You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit b849fd8
Browse filesBrowse the repository at this point in the historyBrowse files
fix(ios): press Open again when the runner watchdog abandons the launch-URL accept (callstack#4017)
* fix(ios): press Open again when the runner watchdog abandons the launch-URL accept
On a freshly started runner, the launch confirmation's Open press can time out
on the runner's main-thread watchdog (MAIN_THREAD_TIMEOUT) while the "Open in"
prompt stays up, and open then failed with launch_confirmation_unanswered with
budget to spare. The answer now presses Open again while the same confirmation
is still up, within a 25-second accept budget that holds a whole press. Any
other accept failure, or a prompt that is gone, ends the answer as before.
Fixescallstack#3861
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(ios): state why a second launch-confirmation press cannot answer a later alert
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ios): bound the launch-confirmation re-press by the open request's deadline
The re-press after a watchdog-abandoned accept used a fresh 25 s budget from the
first press, which ignored the time open had already spent, and reserved only
the press watchdog. The daemon now hands the open its request deadline: the
enforced daemon deadline when the request has one, otherwise the transport
admission time plus what the client envelope leaves the daemon. Another press
starts only while that deadline covers the press and its verification read,
each with a readiness preflight, and every runner alert command and the URL
owner lookup are capped at the time left. A request without a deadline is not
re-pressed.
Refs callstack#3861
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ios): take the open deadline from the admitting transport and keep alert failures typed
A request's result deadline is now recorded by the transport that admitted it,
from the top-level command's client envelope, and an attempt registered under a
parent inherits the parent's. A replay test attempt under an unbounded `test`
request therefore has no deadline, instead of one counted from its own
registration with the nested open's budget.
An accept whose deadline already passed floored its retry window to 1 ms and
could skip the loop and throw undefined; the first attempt now always runs.
Refs callstack#3861
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(ios): keep the expired-deadline accept test off real time and name the deadline tests exactly
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ios): refuse launch-confirmation runner steps the deadline cannot hold
The deadline only shrank the alert payload's timeout before the readiness
preflight, and a spent deadline still dispatched with a 1 ms floor. An alert
runner command now carries a host-side deadline that the exchange turns into
the runner timeout right before the payload, after the preflight; a press
left less than its whole watchdog, or any command left no time, is refused
unsent with the typed `runner_command_deadline_spent`. Once the deadline has
passed, the launch confirmation sends nothing, and a press refused unsent
leaves the confirmation unanswered. Runner startup and transport overhead stay
outside this bound.
Refs callstack#3861
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(ios): name the URL-owner lookup cap and pin the unsent press disclosure
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ios): escape the open help apostrophe
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(ios): pin the exact minTimeoutMs edge and narrow the alert deadline doc
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(ios): say the alert deadline caps the native timeout
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ios): gate only the launch-confirmation re-press on the open deadline
Review of 101a8ce: the per-command deadline also governed the first press, so a
first press with less than 10 s of the open budget left was refused unsent where
main pressed it. The request deadline now decides only whether to press again,
at the one re-press gate (a whole press and its read, 22 s); the first press and
every read keep their own runner bounds. This removes the opt-in runner command
deadline, its exchange binding and typed refusal, the alert-leg deadline option
and the URL-owner lookup cap.
A new suite drives the answer through the real alert legs and runner client
against a fake runner: an abandoned press is verified through two RUNNER_BUSY
resends and pressed again with 49.4 s of a 60 s open left, and a first press
with 5 s left is still sent with its full watchdog (refused on 101a8ce).
Refs callstack#3861
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(ios): name the cross-module launch-confirmation runner suite for the runner package index
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
0 commit comments