test(ocpp): stop test chargers after the server (#1712)

[frahlg]

Part of #1712.

Problem

go/internal/ocpp panicked once in CI with send on closed channel (run 39045946853, on #1703, which does not touch OCPP). The stack ends in the test charge point: ocpp-go@v0.19.0/ws.(*Client).Write ← ocppj.(*Client).SendError ← chargePoint.notSupportedError ← readPump. The test running was TestPerChargerCredentialOverTheWire. Its last log line before the panic is Core's capability probe getting NotSupported from garage-impostor.

Core probes each charger it accepts: GetConfiguration, then a TriggerMessage identity probe. Several tests stop their charge point before the server, because the server's cleanup or defer was registered first and runs last. A probe can then reach the charger while it closes. ws.Client.Write checks IsConnected() and then sends on outQueue without its lock, while cleanup() closes outQueue. An answer that passes the check just before the close panics.

The ocpp-go client runs only in tests and cmd/sim-ocpp; Core uses its server. There is no newer ocpp-go release than v0.19.0.

Change

stopAfterServer(t, srv, stop) (in server_test.go) stops the server first at cleanup, then the charger. Server.Stop is idempotent (stopOnce) and stops the identity probes before it closes connections. It replaces the charger cleanup where the charger used to stop first:

• auth_test.go (TestPerChargerCredentialOverTheWire)
• identity_test.go (TestBootNotificationTakesEitherSerialField)
• server_test.go (three tests that used defer cp.Stop() after defer srv.Stop())
• control_test.go connectCharger, control_v201_test.go connectStationV201, charging_needs_test.go connectStationForNeeds

Tests that stop the server with defer already stopped it before their t.Cleanup chargers, so capabilities_test.go and most of control_test.go were safe and do not change.

Not covered

• Tests that disconnect a charger on purpose mid-test, while the server runs (TestCapabilityProbeMarkerClearedOnDisconnect, the socket and disconnect tests), keep the library's narrow window. Closing it needs a fix in ocpp-go.
• The other tests listed in Flaky api and ocpp tests fail CI on unrelated PRs #1712: TestForecastRetentionLargeOverflowWithinScoringBudget is already skipped in the race job by .github/scripts/race.sh and still runs in the plain job. TestTheBoxOpensAnExchangeBesideItsUplink lives in appuplink/exchange_test.go, which fix(cloud): remove retired relay code and DNS dependency #1691 changes. TestAPlugInReachesAStreamingPhoneAsAPlan lives in cmd/ftw/app_link_plan_push_test.go, which fix(ev): preserve confirmed power history and verify charging recovery #1675 changes and already makes wait for the published revision. Both open PRs have right of way.

Evidence

• The panic did not reproduce locally in 2,000 runs of TestPerChargerCredentialOverTheWire with -race (four parallel runs of 500, GOMAXPROCS=2); the window is a few instructions wide. The fix is by construction: no probe can be sent once the server has stopped.
• go test -race -count=20 ./internal/ocpp/: ok (57.8 s).
• go vet ./internal/ocpp/: ok.
• make verify was not run locally; CI runs the full suite.

Test-only change, so it carries the no-changeset label.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-17T18:35:12.478666Z	194362c	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.