Skip to content

test(e2e): re-enable the configuration export/import suite (#1187) - #1568

Merged
rubenvdlinde merged 8 commits into
developmentfrom
fix/config-export-import-drives-the-real-ui
Aug 25, 2026
Merged

rubenvdlinde merged 8 commits into
developmentfrom
fix/config-export-import-drives-the-real-ui

Conversation

@rubenvdlinde

Copy link
Copy Markdown
Contributor

Four tests, disabled since they were written, now drive the real UI

The stand-down reason named three obstacles. Each is dealt with rather than
described:

the reason said what this PR does
"assumes top-level Export/Import buttons" openStoreActions() opens the NcActions overflow, where CnActionsBar renders manifest headerActions
"needs a seeded configuration group" ensureConfigurationGroup() creates one via OpenRegister's configurations API
"upload fixtures" already untrue — every import test builds its payload in-process with setInputFiles({ buffer })

Configuration groups are a native OpenRegister table
(lib/Db/Configuration.php) exposed at /apps/openregister/api/configurations;
create() takes the entity's own fields and requires admin — which the e2e
session already is. Seeding is cheaper and more honest than standing the suite
down.

A fourth obstacle was never in the reason, because nobody had checked

The helper opened ${APP_BASE}/catalog, and its docblock asserted that page
"hosts the configuration import/export header actions".

No route containing "catalog" exists among this app's 36 manifest pages.
export-configuration / import-configuration are headerActions on the
Store page (/store) — as that page's own _configurationUiNote records,
explaining that configuration groups cannot back a type:index page.

Navigating to a route that does not exist resolves to the default page, so every
assertion here would have run against the Dashboard even with the menu
handled.

The same helper also carried:

// Hash-routed SPA (createWebHashHistory): deep-link via the hash fragment,
await page.goto(`${APP_BASE}/catalog`, …)   // ← a path URL, no hash

The note and the code contradicted each other. integriq builds
createWebHistory(generateUrl('/apps/integriq')), so the path form is right and
the comment was wrong.

Verified before enabling

Every getByTestId in the file resolves to a real element in
ExportConfigurationDialog.vue / ImportPreviewDialog.vue — checked one by
one. The single exception is cn-actions, which belongs to nextcloud-vue's
CnActionsBar, as intended.

The file also failed prettier --check before this change, and CI's format glob
(**/*.{js,ts,vue,css,scss}) does cover it — fixed here rather than left.

Surfaced by the e2e skip-discipline gate, ConductionNL/.github#559.

Four tests, disabled since they were written, now run against the real UI.

The stand-down reason named three obstacles. Each is dealt with rather than
described:

  "assumes top-level Export/Import buttons"
      CnActionsBar renders manifest `headerActions` as NcActionButtons inside
      <NcActions data-testid="cn-actions">, after the built-in Refresh entry.
      openStoreActions() opens that menu and the tests click by menuitem.

  "needs a seeded configuration group"
      ensureConfigurationGroup() makes one. Configuration groups are a NATIVE
      OpenRegister table (lib/Db/Configuration.php) exposed at
      /apps/openregister/api/configurations; its create() takes the entity's own
      fields and requires admin, which the e2e session already is. Seeding is
      cheaper and more honest than standing the suite down.

  "upload fixtures"
      Already untrue when it was written: every test in the import block builds
      its payload in-process with setInputFiles({ buffer }). No checked-in
      binary was ever required.

A FOURTH OBSTACLE WAS NEVER IN THE REASON, BECAUSE NOBODY HAD CHECKED

The helper opened `${APP_BASE}/catalog` and its docblock asserted that page
"hosts the configuration import/export header actions".

No route containing "catalog" exists among this app's 36 manifest pages.
`export-configuration` and `import-configuration` are headerActions on the STORE
page (/store) — as that page's own `_configurationUiNote` records, explaining
that configuration groups cannot back a type:index page. Navigating to a route
that does not exist resolves to the default page, so every assertion in this
file would have run against the Dashboard even with the menu handled.

The same helper carried "Hash-routed SPA (createWebHashHistory): deep-link via
the hash fragment" directly above a path URL with no hash — the note and the
code contradicted each other. integriq builds
`createWebHistory(generateUrl('/apps/integriq'))`, so the path form is right and
the comment was wrong.

