fix(e2e): the seeds create draft case types, and two stale assertions by rubenvdlinde · Pull Request #1944 · ConductionNL/dossiq

rubenvdlinde · 2026-09-08T13:53:38Z

Ten of the twenty-three failing E2E tests on development. None of these is an
application defect: two behaviours changed on purpose and the tests had not
caught up, and one fixture was invalidated by a shipped schema change.

The other thirteen are separate. Three are the transition 500 in #1941, and ten
still need work, listed at the end.

Eight: the seeded case types are drafts

#1918 added x-relation-filter: {isDraft: false} to case.caseType, and the
caseType schema defaults isDraft to true. So every fixture that creates
a case type without naming the field creates a draft, and the New case picker
correctly refuses to offer it:

Locator: getByRole('option').filter({ hasText: 'E2EZAAK-mtsnt8sg-7325 Subsidie' })
Error: element(s) not found

That reads as a missing option, not as a draft, which is why it looked like
eight unrelated breakages.

Three seeding sites gain isDraft: false: case-create-form, case-identity,
and the shared ensureCaseType helper. The natural experiment is already in
the tree: every spec that already passes the flag has no missing-option
failure, namely seedStateMachine, dashboard-tiles, checklist-per-status
and case-types-tabs.

Excluding drafts from the picker is exactly what #1918 set out to do, and the
schema default means a real user's unpublished case type is correctly hidden
too. So the fixtures are what was wrong.

One: three elements now say Tags

tags became facetable and gained a column, so the index sidebar renders all
of these at once:

element	what it is
`.cn-index-sidebar__filter-label`	the filter label, what the test means
`label.select__label`	that filter's own NcSelect label
`.checkbox-content__text`	the Columns picker checkbox

All three are correct, and a bare getByText(/^(Tags|Labels)$/) is a strict
mode violation reporting "resolved to 3 elements". The assertion now names the
filter label class that CnIndexSidebar.vue renders, which is what the comment
above it always said it meant.

One: the guard refusal moved ahead of the POST, as the spec always asked

Scenario: Required document guard evaluation has always read:

THEN the transition button SHALL be disabled
AND the system SHALL display: "Vereist document ontbreekt: Besluit"

The test was clicking the button, confirming the dialog, and reading the error
out of the failed response. That is the behaviour the spec did not describe,
so the application is now right and the test was asserting the old shape.

It now asserts the disabled button and the reason beside it, and still asserts
the case has not moved, because a disabled button that transitioned anyway
would pass everything else.

That test also anchored at #a-failed-guard-keeps-the-case-where-it-is, which
matches no scenario in that spec. A broken anchor looks exactly like coverage.
It now points at the scenario that exists.

Verification

eslint and prettier are clean on all four files, and all four parse. The suite
itself needs CI: E2E runs only on the development push.

Still open, not in this PR

case-documents.spec.ts 503, 544, 611: an uploaded row never appears, with
no server-side error anywhere in nextcloud.log.
case-list-lenses.spec.ts 281, 300, 320, 334: rows not assigned to the
current user never show on the Cases index. The Mine lenses pass.
case-documents.spec.ts:697 and case-identity.spec.ts:231: two more
application defects, a Generate document 500 and a case number overwritten on
edit.

One process point worth deciding separately: 15 of the last 20 development
runs never reached E2E, because the next merge cancelled them. Every cause
above shipped through that gap.

Ten of the twenty-three failing E2E tests on development, none of which is an application defect. EIGHT: the seeded case types are drafts. #1918 added `x-relation-filter: {isDraft: false}` to `case.caseType`, and the caseType schema defaults `isDraft` to TRUE. So every fixture that creates a case type without naming the field creates a draft, and the New case picker correctly refuses to offer it. The failure reads as a missing option rather than as a draft, which is why it looked like eight unrelated breakages. Three seeding sites gain `isDraft: false`: case-create-form, case-identity and the shared `ensureCaseType` helper. The natural experiment is already in the tree, because every spec that already passes the flag has no missing-option failure: seedStateMachine, dashboard-tiles, checklist-per-status and case-types-tabs. ONE: three elements now say Tags. `tags` became facetable and gained a column, so the index sidebar renders the filter label, that filter's own NcSelect label, and the Columns picker checkbox. All three are correct and a bare getByText is a strict mode violation. The assertion now names the filter label class that CnIndexSidebar renders, which is what the surrounding comment always said it meant. ONE: the guard refusal moved ahead of the POST, and the spec always asked for that. Scenario: Required document guard evaluation reads "the transition button SHALL be disabled" and "SHALL display: Vereist document ontbreekt". The test was clicking through to a dialog and reading the error out of a failed response, which is the behaviour the spec did not describe. It now asserts the disabled button and the reason beside it, and still asserts the case has not moved, because a disabled button that transitioned anyway would pass everything else. That test also anchored at #a-failed-guard-keeps-the-case-where-it-is, which matches no scenario in that spec. A broken anchor looks exactly like coverage. It now points at the scenario that exists. eslint and prettier clean on all four files.

