[kris7011]

Purpose

Let a planner import a week of shifts from a spreadsheet safely: preview every row before anything is saved, see row-level problems, confirm explicitly, and export a week back to CSV. Previously listed under "Future Improvements" as "CSV import/export demo".

Implementation

API

• POST /api/shift-imports/preview (raw CSV body): parses, validates every row, stores a pending import, creates no shifts. Unreadable files → 400 with fileErrors.
• POST /api/shift-imports/{id}/confirm: re-validates the stored file against the current schedule; if still valid, inserts all shifts and marks the import confirmed in one transaction. 409 if already confirmed or no longer valid (with a fresh preview), 404 if unknown.
• GET /api/schedule/export?weekStart=: the Monday–Sunday week in the import format (round-trips).

Validation (ShiftImportValidator)

• Field checks (date, shift type, skill), formula-injection guard, employee matching by name (unknown/ambiguous/lacks skill).
• Reuses the registered ISchedulingRules (max assignments, same day, night→day, high load), with the saved schedule plus earlier file rows as planned shifts, just as schedule generation does.
• Duplicates: same assignment twice in the file → error pointing to the first row; already in schedule → error; repeated open shifts → warning (that is how several employees are requested); same file content already imported → blocked.

Consistency

• One-time confirmation: conditional UPDATE … WHERE Status = 'Pending' inside the insert transaction.
• One import per file content: filtered unique index on ContentHash where confirmed.
• AddShiftImports migration is additive. Demo reset also clears import history.

Frontend: new "Import og eksport" page (Danish), preview table with errors and warnings, confirm button enabled only when valid, export link; rows stack as cards at phone width.

Existing scheduling rules, endpoints and tests are unchanged. The only refactors: ShiftEntity.FromDomain (shared mapping) and a shared test helper for the rule list.

Testing

• dotnet test: 101 unit (38 new) and 46 integration (17 new) passed.
• Frontend: npm test 13 (5 new), npm run lint clean, npm run build OK.
• Mutation checks: removing the Pending condition, skipping re-validation, or removing the file-input reset each makes a test fail.
• Real local demo (Edge via Playwright, API :5026 with a separate SQLite file, Vite :5173):
  • Error sample: 4/11 valid with seven kinds of row errors shown, confirm disabled, shifts unchanged at 14.
  • Valid sample: preview left 14 shifts; confirm → 25.
  • Same file again: blocked.
  • Export shows the 11 shifts; a real browser download produced shift-schedule-2026-05-11.csv.
  • 390 px: no overflow on the import page; no console errors.
  • Screenshots in docs/images/import-*.png.
• The demo found two UI issues, fixed here: the file input kept showing the old file after confirming, and the preview table widened the page on phones.

Notes

• The "Ugeplan" page already overflowed horizontally at 390 px before this change; it is left unchanged.
• The README test counts were stale (59/20); they are now updated to the real numbers.
• Design, trade-offs and interview notes: docs/PROJECT_WALKTHROUGH.md.

🤖 Generated with Claude Code