VERIFIED BEFORE ENABLING

Every `getByTestId` in the file resolves to a real element:
export-configuration-dialog / -confirm in src/dialogs/ExportConfigurationDialog
.vue; import-preview-dialog, import-file-input, preview-creates, preview-updates,
preview-unresolved, preview-credentials, unresolved-ack, confirm-import,
import-success, import-credentials-summary in src/dialogs/ImportPreviewDialog
.vue. The one exception is `cn-actions`, which belongs to nextcloud-vue's
CnActionsBar, as intended.

The file also failed `prettier --check` before this change, and CI's format glob
(`**/*.{js,ts,vue,css,scss}`) does cover it — fixed here rather than left.
… filter

All four re-enabled tests failed on the same line: "the configurations
endpoint must be reachable". I had written the probe as

    GET /apps/openregister/api/configurations?_limit=50

borrowing the paging parameter from OpenRegister's OBJECTS API. This is a
native-table controller, and it does not page that way:
ConfigurationsController::index() takes $this->request->getParams(), unsets
only `_route`, and hands the REST to the mapper as column filters. So `_limit`
became a filter on a column that does not exist and the request failed.

Exactly the defect this session fixed in learniq, walked into from the other
side. There the objects API took an unrecognised parameter as a property filter
and answered HTTP 200 with an empty set, so six spec files blamed their
fixtures; here the native endpoint fails outright instead of lying. Same root
cause, opposite symptom — and the loud one is the kinder of the two.

The probe needs no paging: it only asks whether any configuration group exists.

Also puts the status code in the assertion message, so the next reader of this
failure is told 400/403/500 rather than "not ok".
@github-actions

Copy link
Copy Markdown
Contributor

Quality Report — ConductionNL/integriq @ be38e03

Check PHP Vue Security License Tests
lint ✅
phpcs ✅
phpmd ✅
psalm ✅
phpstan ✅
phpmetrics ✅
eslint ✅
stylelint ✅
build ✅
build ✅
check-specs ✅
test-l10n ✅
format ✅
check-schema-l10n ✅
composer ✅ ✅ 145/145
npm ✅ ✅ 551/551
app:check-code ⏭️
info.xml ✅
REUSE ❌
PHPUnit ✅
Newman ✅
Playwright ❌
Hydra gates ✅

Quality workflow — 2026-08-24 22:01 UTC

Download the full PDF report from the workflow artifacts.

@github-actions

Copy link
Copy Markdown
Contributor

Quality Report — ConductionNL/integriq @ b45edca

Check PHP Vue Security License Tests
lint ✅
phpcs ✅
phpmd ✅
psalm ✅
phpstan ✅
phpmetrics ✅
eslint ✅
stylelint ✅
build ✅
build ✅
check-specs ✅
test-l10n ✅
format ✅
check-schema-l10n ✅
composer ✅ ✅ 145/145
npm ✅ ✅ 551/551
app:check-code ⏭️
info.xml ✅
REUSE ❌
PHPUnit ✅
Newman ✅
Playwright ❌
Hydra gates ✅

Quality workflow — 2026-08-24 22:49 UTC

Download the full PDF report from the workflow artifacts.

Two of the three remaining failures were the same thing:

    Test timeout of 60000ms exceeded.
    Call log: waiting for getByRole('option').first()

NcSelect wraps vue-select, whose options exist in the DOM only while the
dropdown is OPEN. Clicking the combobox focuses it but does not reliably open
it, so the test waited out the full minute against a listbox that was never
rendered. The failure snapshot shows the dialog with `combobox [active]`, no
listbox, no options, and — because `loading` never cleared — not even the
"no configuration groups" note that would have made the state obvious.

ArrowDown opens the list and highlights the first entry, Enter takes it. The
selection is then ASSERTED rather than assumed: the Export button is disabled
until a group is chosen, so it becoming enabled is the proof, and a failure
there names the real problem instead of resurfacing later as a stuck download.

THE THIRD FAILURE IS LEFT ABLE TO FAIL, DELIBERATELY

"confirming the import writes the entities" got all the way through — preview,
confirm, import-success, and the credentials summary naming the stripped
`apikey` — and then could not find the imported source on /sources. That route
does exist (unlike the /catalog this suite used to open), so the interesting
question is whether the write happened at all.

