 258.43 KB 530 KB 117.78% (🟡 +1.18%)
/reset 250.8 KB 522.37 KB 116.08% (🟡 +1.15%)
/settings/accounts 314.31 KB 585.88 KB 130.19% (🟡 +1.24%)
/settings/delete 248.74 KB 520.31 KB 115.62% (🟡 +1.23%)
/settings/emails 292.98 KB 564.54 KB 125.45% (🟡 +1.37%)
/settings/profile 256.83 KB 528.4 KB 117.42% (🟡 +1.80%)
/settings/security 261.44 KB 533.01 KB 118.45% (🟡 +1.32%)
/update-registration/[updateToken] 270.21 KB 541.78 KB 120.39% (🟡 +4.81%)
/verify 227.75 KB 499.31 KB 110.96% (🟡 +1.24%)
Details

Only the gzipped size is provided here based on an expert tip.

First Load is the size of the global bundle plus the bundle for the individual page. If a user were to show up to your website and land on a given page, the first load size represents the amount of javascript that user would need to download. If next/link is used, subsequent page loads would only need to download that page's bundle (the number in the "Size" column), since the global bundle has already been downloaded.

Any third party scripts you have added directly to your app using the <script> tag are not accounted for in this analysis

The "Budget %" column shows what percentage of your performance budget the First Load total takes up. For example, if your budget was 100kb, and a given page's first load size was 10kb, it would be 10% of your budget. You can also see how much this has increased or decreased compared to the base branch of your PR. If this percentage has increased by 20% or more, there will be a red status indicator applied, indicating that special attention should be given to this. If you see "+/- <0.01%" it means that there was a change in bundle size, but it is a trivial enough amount that it can be ignored.

ccruzkauppila and others added 9 commits November 21, 2026 08:16
Rewrite the payments migration around two rules: a registrant pays only
after registration has closed, and nothing is ever refunded.

- An event is paid when one of its quotas has a price; drop the payment
  mode, the close email setting and the per-registration price snapshot.
- A PAID payment keeps a registration paid whatever its quota costs now.
- Drop the refund trigger, refund settings and early refund table, and
  restore the original delete functions.
- Clear unpaid_close_email_sent_at when registration_end_time moves
  later, so that the close email goes out again at the new close.
- Count card payments not linked to a registration in the payment
  summary, for manual refunds in Stripe.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
…funds

- Pay (Checkout) is blocked with PAYMENT_NOT_OPEN while registration is
  open, for every paid event and again when registration is reopened.
- Checkout charges the quota price at checkout time.
- The webhook handles checkout.session.completed and .expired only. A
  payment that can't be applied to its registration (deleted, second
  payment, amount mismatch) is kept as a detached PAID row and logged;
  nothing is refunded.
- Remove the refund fields, PAYMENT_REFUNDED and paymentMode from the
  GraphQL API and the refund variables from the email previews.
- Add a closeEventRegistration Cypress server command.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
- Remove the payment__refund and payment__send_refund_email tasks, the
  refund email and the worker's Stripe client; the worker no longer
  depends on stripe.
- The GDPR registration cleanup is back to a plain delete.
- Payment emails use the quota's current price, include the payment
  link only once registration has closed, and the confirmation no
  longer has a paid variant.
- The close-of-registration email goes out for every paid event, again
  after a reopened registration closes.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
Update the client documents to the simplified payments schema: the
payment summary reads the count of card payments not linked to a
registration, and the queue status query of the registration form is
gone. Regenerate the persisted operations.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
quotaRequiresPayment depends on the flag and the quota price only, the
registration form has a single payment notice instead of variants, the
pay page has no refunded state and cancelRefund is gone. The CSV price
column shows the quota price.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
Remove the per-event payment settings from the event form; an event is
paid when one of its quotas has a price. The registration form shows a
single notice that the payment link is emailed when registration
closes, and no longer redirects to Checkout or waits for the claimed
spot's status.

Deleting a paid registration, by the participant or by an admin, warns
that nothing is refunded automatically. The admin payment summary shows
how many card payments are not linked to a registration, with a hint to
refund them by hand in Stripe. The quota price warning now says that
paid registrations stay paid and unpaid ones pay the new price.

