51 lines
2.8 KiB
Markdown
51 lines
2.8 KiB
Markdown
|
# Automated journey acceptance
|
||
|
|
|
||
|
|
Run `make test` for all regression and layer checks, then `make test-journeys` for
|
||
|
|
the role-based acceptance selection. Tests exercise WSGI routes, authorization,
|
||
|
|
CSRF, persisted records, provider failure/retry and confirmation state. They do
|
||
|
|
not send email or modify real directory users.
|
||
|
|
|
||
|
|
Examples:
|
||
|
|
|
||
|
|
```sh
|
||
|
|
make test-journeys JOURNEY_ARGS="--role user"
|
||
|
|
make test-journeys JOURNEY_ARGS="--role tenant_admin"
|
||
|
|
make test-journeys JOURNEY_ARGS="--role platform_admin"
|
||
|
|
make test-journeys JOURNEY_ARGS="--report /tmp/user-engine-journeys.json"
|
||
|
|
make test-journeys JOURNEY_ARGS="--require-complete"
|
||
|
|
```
|
||
|
|
|
||
|
|
The last command is deliberately a release-completeness gate: it fails while any
|
||
|
|
journey still has an implementation or external acceptance gap. A local suite
|
||
|
|
passing is distinct from all 29 journeys being complete. `tests/journey-coverage.json`
|
||
|
|
contains every journey ID, executable test selectors, implementation status and
|
||
|
|
named remaining work. In particular, OTP help tests do not claim provider OTP
|
||
|
|
activation coverage. KeyCape's Go suite and NetKingdom's provider suite are
|
||
|
|
separate owner checks.
|
||
|
|
|
||
|
|
CI `.forgejo/workflows/journeys.yaml` runs full regression and journey selection
|
||
|
|
against the exact pushed commit, and prints the JSON report. All fake identities,
|
||
|
|
provider failure controls and test sessions live in tests, not production routes.
|
||
|
|
|
||
|
|
Tenant lifecycle and role changes hold an in-memory lock or tenant-scoped
|
||
|
|
PostgreSQL advisory lock across provider calls and local mutation. Tests cover
|
||
|
|
concurrent administrator suspension with independent database connections, lock
|
||
|
|
release on exception and preservation of caller transactions. Nested first-admin
|
||
|
|
bootstrap transactions roll back all local records after a failure. Directory
|
||
|
|
changes and local state still are not one database transaction: desired-state
|
||
|
|
retry/readback is required after a crash between external success and local save.
|
||
|
|
|
||
|
|
`make test-browser-journeys` starts a loopback-only synthetic portal and isolated
|
||
|
|
Chromium profile, then drives user/admin/operator navigation and confirmation
|
||
|
|
with Node's native CDP client. Set `JOURNEY_CHROME` if Chromium cannot be found.
|
||
|
|
Missing browser dependencies fail explicitly; no test login endpoints are added
|
||
|
|
to the production application. The ordinary CI job runs the WSGI suites; the
|
||
|
|
browser command can also run on a runner provisioned with Node and Chromium.
|
||
|
|
|
||
|
|
Database suite: `tests/test_journey_postgres.py` uses the existing
|
||
|
|
`USER_ENGINE_POSTGRES_TEST_DSN` plus `USER_ENGINE_POSTGRES_TEST_RESET=1` contract.
|
||
|
|
Use only a disposable database: conformance tests reset user-engine tables.
|
||
|
|
The 2026-09-13 run used a fresh local Docker PostgreSQL container and completed
|
||
|
|
all 210 regression tests with zero skips. The normal dependency-free run skips
|
||
|
|
seven opt-in PostgreSQL tests. See the rollout evidence for exact commands.
|