…eds-create-draft-case-types

github-actions · 2026-09-08T14:20:49Z

Quality Report — ConductionNL/dossiq @ `cb91457`

Check	PHP	Vue	Security	License	Tests
lint	✅
phpcs	✅
phpmd	✅
psalm	✅
phpstan	✅
phpmetrics	✅
eslint		✅
stylelint		✅
build		✅
check-manifest		✅
check-vue3-compile		✅
test-l10n		✅
format		✅
check-schema-l10n		✅
check-l10n-js		✅
composer			✅	✅ 106/106
npm			✅	✅ 541/541
app:check-code	⏭️
info.xml	✅
REUSE	❌
PHPUnit					✅
Newman					⏭️
Playwright					⏭️ deferred — runs on the promotion into beta/main, not on a pull request into development
Hydra gates					✅

Quality workflow — 2026-09-08 14:20 UTC

Download the full PDF report from the workflow artifacts.

`case.caseType` has carried `x-relation-filter: {isDraft: false}` since #1918, and the caseType schema DEFAULTS `isDraft` to true. A spec that takes `caseTypes[0]` blind can therefore hand a case a draft type, the relation filters it out, and the case arrives without a usable case type. That failure is invisible in the worst way HERE specifically: the cases list on the contact page comes back empty, which reads as 'this person has no cases' — the exact reading this spec exists to tell apart from a broken filter. The assertion would have been green for the wrong reason on the empty-person test and red without explanation on the seeded one. #1944 fixed the fixtures that CREATE a case type. Six other specs still take `caseTypes[0]` blind and are the same hazard; this is the one file this change owns, and the failure message names the count it saw so the next reader is not left guessing.

Completes the fix. Since #1918 `case.caseType` carries `x-relation-filter: {isDraft: false}` and the caseType schema DEFAULTS `isDraft` to true, so a case seeded against a draft type has no usable type: the list comes back empty and reads as 'there is nothing here' rather than as a bad fixture. #1944 fixed the fixtures that CREATE a type; the six specs that ADOPT one were not covered. Measured, not assumed: of 21 live case types on the dev rig 13 are drafts, so blind [0] adoption picks one more often than not. The register import ships 6 of its 14 with the field ABSENT, which the schema default makes drafts. The two filters had to compose or the change was a regression. Another worker's seeded type is isDraft:false and would be a valid adoption; the FIXTURE_PREFIX filter excludes it, and without this second half those six specs would be pushed onto a DRAFT shipped type instead — making the multi-worker case worse than the serial one it exists to enable. `=== false` not `!== true`: absent counts as a draft, which is what the schema default makes it. The six assertion messages said 'must ship at least one case type', which this filter makes misleading — an instance with 21 case types and no published one would send a reader looking in the wrong place. They now name the filter.