Drop the refund breakdown, the refunded payment states and their
translations.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
Register into a paid quota without going to Checkout, then pay on /pay
once registration has closed. A new closeEventRegistration server
command ends an event's registration for the tests. Deleting a paid
registration, as a participant or as an admin, shows the no-refund
warning, and the admin summary then counts the unlinked card payment.
The event form no longer has a payment mode.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
Payment opens when registration closes, and an event is paid when a
quota has a price. Ilmo handles two Stripe events, the restricted key
needs no refund permission and only the server needs the Stripe
variables. Refunds are made by hand in the Stripe dashboard; describe
when payments end up unlinked from a registration and how admins see
them.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
Admins only set quota prices, registrants pay after registration closes
and nothing is refunded, so the flag's description no longer mentions
choosing when to pay or refunds.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
@github-actions

Copy link
Copy Markdown

📦 Next.js Bundle Analysis for @app/client

This analysis was generated by the Next.js Bundle Analysis action. 🤖

⚠️ Global Bundle Size Increased

Page Size (compressed)
global 271.31 KB (🟡 +5.3 KB)
Details

The global bundle is the javascript bundle that loads alongside every page. It is in its own category because its impact is much higher - an increase to its size means that every page on your website loads slower, and a decrease means every page loads faster.

Any third party scripts you have added directly to your app using the <script> tag are not accounted for in this analysis

If you want further insight into what is behind the changes, give @next/bundle-analyzer a try!

New Pages Added

The following pages were added to the bundle from the code in this PR:

Page Size (compressed) First Load % of Budget (450 KB)
/admin/features 374.49 KB 645.8 KB 143.51%
/pay/[updateToken] 248.05 KB 519.35 KB 115.41%
/settings/reminders 303.27 KB 574.58 KB 127.68%

Twenty-six Pages Changed Size

The following pages changed size from the code in this PR compared to its base branch:

Page Size (compressed) First Load % of Budget (450 KB)
/ 347.8 KB 619.11 KB 137.58% (🟡 +0.79%)
/404 227.2 KB 498.51 KB 110.78% (🟡 +1.23%)
/_error 232.13 KB 503.44 KB 111.88% (🟡 +1.22%)
/admin/emails 257.25 KB 528.56 KB 117.46% (🟡 +1.46%)
/admin/event-category/create 287.13 KB 558.43 KB 124.10% (🟡 +0.99%)
/admin/event-category/list 340 KB 611.3 KB 135.84% (🟡 +1.27%)
/admin/event-category/update/[id] 287.31 KB 558.61 KB 124.14% (🟡 +0.99%)
/admin/event/create 544.05 KB 815.36 KB 181.19% (🟡 +5.08%)
/admin/event/list 347.12 KB 618.42 KB 137.43% (🟡 +1.09%)
/admin/event/update/[id] 544.18 KB 815.48 KB 181.22% (🟡 +5.00%)
/admin/organization/[slug] 314.77 KB 586.07 KB 130.24% (🟡 +0.90%)
/admin/organization/create 263.13 KB 534.44 KB 118.76% (🟡 +0.80%)
/admin/users/list 342.71 KB 614.02 KB 136.45% (🟡 +1.80%)
/event/[slug] 402.42 KB 673.72 KB 149.72% (🟡 +7.87%)
/event/[slug]/register/[quotaId] 298.23 KB 569.53 KB 126.56% (🟡 +2.41%)
/forgot 244.61 KB 515.91 KB 114.65% (🟡 +1.14%)
/invitations/accept 228.67 KB 499.98 KB 111.11% (🟡 +1.20%)
/login 258.29 KB 529.59 KB 117.69% (🟡 +1.15%)
/reset 250.66 KB 521.96 KB 115.99% (🟡 +1.12%)
/settings/accounts 314.17 KB 585.47 KB 130.10% (🟡 +1.21%)
/settings/delete 248.61 KB 519.91 KB 115.54% (🟡 +1.20%)
/settings/emails 292.84 KB 564.15 KB 125.37% (🟡 +1.33%)
/settings/profile 256.69 KB 528 KB 117.33% (🟡 +1.77%)
/settings/security 261.31 KB 532.62 KB 118.36% (🟡 +1.29%)
/update-registration/[updateToken] 259.21 KB 530.51 KB 117.89% (🟡 +2.36%)
/verify 227.6 KB 498.91 KB 110.87% (🟡 +1.20%)
Details

Only the gzipped size is provided here based on an expert tip.

First Load is the size of the global bundle plus the bundle for the individual page. If a user were to show up to your website and land on a given page, the first load size represents the amount of javascript that user would need to download. If next/link is used, subsequent page loads would only need to download that page's bundle (the number in the "Size" column), since the global bundle has already been downloaded.