Rather than soften the assertion, it is now two: first the store is queried
directly for the written source, then the Sources index is checked as a user
would. If the first fails, the import dialog reported success over a write that
did not happen — this session has already found four defects of exactly that
shape, and it should be stated plainly rather than inferred from a missing
table row. If only the second fails, the write landed and the row is merely not
on the first page.
@github-actions

Copy link
Copy Markdown
Contributor

Quality Report — ConductionNL/integriq @ 470cbec

Check PHP Vue Security License Tests
lint ✅
phpcs ✅
phpmd ✅
psalm ✅
phpstan ✅
phpmetrics ✅
eslint ✅
stylelint ✅
build ✅
build ✅
check-specs ✅
test-l10n ✅
format ✅
check-schema-l10n ✅
composer ✅ ✅ 145/145
npm ✅ ✅ 551/551
app:check-code ⏭️
info.xml ✅
REUSE ❌
PHPUnit ❌
Newman ✅
Playwright ❌
Hydra gates ✅

Quality workflow — 2026-08-24 23:34 UTC

Download the full PDF report from the workflow artifacts.

All six PHPUnit cells died on the same line:

    ArgumentCountError: Too few arguments to function
    OCA\OpenRegister\Service\Flow\Nodes\TriggerScheduleNode::__construct(),
    2 passed in tests/Unit/Service/JobToFlowGeneratorTest.php on line 374
    and exactly 3 expected

    Tests: 2671, Assertions: 9355, Errors: 1

OpenRegister's node now takes `(IL10N, IURLGenerator, IUserManager)`; the third
resolves the declared `runAs` to prove the account exists before a flow is
stored. This call still passed two.

The break arrives here without a commit in this repository: this suite installs
openregister from `development` rather than pinning it, so a constructor change
there lands in the next run here. The `class_exists()` guard immediately above
cannot catch it — the class exists; it is its constructor that moved.

`IUserManager` is already imported in this file and mocked a few lines earlier,
so the fix is the argument itself.

Unrelated to the e2e work on this branch, and fixed here because it blocks it.
Both remaining failures in the re-enabled configuration import/export suite
were in the test, not the product. Neither was a stale selector.

1. "an unresolvable slug reference blocks confirmation until acknowledged"
   timed out for the full 60s on `getByTestId('unresolved-ack').click()`,
   reporting "...checkbox-content... intercepts pointer events".

   `data-testid` lands on the `<input>` -- NcCheckboxRadioSwitch spreads
   `$attrs` onto the input, not the wrapper -- and that input is
   `position: absolute; z-index: -1; opacity: 0 !important`. The component
   renders no `<label for>` either: its wrapper element defaults to a plain
   `<span>`. So the visible content span sits on top of the input and a
   click can never reach it. The input is still focusable and `change` is
   the handler the component binds, so Space on the focused checkbox is
   both the accessible interaction and the one that toggles the model --
   the same reason this suite already drives NcSelect by keyboard.

   `toBeChecked()` is asserted explicitly so this cannot decay into a test
   that passes without interacting: a swallowed keypress now names the box
   that was never ticked instead of the button that stayed disabled.

2. "confirming the import writes the entities" failed on
   "the confirmed import must have written the source" -- while the very
   same run's import response said
   `"written": {"sources": ["e2e-import-source"]}`.

   The write happened. The check listed sources unpaged and the objects
   API pages at 20: the response carried `total: 23, pages: 2, limit: 20`,
   and the new row was on page 2. The number being asserted on was the
   page size. That is the trap tests/e2e/ci-seed.sh already documents
   against three earlier specs, reaching this file by a different route.

   The API half now asks for the whole list, and asserts that it really
   got the whole list -- without that guard the `_limit` expires silently
   the next time the fixture grows. The UI half walks the pager the way
   the Source CRUD cycle does, which makes it stricter than a page-1
   check: it fails only when the row is absent from the entire list.
@github-actions

Copy link
Copy Markdown
Contributor

Quality Report — ConductionNL/integriq @ ac58ed6

Check PHP Vue Security License Tests
lint ✅
phpcs ✅
phpmd ✅
psalm ✅
phpstan ✅
phpmetrics ✅
eslint ✅
stylelint ✅
build ✅
build ✅
check-specs ✅
test-l10n ✅
format ✅
check-schema-l10n ✅
composer ✅ ✅ 145/145
npm ✅ ✅ 551/551
app:check-code ⏭️
info.xml ✅
REUSE ❌
PHPUnit ❌
Newman ❌
Playwright ❌
Hydra gates ❌

Quality workflow — 2026-08-25 05:02 UTC

Download the full PDF report from the workflow artifacts.

