Skip to content

testing.md

Latest commit

 

History

History
70 lines (52 loc) · 4.76 KB

testing.md

File metadata and controls

70 lines (52 loc) · 4.76 KB
Raw

Testing

Frontend tests (Jest)

cd Applications/Pgan.PoracleWebNet.App/ClientApp
npm test

Uses Jest with jest-preset-angular. Tests cover:

  • Services (user-geofence.service.spec.ts, admin-geofence.service.spec.ts, profile.service.spec.ts)
  • Components (region-selector.component.spec.ts, geofence-submissions.component.spec.ts)
  • Dialogs (geofence-name-dialog.component.spec.ts, geofence-approval-dialog.component.spec.ts, active-hours-editor-dialog.component.spec.ts)
  • Utilities (geo.utils.spec.ts, active-hours.models.spec.ts)
  • Active hours (active-hours-chip.component.spec.ts, location-warning.component.spec.ts)
  • Pipes
  • Pokemon availability (pokemon-availability.service.spec.ts)

Backend tests (xUnit)

dotnet test

Uses xUnit with Moq. Tests cover:

  • Controllers (UserGeofenceControllerTests, AdminGeofenceControllerTests, GeofenceFeedControllerTests, LocationControllerTests, ProfileControllerTests, AreaControllerTests, AdminControllerTests, SettingsControllerTests, ScannerControllerTests, PokemonAvailabilityControllerTests, and all alarm controller tests)
  • Alarm services (MonsterServiceTests, RaidServiceTests, EggServiceTests, QuestServiceTests, InvasionServiceTests, LureServiceTests, NestServiceTests, GymServiceTests) -- these mock IPoracleTrackingProxy
  • Proxy classes (PoracleTrackingProxyTests, PoracleHumanProxyTests) -- verify HTTP request construction, URL encoding, response unwrapping
  • Human/profile services (HumanServiceTests, ProfileServiceTests) -- mock IPoracleHumanProxy for single-user ops, IHumanRepository for admin bulk ops
  • Active hours validation (ActiveHoursValidationTests) -- server-side active hours validation rules
  • Other services (UserGeofenceServiceTests, DiscordNotificationServiceTests, GeoMathTests, CleaningServiceTests, DashboardServiceTests, SiteSettingServiceTests, WebhookDelegateServiceTests, SettingsMigrationServiceTests, QuickPickServiceSecurityTests, PokemonAvailabilityServiceTests)
  • Mapping extensions (MappingExtensionTests) -- alarm DTO To*() / ApplyUpdate() and entity ToModel() / ToEntity() / ApplyTo()

!!! info "Alarm service tests mock IPoracleTrackingProxy" Since alarm services no longer use repositories, their tests mock IPoracleTrackingProxy instead of IRepository. The mock returns JsonElement values matching PoracleNG's snake_case JSON format.

!!! info "Human/profile tests mock IPoracleHumanProxy" HumanServiceTests and ProfileServiceTests mock IPoracleHumanProxy for single-user operations (get, create, exists, location, areas, profile switch, active hours). Admin bulk operations still mock IHumanRepository. LocationControllerTests and AreaControllerTests verify proxy calls with no direct DB interaction. ProfileControllerTests and ProfileServiceTests include extended coverage for active hours CRUD and validation.

Auditing fixes for the defects they introduce

Roughly one in five defects found in this project's audit sweeps was caused by an earlier fix in the same sweep. They cluster into two shapes: a constraint added without enumerating who legitimately depended on the loose rule, and a fix applied to one member of a set of ten while nine siblings are left alone.

.claude/commands/regression-lens.md is a Claude Code slash command (/regression-lens) that audits recent merges asking only what did these fixes break, and which siblings did they miss. Run it after a batch of fixes, scoping each pass to the previous pass's changes, until a pass reports nothing. When it was first used it converged 8 → 5 → 2 → 1 → 0; stopping after one pass would have left five defects live, including a profile-create path that answered 400 while leaving an orphan profile behind.

Two habits that came out of it are worth applying by hand, with or without the tool:

  • Give every guard a legitimate-case-still-passes test, not just a refusal test. Check what real data looks like before tightening a rule — an invasion grunt-type allowlist would have refused blanche and npc 0, both of which exist in production.
  • Revert the fix and confirm the new test goes red. A test written alongside a fix encodes that fix's own assumptions and passes either way. One spec in this repo was asserting a broken request shape, so the suite was defending the bug rather than catching it.

The full rationale is in the "Fixing Defects Without Causing Them" section of CLAUDE.md.

CI

Both test suites run automatically on pushes and pull requests for both main and develop, and against the merge queue. Since pull requests target develop, a workflow filtered to main alone would mean PRs merged with no checks at all. See CI/CD for workflow details.