Any third party scripts you have added directly to your app using the <script> tag are not accounted for in this analysis

The "Budget %" column shows what percentage of your performance budget the First Load total takes up. For example, if your budget was 100kb, and a given page's first load size was 10kb, it would be 10% of your budget. You can also see how much this has increased or decreased compared to the base branch of your PR. If this percentage has increased by 20% or more, there will be a red status indicator applied, indicating that special attention should be given to this. If you see "+/- <0.01%" it means that there was a change in bundle size, but it is a trivial enough amount that it can be ignored.

admin_delete_registration takes only the registration id again, so the
PL/pgSQL context in the error snapshots no longer names a boolean
argument and the not-found RAISE moves to line 12.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
@github-actions

Copy link
Copy Markdown

📦 Next.js Bundle Analysis for @app/client

This analysis was generated by the Next.js Bundle Analysis action. 🤖

⚠️ Global Bundle Size Increased

Page Size (compressed)
global 271.31 KB (🟡 +5.3 KB)
Details

The global bundle is the javascript bundle that loads alongside every page. It is in its own category because its impact is much higher - an increase to its size means that every page on your website loads slower, and a decrease means every page loads faster.

Any third party scripts you have added directly to your app using the <script> tag are not accounted for in this analysis

If you want further insight into what is behind the changes, give @next/bundle-analyzer a try!

New Pages Added

The following pages were added to the bundle from the code in this PR:

Page Size (compressed) First Load % of Budget (450 KB)
/admin/features 374.49 KB 645.8 KB 143.51%
/pay/[updateToken] 248.05 KB 519.35 KB 115.41%
/settings/reminders 303.27 KB 574.58 KB 127.68%

Twenty-six Pages Changed Size

The following pages changed size from the code in this PR compared to its base branch:

Page Size (compressed) First Load % of Budget (450 KB)
/ 347.8 KB 619.11 KB 137.58% (🟡 +0.79%)
/404 227.2 KB 498.51 KB 110.78% (🟡 +1.23%)
/_error 232.13 KB 503.44 KB 111.88% (🟡 +1.22%)
/admin/emails 257.25 KB 528.56 KB 117.46% (🟡 +1.46%)
/admin/event-category/create 287.13 KB 558.43 KB 124.10% (🟡 +0.99%)
/admin/event-category/list 340 KB 611.3 KB 135.84% (🟡 +1.27%)
/admin/event-category/update/[id] 287.31 KB 558.61 KB 124.14% (🟡 +0.99%)
/admin/event/create 544.05 KB 815.36 KB 181.19% (🟡 +5.08%)
/admin/event/list 347.12 KB 618.42 KB 137.43% (🟡 +1.09%)
/admin/event/update/[id] 544.18 KB 815.48 KB 181.22% (🟡 +5.00%)
/admin/organization/[slug] 314.77 KB 586.07 KB 130.24% (🟡 +0.90%)
/admin/organization/create 263.13 KB 534.44 KB 118.76% (🟡 +0.80%)
/admin/users/list 342.71 KB 614.02 KB 136.45% (🟡 +1.80%)
/event/[slug] 402.42 KB 673.72 KB 149.72% (🟡 +7.87%)
/event/[slug]/register/[quotaId] 298.23 KB 569.53 KB 126.56% (🟡 +2.41%)
/forgot 244.61 KB 515.91 KB 114.65% (🟡 +1.14%)
/invitations/accept 228.67 KB 499.98 KB 111.11% (🟡 +1.20%)
/login 258.29 KB 529.59 KB 117.69% (🟡 +1.15%)
/reset 250.66 KB 521.96 KB 115.99% (🟡 +1.12%)
/settings/accounts 314.17 KB 585.47 KB 130.10% (🟡 +1.21%)
/settings/delete 248.61 KB 519.91 KB 115.54% (🟡 +1.20%)
/settings/emails 292.84 KB 564.15 KB 125.37% (🟡 +1.33%)
/settings/profile 256.69 KB 528 KB 117.33% (🟡 +1.77%)
/settings/security 261.31 KB 532.62 KB 118.36% (🟡 +1.29%)
/update-registration/[updateToken] 259.21 KB 530.51 KB 117.89% (🟡 +2.36%)
/verify 227.6 KB 498.91 KB 110.87% (🟡 +1.20%)
Details

Only the gzipped size is provided here based on an expert tip.

First Load is the size of the global bundle plus the bundle for the individual page. If a user were to show up to your website and land on a given page, the first load size represents the amount of javascript that user would need to download. If next/link is used, subsequent page loads would only need to download that page's bundle (the number in the "Size" column), since the global bundle has already been downloaded.

Any third party scripts you have added directly to your app using the <script> tag are not accounted for in this analysis

The "Budget %" column shows what percentage of your performance budget the First Load total takes up. For example, if your budget was 100kb, and a given page's first load size was 10kb, it would be 10% of your budget. You can also see how much this has increased or decreased compared to the base branch of your PR. If this percentage has increased by 20% or more, there will be a red status indicator applied, indicating that special attention should be given to this. If you see "+/- <0.01%" it means that there was a change in bundle size, but it is a trivial enough amount that it can be ignored.