@github-actions

Copy link
Copy Markdown
Contributor

Quality Report — ConductionNL/integriq @ dd560da

Check PHP Vue Security License Tests
lint ✅
phpcs ✅
phpmd ✅
psalm ✅
phpstan ✅
phpmetrics ✅
eslint ✅
stylelint ✅
build ✅
build ✅
check-specs ✅
test-l10n ✅
format ✅
check-schema-l10n ✅
composer ✅ ✅ 145/145
npm ✅ ✅ 551/551
app:check-code ⏭️
info.xml ✅
REUSE ❌
PHPUnit ❌
Newman ✅
Playwright ✅
Hydra gates ✅

Quality workflow — 2026-08-25 05:34 UTC

Download the full PDF report from the workflow artifacts.

…ntity

Every PHPUnit cell was red on a REAL cross-app break, not a test bug:

    1) testGeneratedConfigPassesTheOpenRegisterNodesOwnValidateConfig
       .../openregister/lib/Service/Flow/Nodes/TriggerScheduleNode.php:176
       Tests: 2671, Errors: 1

OpenRegister's TriggerScheduleNode now calls validateActingIdentity(), which
requires an explicit `runAs` naming an existing account, and says why:

    Nobody is present when a schedule fires, so there is no session to take an
    identity from, and the flow's owner is not used as a fallback — authoring a
    flow is not consent to unattended execution as its author.

This generator emitted `config => ['cron' => $cron]` and nothing else, so every
document it produced is now rejected at validation. The test is named for
exactly this and did its job.

THE PREMISE THIS CODE WAS BUILT ON HAS BEEN REVERSED

scheduleRefusals() refused any job that HAD a userId, reasoning that "a
scheduled flow runs as its OWNER — the account that creates it — and a
generated document cannot set an owner, so the run identity would silently
change". OpenRegister has now made the opposite true: the owner is explicitly
NOT a fallback, and the identity must be named.

So the userId is no longer the problem — it is the answer:

  * nodesFor() carries `runAs` from the job's userId into the trigger config.
  * The refusal is INVERTED: a job that names nobody is refused, because
    generating without a runAs produces a document OpenRegister rejects, and
    choosing an account on the job's behalf would be precisely the silent
    identity change the old reason existed to prevent.

Chosen deliberately over falling back to admin/system, which would make
unattended runs act as a privileged account nobody picked — the thing
OpenRegister's new rule is written to stop.

TESTS

  * The `job()` fixture names a user, since a convertible job now must.
  * testAUserScopedJobIsRefused -> testAJobWithNoActingIdentityIsRefused,
    asserting the new refusal.
  * A new test asserts the trigger carries the job's user as `runAs`.
  * The three-reason refusal test swaps its userId for the absent case, so it
    still exercises what it was written for.
  * The OR-node test configures its IUserManager double to RESOLVE 'alice'. A
    bare mock returns null for everything, so validateActingIdentity() would
    have rejected for a reason unrelated to the generator — and
    `IUserManager::get()` is typed `?IUser`, so the double returns an IUser
    mock rather than a truthy placeholder.
  * The command test's hand-written fixture reason and its assertion no longer
    quote the retired sentence.

NOT DONE, AND WORTH SAYING: no openspec spec describes this generator's refusal
set, so there is nothing to amend alongside this. A rule about which identity
unattended runs act as probably belongs in one.

Verified: `php -l` clean on all three files. PHPUnit could not be run locally —
`composer install` refuses on a lockfile that is already out of date with
composer.json in this repo, which is a pre-existing condition and not touched
here.
rubenvdlinde added a commit to ConductionNL/buildiq that referenced this pull request Aug 25, 2026
Two tests failed on flow creation, not on anything they assert:

    creating the fixture flow must succeed (got 400)
    seeding the definition must succeed (got 400)

Both fixtures build `openregister.trigger-schedule` nodes with `config: {}`.
OpenRegister validates that config on save now and refuses an empty one: the
node requires a five-field `cron`, and TriggerScheduleNode::validateActingIdentity()
requires a `runAs` naming an existing account — "nobody is present when a
schedule fires, so there is no session to take an identity from, and the flow's
owner is not used as a fallback".

These fixtures exist to be exported and re-imported, never to fire, but they
still have to be documents the platform accepts.

