Repository navigation
Process and Review runs that skip command::start may not configure logs.waiting_line_minutes #1983
Description
This was generated by AI while reviewing the pull request for #1495.
Finding
crate::poll::configure_waiting_lines(config.effective_waiting_line_interval()) runs in command::start (src/command.rs:116). The fix for #1495 adds the same call to the Experiment's own start, because the Experiment skips command::start and so its waiting lines ignored logs.waiting_line_minutes.
Other entry points also call logs::configured(&config) themselves, apparently without configure_waiting_lines:
src/processes.rs:650and:742: Process runs, including Passessrc/processes/review.rs:22and:202: Review runs by authorsrc/processes/source.rs:919
If any of these reaches a waiting stage without first configuring the interval, its waiting lines use the built-in 5 minutes. Examples are CI watch, Self-merge and host-resource waits. They would then ignore both logs.waiting_line_minutes and the half-silence fallback for a low inbox.silent_minutes. With inbox.silent_minutes = 4, for example, a waiting stage would still repeat every 5 minutes and could show as Silent.
Evidence
- This is a Standards review finding (possible Duplicated Code / Shotgun Surgery) on the pull request for An Experiment shows as Silent in Needs you while a Trial runs longer than inbox.silent_minutes #1495 (branch
issue-1495). - It has not been reproduced. Neither the reviewer nor the implementer checked whether those paths reach a waiting stage in the same process before some other call configures the interval.
WAITING_LINE_INTERVALis a process-wideOnceLock, so whichever call comes first wins.
Why this was not decided in that pull request
#1495 is limited to an Experiment's Trial waiting lines. Changing how every other Command configures waiting lines is a separate behaviour change, and it has no Spec. It could mean folding the call into logs::configured or a shared start helper. Whether it is a real bug, and which shape to use, needs triage.
Activity
JacobStephens2 commentedon Nov 11, 2026
Agent Brief
Confirmed on Base branch main at d7f9f18. Fix #1983 as one bounded startup-configuration change.
Diagnosis
Process and Review commands bypass command::start (src/main.rs:471–475). Their startup paths call logs::configured at src/processes.rs:650,742, src/processes/source.rs:919 and src/processes/review.rs:22,202, but that function only configures logging (src/logs.rs:313; src/logs/recording.rs:151). Waiting-line configuration currently happens only in command::start and Experiment startup.
These paths reach real waits: Review Delivery calls CI watch (src/processes/review/delivery.rs), and a Custom process copied from processes/ticket-run.toml reaches Delivery and Self-merge. CI watch constructs WaitingLine at src/ci.rs:528. With no configured interval, WaitingLine uses 300 seconds (src/poll.rs:37), bypassing both logs.waiting_line_minutes and UserConfig::effective_waiting_line_interval (src/config.rs:145). Host-resource and Self-merge waits use the same mechanism.
Reproduction
A scratch probe reused tests/support::Scenario with real Git, fake GitHub and fake Harnesses. It held one successful check pending for 14 observations and set THIRDSHIFT_WAITING_MINUTE_MS=150.
With logs.waiting_line_minutes=1, a normal Ticket run emitted seven unchanged CI waiting lines; a Custom process copied from the Built-in ticket-run, an explicit Review run and a Review --author Pass each emitted one. With waiting_line_minutes=5 and inbox.silent_minutes=4, the normal Ticket run emitted five lines; each affected command again emitted one. All eight commands completed successfully after CI passed. The reserved ticket-run name cannot be run directly as a Custom process; copy its definition under another name.
Implementation and acceptance
Capture the effective waiting interval at the existing common startup boundary, logs::configured or a helper it calls, before waits can begin. Remove redundant configuration calls if centralized. Preserve the existing OnceLock first-initialization behavior and test-minute scaling.
Add CLI regression coverage for explicit waiting intervals and the half-silence fallback through Custom process and Review startup, including Review --author. Cover Process Pass and issue startup paths, and verify repeated waiting lines reach both stderr and the Command log. Reuse the fixtures in tests/ci.rs and tests/command_logs.rs. Keep the normal Ticket and Experiment behavior passing.
The 300-second fallback exceeding a four-minute Silent threshold is confirmed by code; an actual Inbox transition to Silent was not observed. No production access or human decision is needed to implement this fix.