ci(ios): keep every step's request log in the Settings replay smoke - #3374

Size Report

Startup median (7 runs, lower is better):

Thanks for the change. I reviewed 84623cc and found no code problems in the diff, but I can't call it ready yet because the PR doesn't show that --debug on test writes a request log for every step, passing ones included, under --artifacts-dir. Please show one iOS Settings replay smoke run, or a local test ... --debug --artifacts-dir run, where the artifact directory holds a request log for each step, and confirm the upload step at .github/workflows/ios.yml#L175 collects that path (https://github.com/callstack/agent-device/blob/84623cc/.github/workflows/ios.yml#L175). Without that, the flag may only raise verbosity while logs stay limited to failing steps, and #3367 would still be open.

The failing check is the Android Smoke Tests job, which fails on the helper-transport assertion expecting 'persistent-session'. It is likely unrelated, because this diff only touches the iOS Settings replay step in .github/workflows/ios.yml. I judged that from the diff scope and the log excerpt, and I did not read the iOS upload step or any iOS run output. There are no conflicts. The remaining step before merge is that run output showing per-step request logs for passing steps.

Not blocking: the same evidence would also confirm what --debug does in the test command, which I did not check in source, and you can take or leave that.

Shown on this PR's own iOS run: 38062260581 at 84623cce9.

• Upload path. "Upload iOS artifacts" uses upload-agent-device-artifacts with agent-state-dir: $AGENT_DEVICE_STATE_DIR. That action uploads ${agent-state-dir}/sessions/** (.github/actions/upload-agent-device-artifacts/action.yml:28). Request logs live there, not under --artifacts-dir.
• Per-step logs. 01-settings.ad passed on attempt 1. The attempt's request log (sessions/…_1-01-settings_attempt-1/requests/<requestId>_attempt_1.ndjson) has 96 entries, with a request_start for each step: open, appstate, click, wait, is, find, back. All 7 steps passed. By command: open 28, click 20, wait 15, find 15, appstate 8, back 7, is 3.
• Before this PR. On main, a failed attempt's log kept only the failing wait (3 entries in run 37815426076), and a passing attempt kept nothing.
• Why. --debug sets the diagnostics scope's debug. emitDiagnostic then writes each event live to the request log (packages/host-kit/src/internal/diagnostics.ts:155-160). Without it, flushDiagnosticsToSessionFile drops the events of a successful request (:218).

This run's open also reported postOpenObservation: "probe-failed", which #3376 relies on. The failing Android Smoke job doesn't touch this diff.

Thanks, that settles it for 84623cc. The iOS run 38062260581 passed at this commit, and the upload step collects sessions/** from the state dir, which is where the request logs live. The source matches your trace: with --debug, each event is written to the request log as it happens, so passing steps keep their logs too. The failing Android Smoke Tests job does not touch this diff. I'm marking this ready for human review.