`runAs` is the admin, which is the account the e2e session actually runs as, so
the fixture names something true rather than a placeholder that happens to
exist.

THIRD APP HIT BY THE SAME UPSTREAM CHANGE

integriq's JobToFlowGenerator emitted schedule triggers with no `runAs` (fixed
in ConductionNL/integriq#1568) and its PHPUnit went red on the node's own
validateConfig. decidiq and learniq were untouched. Worth noting for anyone
else seeing a sudden 400 on flow-create: the request shape did not change, the
node's contract did.

Verified: `npm run format` clean across the tree.
Down to one error from six:

    1) JobToFlowGeneratorTest::testGenerateForReadsTheJobAndRendersIt
       EntityNotMigratableException: … 1 unsupported feature(s)

testGenerateFor… builds its job record inline through ObjectEntity::setObject()
rather than through the `job()` helper, so it did not pick up the `userId` that
helper gained. A schedule trigger must name the account its runs act as, and a
job naming none is now refused before it can be rendered — which is the new
behaviour working, on a fixture that had not been told about it.

Checked the rest of the suite rather than fixing only the one that shouted:
the other `setObject()` call in this file passes `$this->job()` (already
covered), and every other one in tests/Unit builds a SOURCE, not a job.
@github-actions

Copy link
Copy Markdown
Contributor

Quality Report — ConductionNL/integriq @ a7db70c

Check PHP Vue Security License Tests
lint ✅
phpcs ✅
phpmd ✅
psalm ✅
phpstan ✅
phpmetrics ✅
eslint ✅
stylelint ✅
build ✅
build ✅
check-specs ✅
test-l10n ✅
format ✅
check-schema-l10n ✅
composer ✅ ✅ 145/145
npm ✅ ✅ 551/551
app:check-code ⏭️
info.xml ✅
REUSE ❌
PHPUnit ❌
Newman ✅
Playwright ❌
Hydra gates ❌

Quality workflow — 2026-08-25 05:49 UTC

Download the full PDF report from the workflow artifacts.

…ort-import-drives-the-real-ui

# Conflicts:
#	lib/Service/JobToFlowGenerator.php
#	tests/Unit/Service/JobToFlowGeneratorTest.php
@rubenvdlinde

Copy link
Copy Markdown
Contributor Author

Note on the runAs implementation in this branch

I implemented runAs here derived from the job's own userId, refusing a job
that names nobody. While that was in flight, development gained a different
and better design
(#1569 / d012ded): a per-instance app-config key
job_flow_run_as, read through IAppConfig, with IUserManager proving the
account exists — and, as its own comment puts it:

there is deliberately no default, because every candidate default is a guess
about consent

That is the stronger answer. A job's userId is whoever happened to create it,
which is a weaker basis for unattended execution than an administrator naming
the account once, explicitly.

So this merge takes development's implementation for
JobToFlowGenerator.php and its unit test, and drops mine. Nothing of the
runtime behaviour in this PR is mine any more; what remains is the e2e work the
PR is actually for.

Recording it because the two changes were made in parallel and the conflict
resolution is otherwise invisible in the diff.

@github-actions

Copy link
Copy Markdown
Contributor

Quality Report — ConductionNL/integriq @ 81e65ab

Check PHP Vue Security License Tests
lint ✅
phpcs ✅
phpmd ✅
psalm ✅
phpstan ✅
phpmetrics ✅
eslint ✅
stylelint ✅
build ✅
build ✅
check-specs ✅
test-l10n ✅
format ✅
check-schema-l10n ✅
composer ✅ ✅ 145/145
npm ✅ ✅ 551/551
app:check-code ⏭️
info.xml ✅
REUSE ❌
PHPUnit ✅
Newman ✅
Playwright ✅
Hydra gates ✅

Quality workflow — 2026-08-25 07:22 UTC

Download the full PDF report from the workflow artifacts.

@rubenvdlinde
rubenvdlinde merged commit f1ee1bd into development Aug 25, 2026
88 checks passed
@rubenvdlinde
rubenvdlinde deleted the fix/config-export-import-drives-the-real-ui branch August 25, 2026 08:04
rubenvdlinde added a commit to ConductionNL/buildiq that referenced this pull request Aug 25, 2026
…doned (#459)

* spec(runtime): REQ-OBR-002/003 described an architecture the app abandoned

Both requirements mandated a NESTED CnAppRoot inside the Buildiq shell, with
the outer CnAppNav, header and chrome staying visible, mounted by
BuilderHost.vue, and an OUTER router forwarding an opaque path suffix to an
INNER one.