…1984) Seven sites take `listObjects(api, 'caseType')[0]` blind: `ensureCaseType` plus six specs that inline the same adoption (case-communication, case-detail-kpis-and-tabs, case-documents, case-objects, case-parties, case-task-pane). Two things are wrong with that, and they compound. DRAFTS. Since #1918 `case.caseType` carries `x-relation-filter: {isDraft: false}`, and the caseType schema DEFAULTS `isDraft` to true. A case seeded against a draft type has no usable type: the list comes back empty and reads as 'there is nothing here' rather than as a bad fixture. #1944 fixed the fixtures that CREATE a type; the six that ADOPT one were not covered. Measured rather than assumed: of 21 live case types on the dev rig 13 are drafts, so blind [0] picks an unusable one MORE OFTEN THAN NOT, and the register import ships 6 of its 14 with the field absent. FIXTURE-OWNED ROWS. A throwaway seeded by another run can be adopted as though it were instance data (`seeded: false`, so it is never cleaned up) and then deleted by its owner's teardown mid-test. `adoptableCaseTypes()` excludes both. The conditions have to compose: filtering only fixture-owned rows would push these specs onto a DRAFT shipped type instead. `=== false` not `!== true`, because absent counts as a draft, which is what the schema default makes it. NOT 'let every caller seed its own type'. `case` is ARCHIVAL, so a seeded case cannot be deleted in teardown, and a caseType that IS deleted therefore strands permanent cases pointing at a type that is gone. Each of the six inline sites already carries a comment saying exactly that. Also: a per-call suffix on the two seeding paths, because RUN_PREFIX is unique per PROCESS but not across two calls in one worker, so the second create reused the first's identifier; and `isDraft: false` on seedStateMachine's caseType, which #1944's message reports as already done but which only its workflowTemplate had. The six assertion messages said 'must ship at least one case type', which this filter makes misleading: an instance with 21 case types and no published one would send a reader looking in the wrong place. They now name the filter. playwright.config.ts is COMMENT-ONLY. `workers: 1` is unchanged; its comment described the adoption defect as the reason, which is no longer accurate, and left as-is would invite someone to re-fix it. The new comment records that raising it is unverified and what remains unmeasured. Co-authored-by: Conduction Release Bot <release-bot@conduction.nl>

…st (#1986) * fix(e2e): adopt a PUBLISHED case type, not whichever one sorts first Seven sites take `listObjects(api, 'caseType')[0]` blind: `ensureCaseType` plus six specs that inline the same adoption (case-communication, case-detail-kpis-and-tabs, case-documents, case-objects, case-parties, case-task-pane). Two things are wrong with that, and they compound. DRAFTS. Since #1918 `case.caseType` carries `x-relation-filter: {isDraft: false}`, and the caseType schema DEFAULTS `isDraft` to true. A case seeded against a draft type has no usable type: the list comes back empty and reads as 'there is nothing here' rather than as a bad fixture. #1944 fixed the fixtures that CREATE a type; the six that ADOPT one were not covered. Measured rather than assumed: of 21 live case types on the dev rig 13 are drafts, so blind [0] picks an unusable one MORE OFTEN THAN NOT, and the register import ships 6 of its 14 with the field absent. FIXTURE-OWNED ROWS. A throwaway seeded by another run can be adopted as though it were instance data (`seeded: false`, so it is never cleaned up) and then deleted by its owner's teardown mid-test. `adoptableCaseTypes()` excludes both. The conditions have to compose: filtering only fixture-owned rows would push these specs onto a DRAFT shipped type instead. `=== false` not `!== true`, because absent counts as a draft, which is what the schema default makes it. NOT 'let every caller seed its own type'. `case` is ARCHIVAL, so a seeded case cannot be deleted in teardown, and a caseType that IS deleted therefore strands permanent cases pointing at a type that is gone. Each of the six inline sites already carries a comment saying exactly that. Also: a per-call suffix on the two seeding paths, because RUN_PREFIX is unique per PROCESS but not across two calls in one worker, so the second create reused the first's identifier; and `isDraft: false` on seedStateMachine's caseType, which #1944's message reports as already done but which only its workflowTemplate had. The six assertion messages said 'must ship at least one case type', which this filter makes misleading: an instance with 21 case types and no published one would send a reader looking in the wrong place. They now name the filter. playwright.config.ts is COMMENT-ONLY. `workers: 1` is unchanged; its comment described the adoption defect as the reason, which is no longer accurate, and left as-is would invite someone to re-fix it. The new comment records that raising it is unverified and what remains unmeasured. * perf(e2e): run the suite in parallel, with instance-mutating specs last The suite truncates at its 38 minute globalTimeout having produced a result for 111 of 371 tests. Fixing the red tests does not close that, and the arithmetic is worth keeping because the opposite reading is intuitive and wrong. Measured off the log timestamps of run 34244366521: 111 tests produced a result in 37.6 min 20.3s each the 25 failures cost 18.5 min of that 22.2s x2 for the retry if every one became a ~5.1s pass saves 16.4 min all 371 tests at the remaining rate 71 min SERIAL So a fully green suite still runs at roughly twice the budget. Parallelism is the only lever that closes the gap. Also worth recording because it was circulating the other way: the cost is NOT per-file setup. Of 37.6 minutes, 2.3 falls at spec-file boundaries and 35.2 between tests inside a file. THREE WORKERS ON CI, ONE LOCALLY. fullyParallel stays false, so this parallelises at FILE granularity: different files on different workers, tests inside a file still in order on one worker. Several files build shared state in beforeAll and read it across their tests, so the aggressive half of parallelism is not what this suite needs. Three and not four because a SQLSTATE[53200] out of shared memory was seen once under four; the extra six minutes buys distance from a failure mode that gets blamed on the tests. One locally, because a developer runs this against the shared dev instance. The instance-mutating specs move to their own project that runs after the rest. The hazard is asymmetric: a spec installing demo data or writing app settings while empty-state assertions run elsewhere makes those fail and reads as a product defect. The reverse costs nothing, so the list errs toward including a spec. The price is stated in the config rather than left to be discovered: Playwright SKIPS a project whose dependency failed, so while anything in the parallel project is red those 31 tests report as did not run. Verified without touching the instance: the config loads, and --list resolves 371 tests in 67 files across 340 parallel and 31 serial, with all four instance-mutating specs landing only in the serial project. eslint and prettier clean. Whether the run COMPLETES is for CI, which is the standard this change was asked to meet. Stacked on #1984, whose adoptableCaseTypes() removes the cross-worker case-type hazard that pinned this to one worker. This branch carries those commits, so CI here exercises the combination. * perf(e2e): let E2E_WORKERS override the worker count Taken from the four-worker experiment on exp/e2e-four-workers, which had this and is worth keeping whatever that run concludes: the next person to question the number should not have to edit a config to answer it. Default is unchanged, three on CI and one locally. * perf(e2e): four workers, because the run says so This started at three, to keep distance from a SQLSTATE[53200] out of shared memory / max_locks_per_transaction that had been seen once under four. Both counts have now been run against the same tree, so the caution can be settled with a measurement instead of carried: workers reached a verdict passed never ran postgres locks 1 144 108 227 - 3 282 205 89 0 4 309 230 60 0 Four reached more and the lock error did not reappear. One sighting against a clean run is a weak argument, so the measurement wins. The three runs are comparable, which matters because the decision rests on it: each reports Running 371 tests, so the project split does not change what is collected. The four-worker tally sums to 369 rather than 371 because two errors fell outside any test, which both parallel runs report. ⚠️ FOUR WORKERS DOES NOT MAKE THE SUITE FIT and the config now says so. It still truncates with about 60 tests unreached. The rest of the gap is the 44 failures, which cost roughly 32 of the 38 minutes because each is retried. That closes as they are fixed, not by adding workers, and the next raise should come behind a run rather than behind arithmetic. --------- Co-authored-by: Conduction Release Bot <release-bot@conduction.nl>

