Stack panel says a pull request is queued when it has no queue entry

[grubmeshi]

Summary

On a stacked pull request that is not in the merge queue, the stack panel says it is, and offers to remove it. The badges in the same panel report the state correctly, so the panel contradicts itself.

Evidence

Stack main ← #101 ← #102. After Enqueue stack (2), only #101 is admitted. The page of #102 then shows:

• the heading "Queued to merge…"
• "This pull request is next up in the merge queue."
• a "Remove from queue" button
• badges #102 Ready and #101 Queued, which are correct

The merge queue page for the same branch shows 1 Queued and lists only #101:

The API agrees with the queue page, not with the heading:

mergeQueue(branch:"main").entries.totalCount = 1
mergeQueue(branch:"main").entries.nodes      = [{ position: 1, pullRequest: #101 }]

#101   isInMergeQueue true    position 1
#102   isInMergeQueue false   mergeQueueEntry null

#102 reported isInMergeQueue=false at every sample across the whole landing.

Expected

A stacked pull request with no queue entry should not be described as queued:

• do not head the panel "Queued to merge…" on a pull request that is not queued;
• do not say "This pull request is next up in the merge queue" when isInMergeQueue is false. If the intent is that this layer goes after the one below it, say that. It is a statement about the stack, not about queue membership;
• do not offer "Remove from queue" for a pull request with no mergeQueueEntry. The button does not say whether it would remove the layer below, which would be destructive.

Actual

The heading, the sentence and the button all claim queue membership that the API denies, while the badges in the same panel report it correctly.

Environment

• gh 2.100.0, gh stack v0.0.8
• Trunk ruleset:

  merge_queue:   grouping_strategy HEADGREEN, merge_method REBASE,
                 max_entries_to_build 8, max_entries_to_merge 8,
                 min_entries_to_merge 1, min_entries_to_merge_wait_minutes 5,
                 check_response_timeout_minutes 40
  pull_request:  allowed_merge_methods ["rebase"], required_approving_review_count 1,
                 require_last_push_approval false, dismiss_stale_reviews_on_push false
  

• delete_branch_on_merge: true, allow_auto_merge: false

Reproduction

1. Protect a trunk with a merge queue. Run gh stack link <bottom> <top> and get both pull requests approved and green.
2. Press Enqueue stack (2).
3. Open the top pull request and compare its panel against the branch's merge queue page and against pullRequest{ isInMergeQueue mergeQueueEntry{ position } }.

Related

• Stack merge hangs after partial merge invalidates latest-push approval #174 — why the upper layer is out of the queue in the first place
• "Merge stack" on a stacked PR appears to enqueue only the bottom PR #172 — "'Merge stack' appears to enqueue only the bottom PR"
• Merge queue rebases a fast-forward and re-runs CI above it #502 — the wording matters most there, because the rewrite forces a manual re-enqueue and this panel gives no sign of it

[grubmeshi]

A second wording in the same panel describes a state the queue does not have, this time while the enqueue is in flight.

After the layer below merged, #102 had to be enqueued by hand. From the click, the panel showed this for about 60 seconds:

• the heading "Merging this pull request in the stack…"
• "This pull request is being merged."
• a "Cancel merge" button

Throughout that minute the branch's merge queue was empty and #102 reported isInMergeQueue: false. The entry appeared only afterwards, at position 1.

Nothing is lost in the end, so this is milder than the case in the issue above. It is the same shape though: the panel states a queue fact that the queue does not yet hold, and it states it for long enough that a user cannot tell whether the click registered. "Being merged" and "Cancel merge" both suggest the work has started, when what is actually pending is admission to the queue.

Wording that matched the state would say the request is being submitted, and the panel would show the queue position once there is one.

One note on evidence, to keep this honest. gh pr merge answered is already queued to merge at a point when an earlier GraphQL query had reported isInMergeQueue: false, but those two calls were minutes apart, so the entry may simply have been created in between. I am not reporting that as a contradiction. The 60 seconds of "being merged" against an empty queue is what was observed directly.