test/close-merged-issues.test.js hardcodes the open-request queue, so every device PR that answers a request goes red on all three Node legs #1963
Labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
jaylouisw/packetforge#1963
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
The device stream cannot merge anything: both open device PRs are red on all three Node legs and green on everything else (Classify, PR rebase, Server smoke, Docker build, Arm auto-merge all succeed). The failing assertion is not about the devices.
Evidence —
node test/close-merged-issues.test.jsat each PR head, on current main9d852ed8:mergeable: true)none of the six open requests is in the catalogue ... [#808]... [#810, #808, #807]Everything else in the suite passes. The device specs themselves are fine.
Cause —
test/close-merged-issues.test.js:306-317hardcodes a snapshot of Jay's open device-request queue and asserts that none of those requests is in the catalogue:The assertion is a snapshot of a moving queue. Every PR that answers a request necessarily breaks it — the guard makes "correctly modelling a device" indistinguishable from "regressing the sweep".
Precedent — #1955 (
1304b0bd) walked through this by editing the same literal in the same commit: it dropped[807, ...]and left the comment at line 318 recording why. That is why it went green and the other two did not. The two open PRs have no such edit.Minimal patch (for whoever owns
test/— this stream may not touch it):[810, ...],[808, ...]) as those requests become answered.Durable fix, so this stops recurring —
research/open-device-requests.jsonalready holds the machine-readable queue ([{issue, models, missing}, ...]). Buildqueuefrom it instead of a literal, and assert the invariant that actually matters: no request in the queue is answered by the catalogue unless the sweep would also close it. As written, the test has to be hand-edited by every device PR until the queue empties, and each edit is an opportunity to write a queue that the PR's own devices answer — which is exactly what #1934 did (it edited the literal to seven entries and still lists #807, #808 and #810, all three of which its own catalogue answers).Blocked on this: #1962 is one line from green and mergeable with zero drift from main; #1934 is 9 behind and cannot be rebased mechanically (see that PR). Nothing in the device stream can land until this test stops snapshotting the queue.
Needs Jay's decision on who may edit
test/.