rubenvdlinde requested review from SudoThijn and remko48 as code owners September 8, 2026 13:53

Conduction Release Bot added 2 commits September 8, 2026 15:59

Merge remote-tracking branch 'origin/development' into fix/the-e2e-se…

4e2e0d8

…eds-create-draft-case-types

Merge remote-tracking branch 'origin/development' into fix/the-e2e-se…

57698fa

…eds-create-draft-case-types

rubenvdlinde merged commit 6f5026d into development Sep 8, 2026
46 of 48 checks passed

rubenvdlinde deleted the fix/the-e2e-seeds-create-draft-case-types branch September 8, 2026 15:01

This was referenced Sep 8, 2026

feat(contacts): contacts as a place you can go #1947

Merged

chore(openspec): sync and archive five completed round 2 changes #1963

Merged

rubenvdlinde mentioned this pull request Sep 8, 2026

fix(e2e): adopt a PUBLISHED case type, not whichever one sorts first #1984

Merged

Navigation Menu

Uh oh!

fix(e2e): the seeds create draft case types, and two stale assertions - #1944

fix(e2e): the seeds create draft case types, and two stale assertions#1944
rubenvdlinde merged 3 commits into
developmentfrom
fix/the-e2e-seeds-create-draft-case-types

rubenvdlinde commented Sep 8, 2026

Uh oh!

github-actions Bot commented Sep 8, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

rubenvdlinde commented Sep 8, 2026

Eight: the seeded case types are drafts

One: three elements now say Tags

One: the guard refusal moved ahead of the POST, as the spec always asked

Verification

Still open, not in this PR

Uh oh!

github-actions Bot commented Sep 8, 2026

Quality Report — ConductionNL/dossiq @ cb91457

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Quality Report — ConductionNL/dossiq @ `cb91457`