ActorClient.call and TaskClient.call apply timeout only to the start request #1171

ActorClient.call(timeout=...) passes timeout to start (actor.py#L394-L407) and then calls run_client.wait_for_finish(wait_duration=wait_duration) without it (#L416-L417, #L428-L429). wait_for_finish defaults to timeout='no_timeout' (run.py#L140-L145), so every poll runs without a timeout. TaskClient.call does the same (task.py#L349-L369), and so do both async variants (actor.py#L954-L990, task.py#L757-L776). The docstring says only "Timeout for the API HTTP request".

A caller who passes timeout=timedelta(seconds=90) to guard against a stuck connection gets that guard on the start request only. If a poll connection hangs (a half-open TCP connection, a proxy that never answers), call() hangs with it. The server holds each poll for at most about 60 seconds, so a 90 s timeout would never cut off a healthy poll.

The JS client passes timeoutSecs to both start and waitForFinish (actor.ts#L313-L331, task.ts#L256-L267) and documents it as "Timeout for each API request, the start and every poll alike".

Expected behavior

call() forwards timeout to wait_for_finish in all four variants, and the docstring says it applies to the start request and to every poll.

✍️ Drafted by Claude Code