The product deliberately does the opposite, and says so in its own source.
appinfo/routes.php maps the bare `/builder/{slug}` to `dashboard#builder`, a
standalone page booting the `src/builder.js` webpack entry, whose header states
the reason:

    It is deliberately NOT the Buildiq SPA: rendering the app inside Buildiq's
    shell nests one NcContent in another (double chrome) and, worse, shares
    Buildiq's router — which has none of the app's page routes, so page content
    never resolves.

BuilderHost.vue still exists and is still registered, but only ever mounts for
builder sub-paths falling through to the SPA catch-all — never for the runtime
route these scenarios navigate to. So `[data-testid="buildiq-builder-host"]`
was genuinely absent while the app rendered correctly, and the two e2e tests
could not pass however they were written. They were left permanently skipped
with a note recommending exactly this rewrite. This is that rewrite.

REQ-OBR-002 now describes the standalone shell, and its scenario asserts the
property the old wording had BACKWARDS: one shell, not two.

REQ-OBR-003 now describes the app's own router resolving its own routes, and
drops the outer/inner forwarding machinery, which has no counterpart in the
shipped design. Its scenario is written around opening a row rather than
deep-linking `00000000-0000-0000-0000-000000000000` — a uuid that exists in no
fixture, so the old scenario could not have been asserted even had the
architecture matched.

THE TEST IS REAL, NOT A STUB

The REQ-OBR-002 test now opens `/builder/hello-world`, asserts the app's own
seeded index content, and then COUNTS mounted CnAppRoot instances via the
existing componentTree helper, requiring exactly one. Counting is the point: the
abandoned design would show two, and a plain visibility check cannot tell those
apart. It also asserts the SPA's builder-host wrapper has count 0 on this route.

No test is written for REQ-OBR-003: "REQ-OBR-004 — the seeded index lists the
three sample messages and opens one" already clicks a row, waits for the URL to
move onto the manifest's `/messages/:id` route, and asserts the detail renders.
That IS the rewritten scenario; a second test would add a passing assertion
without adding coverage.

Verified with the repo's own gate — `npm run format`
(prettier --check "**/*.{js,ts,vue,css,scss}") passes across the tree.

* fix(e2e): a schedule-trigger fixture must carry a cron and a runAs

Two tests failed on flow creation, not on anything they assert:

    creating the fixture flow must succeed (got 400)
    seeding the definition must succeed (got 400)

Both fixtures build `openregister.trigger-schedule` nodes with `config: {}`.
OpenRegister validates that config on save now and refuses an empty one: the
node requires a five-field `cron`, and TriggerScheduleNode::validateActingIdentity()
requires a `runAs` naming an existing account — "nobody is present when a
schedule fires, so there is no session to take an identity from, and the flow's
owner is not used as a fallback".

These fixtures exist to be exported and re-imported, never to fire, but they
still have to be documents the platform accepts.

`runAs` is the admin, which is the account the e2e session actually runs as, so
the fixture names something true rather than a placeholder that happens to
exist.

THIRD APP HIT BY THE SAME UPSTREAM CHANGE

integriq's JobToFlowGenerator emitted schedule triggers with no `runAs` (fixed
in ConductionNL/integriq#1568) and its PHPUnit went red on the node's own
validateConfig. decidiq and learniq were untouched. Worth noting for anyone
else seeing a sudden 400 on flow-create: the request shape did not change, the
node's contract did.

Verified: `npm run format` clean across the tree.

* test(e2e): two real runtime bodies, and the header's defect is fixed

THE DEFECT THIS FILE'S HEADER DESCRIBED NO LONGER EXISTS

It read: "DocumentActions filters attachments by object['@self'].schema, which
OpenRegister returns as the NUMERIC schema id ("21"), while a
runtime.documents[] entry declares a schema SLUG ("hello-message"). The two
never match, so the surface renders nothing for every real object."

The component now injects `cnObjectContext` — whose `schema` is the manifest
slug — and resolves candidates through `objectSchemaKeys()`, which collects
`[ctx.schema, obj.schema, self.schemaSlug, self.schema]` and matches any of
them. Slug and numeric id both hit.

