Found while building the merge queue (#1264, PR #1273). It is out of that PR's scope, so it is filed here rather than folded in.
With the merge queue, every queued PR runs the FULL main lane on merge_group, including docs-only PRs. scope only classifies docs-only changes on pull_request, not on merge_group. The queue's wall time is set by macos / macos-latest running the full suite plus [99i]: about 35 min on main, against 16 on a PR (measured by the #1273 round-1 critic).
Decide: either teach scope to classify the queue candidate's diff against its base on merge_group (so docs-only merges skip the heavy lanes and still report the required checks as success), or accept the cost, since merges are a few a day. Measure the queue wall time over the first week before choosing.
Found while building the merge queue (#1264, PR #1273). It is out of that PR's scope, so it is filed here rather than folded in.
With the merge queue, every queued PR runs the FULL main lane on
merge_group, including docs-only PRs.scopeonly classifies docs-only changes onpull_request, not onmerge_group. The queue's wall time is set bymacos / macos-latestrunning the full suite plus [99i]: about 35 min on main, against 16 on a PR (measured by the #1273 round-1 critic).Decide: either teach
scopeto classify the queue candidate's diff against its base onmerge_group(so docs-only merges skip the heavy lanes and still report the required checks as success), or accept the cost, since merges are a few a day. Measure the queue wall time over the first week before choosing.