test(plantext): exercise the PlanText grammar through the UI

[renemadsen]

Follows #1724. The shared parser has ~130 unit tests, but nothing crossed the browser boundary: PlanText is typed in the UI and parsed server-side, and no test proved that wiring.

The seven cases

Each one answered differently through this exact UI path before the migration.

Typed	Asserted	Before
8-16/1.5	break 01:30, hours 6.5	8.0 — break paid as worked time
6.3-16/1	start 06:30	06:03
7,5-16	start 07:30	07:05
8-16/2 helligdag	break 02:00	0
8-16 hjemme	08:00–16:00	threw, aborting the import
5 segments with breaks	hours 6.25	7.75 — the 1.5 on shift 4 zeroed
Ferie	hours 7.4 survive	guards the regression review caught

planHours is the load-bearing assertion: computed from all five shifts less their breaks, so a wrong start, a dropped break or a lost segment all surface in it — no fragile selectors for shifts 3–5 needed.

The Ferie case is the one that would also have passed before. It is deliberate: it guards the data-loss regression introduced while fixing the others, where non-shift text briefly zeroed PlanHours and seeded the flex chain from it.

Where the parse actually happens

Not the save. The seed holds no planning two weeks out, so the working-hours PUT takes the CreatePlanning branch, which stores PlanText verbatim and never parses. The derivation happens on the next dashboard load, in UpdatePlanRegistrationsInPeriod, and is persisted there. The fill → save → open-dashboard → read flow is not three conveniences in a row; the dashboard load is the step under test. This is documented in the spec because it is not obvious and I had it wrong initially.

It also respects both server gates: a future week so the day is past the day of payment, and the day dialog only ever read and cancelled — saving it sets PlanChangedByAdmin, after which the server stops deriving columns from PlanText and every later case would read stale values.

Why a new shard

Shard b holds the existing PlanText spec but its specs share one worker and week with no DB reset between them, which is why the config pins retries: 0. t is isolated, falls back to a's seed automatically, and carries its own activate-plugin.spec.ts (byte-identical to the other 18, deliberately — divergence would break a future consolidation).

Registered in both workflow matrices. Playwright shards are a directory glob with no equivalent of ShardCoverageTests, so an unlisted directory silently never runs.

Review

Dual gate found six real problems, all fixed:

• Two waitForResponse calls armed after the click that fires them — one would deadlock to the 180s timeout, the other could let a stale site-filter response satisfy the date-range wait and read the wrong week while passing green.
• The grid save doesn't parse on this path at all — my comment claimed it did.
• The five-segment rationale was factually wrong: the old parser read all five segments and gave 7.75, it didn't drop shifts 3–5.
• #cell0_{day} is positional and the dashboard wasn't filtered to the worker, so row 0 could have been someone else.
• exactly() was dead weight — toHaveValue with a string is already exact equality, never substring.
• The row-count assertion overclaimed what it checked.

Not verified locally

A hook blocks local Playwright runs and this needs the full stack, so CI is this spec's first execution — and a brand-new shard directory is exactly where a silent mistake hides. I'll watch it to a verdict rather than assume.

🤖 Generated with Claude Code

[Copilot]

Copilot review overview

🔵 Needs a closer look

The new isolated Playwright shard has not been executed locally and requires CI verification.

Review effort: Lite
Findings: None

What changed in this PR

Adds Playwright coverage for PlanText parsing through the UI and registers isolated shard t in CI.

Changes:

• Adds seven grammar and persistence test cases.
• Adds shard t plugin activation setup.
• Registers shard t in PR and master workflows.

File	Summary
eform-client/​playwright/​e2e/​plugins/​time-planning-pn/​t/​plan-text-grammar.spec.ts	End-to-end PlanText parsing and dashboard persistence coverage; description should distinguish regression cases from the deliberate guard case.
eform-client/​playwright/​e2e/​plugins/​time-planning-pn/​t/​activate-plugin.spec.ts	Enables the plugin for shard t.
.github/​workflows/​dotnet-core-pr.yml	Registers shard t for pull-request CI.
.github/​workflows/​dotnet-core-master.yml	Registers shard t for master CI.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Copilot review overview

🔵 Needs a closer look

The new shard has not been verified locally and CI is its first execution.

Review effort: Lite
Findings: None