A skip reason outliving its cause is how this file came to carry twelve tests
asserting nothing, so it is corrected rather than left to age further. It is the
second such case in this repo today: the `buildiq#41` quarantine cited across 16
files names a PR that MERGED on 2026-07-27.

TWO STUBS BECOME REAL TESTS

Both were `goto('/applications')` + `expect(main).toBeVisible()` — bodies that
would pass without ever reaching a runtime object.

  REQ-DDT-004 "no attachments renders nothing" now opens a real object detail on
  the seeded `hello-world` runtime app (which declares no runtime.documents[] —
  exactly the fixture the scenario needs) and asserts `.ob-document-actions` has
  count 0. The component gates its whole root on `v-if="schemaAttachments
  .length"`, so "renders nothing" is an ABSENT element, not an empty one.

  REQ-DDT-005 "runtime surface degrades without requests" asserts what the DOM
  cannot show: that no request to /apps/docudesk/ is issued at all. A surface
  that renders empty while still calling Docudesk on every object detail is the
  failure this scenario exists to catch, and it is invisible in markup.

THE REQ-DDT-003 STUBS ARE LEFT SKIPPED, DELIBERATELY

They need a published app carrying an attachment — real fixture work, not a
skip to remove. Their notes now say that instead of citing the retired defect.
Enabling them as they stand would produce green tests asserting nothing and
would start crediting their scenarios with coverage they do not have.

Verified: `npm run format` clean across the tree.

* test(e2e): a real nldesign degradation test, and 16 stale quarantine notes

THE QUARANTINE CITED ACROSS THIS SUITE IS STALE

Sixteen spec files deferred to "Conduction/buildiq#41: buildiq admin UI not
functional in this build". #41 is a PULL REQUEST that MERGED on 2026-07-27, and
47 spec files in this suite already pass against that same UI —
applicationDetailOverview.spec.ts alone has 9 passing tests.

Every one of those notes now says what actually blocks its test: the body is a
stub. Leaving them pointed at a merged PR is how a file ends up carrying twelve
tests that assert nothing, which this repo has already done twice.

ONE STUB BECOMES A REAL TEST

REQ-NTS-005 "designer degrades when nldesign is missing" was
`goto('/applications')` + `expect(main).toBeVisible()` — it asserts nothing
about themes and passes on any page that renders.

It now opens the page designer, scrolls to the Theme section, and asserts BOTH
halves of degrading: that the absence is EXPLAINED (`.ob-theme-section__hint`
visible) and that the control which cannot work is disabled rather than
silently inert (`Change` is `:disabled="!nldesignAvailable"`).

Deterministic, not hopeful: this app's CI installs openregister and docudesk and
NOT nldesign (code-quality.yml `additional-apps`), so `nldesignAvailable` is
false and the degraded branch is the one under test.

WHAT IS DELIBERATELY LEFT SKIPPED

The other stubs stay disabled with corrected notes. They need real fixtures — a
published app carrying a theme, a seeded token-set catalogue — and enabling them
as they stand would produce green tests asserting nothing AND start crediting
their scenarios with coverage they do not have. gate-19 counts `skip`/`fixme`
as off, so while skipped they credit nothing; that is the honest state until
bodies exist.

Verified: `npm run format` clean across the tree. The one eslint finding on the
new import is `import-extensions/extensions`, which fires 71 times across
tests/e2e already — no file in this repo uses the `.ts` extension, so the
import follows the established convention rather than inventing a new one.

* test(e2e): dismiss the walkthrough that was covering the runtime row

Both new runtime tests timed out at 30s, and not for want of the element:

    locator.click: Test timeout of 30000ms exceeded
    waiting for getByText('Welcome to Buildiq').first()
      52 x waiting for element to be visible, enabled and stable

The page snapshot shows the runtime app rendered correctly — the "Messages"
heading, the table, and the "Welcome to Buildiq" cell all present. The click
spent its whole timeout on actionability, which is what a covering overlay
looks like.

The runtime app declares a walkthrough that pops a beat AFTER navigation
settles. `dismissOverlays()` from appFixture checks instantaneously and races
it; `dismissFirstVisitOverlays()` from support/overlays polls with `waitFor()`,
which is why buildiq-runtime.spec.ts drives this same route successfully with
it. Switched, and the row is scrolled into view before the click.

`dismissOverlays` stays imported — the designer tests in this file still use it
on a surface that has no walkthrough.

---------

Co-authored-by: Conduction Release Bot <release-bot@conduction.nl>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant