Repository navigation
Record Platform-Lane Membership and Report Skipped Platform Issues in Unattended Runs #2670
Description
Problem
Pull request #2668 (issue #2610) adds a platform-lane rule to session-handoff "The Chain", with matching text in backlog-burndown "Ranking" and unattended-handoff "Auto-Resolvable". Its local review left two gaps in how a platform issue reaches its lane. Both are about machinery outside the rule itself, so they are filed here rather than grown into that pull request.
Gap 1: nothing records that an issue belongs to a lane
- The split rule files a platform part "as its own issue for its platform's lane". When that lane is open, nothing adds the new issue to the lane's handoff next steps. A session on another track writes only its own link.
- The lane then closes out "once none of its next steps needs its platform", which reads the lane's next steps rather than the qualifying issues the maintainer's rule names.
- Constructed case: a default-track session splits off a WSL part while the
wsllane is open. Thewslsession finishes its listed steps and closes the lane while the new issue is still open, and no seat names who reopens the lane for it.
Gap 2: an unattended run has no channel for a skipped platform or mixed issue
unattended-handoff"Auto-Resolvable" now excludes an issue that needs a platform, in whole or in part, and the picker never splits.- The picker's reply is one fixed line, it writes nothing but the handoff it creates, and the orchestrator's final message lists only round outcome lines.
- So an unattended run skips such an issue every round and never brings it to the maintainer, unlike
backlog-burndown, whose report names each held-back issue.
Done Looks Like
- A stated way an issue joins a lane, for example by being named in the lane's next steps, and a close-out test that reads that membership.
- A named seat that reopens a closed lane for a new qualifying issue.
- An unattended run that reports the platform and mixed issues it skipped, through a channel its return-line contract allows.
Activity
ptr727 commentedon Oct 9, 2026
A further gap of the same kind, pre-existing rather than introduced by #2668:
session-handoff"The Attended Session" step 1 takes the newest open link with no host check, so "resume the handoff" on a Linux host can pick awslorwindowslink whose work that host cannot verify. Only the one-line pick announcement lets the maintainer correct it.ptr727 commentedon Oct 9, 2026
Another gap for this issue, from the local review of #2668: the platform test's first condition is #2610's own "on that platform alone", so a defect that fails on two non-Linux platforms but not on Linux (a constructed case: a case-insensitive path collision that fails on Windows and on default macOS) meets no single lane's test. The rule then sends it to the ordinary backlog, where a Linux worker cannot verify it. The fix needs a stated destination for a multi-platform issue.
ptr727 commentedon Oct 9, 2026
The reverse direction of gap 1, also from #2668's local review: nothing tells a platform lane's session to drop from its next steps an issue that fails the test. An issue a Linux host can fix and verify, still named in an open platform lane's next steps, stays excluded from the unattended picker by Auto-Resolvable's "No open handoff on any track names it" rule for as long as the lane stays open. Constructed case: a
windowslane open on one Windows issue with two systemd issues listed beside it.State When an Issue Belongs in a Platform Handoff Lane (#2668)
ptr727 commentedon Oct 23, 2026
Closed under the maintainer's decision (ptr727, 2026-10-23) on #3391, item 3: the prose and documentation backlog frozen by #1311 in September is closed rather than worked one issue at a time, keeping open only the issues where a document states something the shipped code or workflow does not do. This issue was read as wording, structure, restatement, or a gap in guidance rather than a falsified claim. Reopen with the sentence and the behavior it misstates if that reading is wrong.