The fake event starts a random time up to seven days after registration
closes, so moving the close a fixed day later sometimes passed the start
and hit the registration end check constraint. Move it halfway to the
start instead.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0156kYgzTdmw7ZBY2vrLwQmA
@github-actions

Copy link
Copy Markdown

📦 Next.js Bundle Analysis for @app/client

This analysis was generated by the Next.js Bundle Analysis action. 🤖

⚠️ Global Bundle Size Increased

Page Size (compressed)
global 271.31 KB (🟡 +5.3 KB)
Details

The global bundle is the javascript bundle that loads alongside every page. It is in its own category because its impact is much higher - an increase to its size means that every page on your website loads slower, and a decrease means every page loads faster.

Any third party scripts you have added directly to your app using the <script> tag are not accounted for in this analysis

If you want further insight into what is behind the changes, give @next/bundle-analyzer a try!

New Pages Added

The following pages were added to the bundle from the code in this PR:

Page Size (compressed) First Load % of Budget (450 KB)
/admin/features 374.49 KB 645.8 KB 143.51%
/pay/[updateToken] 248.05 KB 519.35 KB 115.41%
/settings/reminders 303.27 KB 574.58 KB 127.68%

Twenty-six Pages Changed Size

The following pages changed size from the code in this PR compared to its base branch:

Page Size (compressed) First Load % of Budget (450 KB)
/ 347.8 KB 619.11 KB 137.58% (🟡 +0.79%)
/404 227.2 KB 498.51 KB 110.78% (🟡 +1.23%)
/_error 232.13 KB 503.44 KB 111.88% (🟡 +1.22%)
/admin/emails 257.25 KB 528.56 KB 117.46% (🟡 +1.46%)
/admin/event-category/create 287.13 KB 558.43 KB 124.10% (🟡 +0.99%)
/admin/event-category/list 340 KB 611.3 KB 135.84% (🟡 +1.27%)
/admin/event-category/update/[id] 287.31 KB 558.61 KB 124.14% (🟡 +0.99%)
/admin/event/create 544.05 KB 815.36 KB 181.19% (🟡 +5.08%)
/admin/event/list 347.12 KB 618.42 KB 137.43% (🟡 +1.09%)
/admin/event/update/[id] 544.18 KB 815.48 KB 181.22% (🟡 +5.00%)
/admin/organization/[slug] 314.77 KB 586.07 KB 130.24% (🟡 +0.90%)
/admin/organization/create 263.13 KB 534.44 KB 118.76% (🟡 +0.80%)
/admin/users/list 342.71 KB 614.02 KB 136.45% (🟡 +1.80%)
/event/[slug] 402.42 KB 673.72 KB 149.72% (🟡 +7.87%)
/event/[slug]/register/[quotaId] 298.23 KB 569.53 KB 126.56% (🟡 +2.41%)
/forgot 244.61 KB 515.91 KB 114.65% (🟡 +1.14%)
/invitations/accept 228.67 KB 499.98 KB 111.11% (🟡 +1.20%)
/login 258.29 KB 529.59 KB 117.69% (🟡 +1.15%)
/reset 250.66 KB 521.96 KB 115.99% (🟡 +1.12%)
/settings/accounts 314.17 KB 585.47 KB 130.10% (🟡 +1.21%)
/settings/delete 248.61 KB 519.91 KB 115.54% (🟡 +1.20%)
/settings/emails 292.84 KB 564.15 KB 125.37% (🟡 +1.33%)
/settings/profile 256.69 KB 528 KB 117.33% (🟡 +1.77%)
/settings/security 261.31 KB 532.62 KB 118.36% (🟡 +1.29%)
/update-registration/[updateToken] 259.21 KB 530.51 KB 117.89% (🟡 +2.36%)
/verify 227.6 KB 498.91 KB 110.87% (🟡 +1.20%)
Details

Only the gzipped size is provided here based on an expert tip.

First Load is the size of the global bundle plus the bundle for the individual page. If a user were to show up to your website and land on a given page, the first load size represents the amount of javascript that user would need to download. If next/link is used, subsequent page loads would only need to download that page's bundle (the number in the "Size" column), since the global bundle has already been downloaded.

Any third party scripts you have added directly to your app using the <script> tag are not accounted for in this analysis

The "Budget %" column shows what percentage of your performance budget the First Load total takes up. For example, if your budget was 100kb, and a given page's first load size was 10kb, it would be 10% of your budget. You can also see how much this has increased or decreased compared to the base branch of your PR. If this percentage has increased by 20% or more, there will be a red status indicator applied, indicating that special attention should be given to this. If you see "+/- <0.01%" it means that there was a change in bundle size, but it is a trivial enough amount that it can be ignored.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

7 participants