TOP PRIORITY: consolidate all open PRs onto one next branch, one PR #189
Reference in New Issue
Block a user
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?
This outranks all feature work in this repo. Do it before anything else.
Owner instruction, sneak 2026-08-10:
He squashes on merge, so N open PRs means N-1 conflict the moment the first lands. This repo has already produced exactly that: #178 was built on #169's branch and went un-mergeable the instant #169 landed.
What to do
Fold every open PR here onto a single
nextbranch, resolving the conflicts between them yourself:needs-rebase— it targeted #169's branch, which has since merged tomain)needs-rework; include it only if its rework completes, otherwise leave it out and note why)TODO.mdwill conflict across most of them; #178's edit is deliberately a single insert at the top of Completed Steps to make that trivial.Definition of done
nextbranch containing the included PRs as separate issue-closing commits, each message ending(closes #N).next->main, titled for the milestone.make checkgreen on the combined result, andmake buildstill produces release artifacts withDEBUGoff — the guard added by #178 must pass against the consolidated tree, not just its own branch.nextbefore closing it.Model going forward
nextis the branch for the NEXT MILESTONE, not the next release;mainchanges between releases as milestones land. Every worker works in its own clone, pullsnextbefore starting, and pulls and resolves conflicts itself immediately before pushing. Exactly one open PR in this repo at all times from now on.Correction to the rule as I originally stated it. sneak, 2026-08-10:
> you can have multiple open PRs per repo if the milestone on
nextis completed and hasn't been merged tomainyet. you can work on the next milestone off ofnexton a new PR (mark wip). i just don't want one PR per small feature branchSo the target is not "exactly one open PR, always" — it is one PR per milestone, never one per feature branch. A second open PR is legitimate when a finished milestone PR is waiting on sneak and the following milestone has started off
next(not offmain) under aWIP:title prefix.The consolidation work here is unchanged: these are per-issue PRs, which is exactly the shape he does not want.
Also done, so nobody has to remember not to merge mid-consolidation: every open PR in this repo is now titled
WIP: …, labelledneeds-rebase, and assigned toclawbot. Strip the prefix and reassign only when the milestone PR is genuinely ready. His instruction: "don't tell me not to merge things, rename the PRs with WIP: prefix so they aren't mergeable."Superseded. sneak, 2026-08-10: "branch and pr per issue, landing on next, squash merged by manager on successful review. manager owns dispatching workers to resolve merge conflicts. next is always ready for me to merge to main without notice." And: "manager merges the pr to next i mean, not directly with git."
So do NOT consolidate or close these PRs. Retarget each to base
next, squash-merge it there through Gitea on a passed review, and dispatch a worker for any that will not merge cleanly.mainis only ever touched by the milestone PR fromnext, which sneak merges.Feature branches may be red;
nextandmainmust stay green. Createnextfrommainif it does not exist here yet. Drop theWIP:prefixes once retargeted.Done, in the corrected shape:
nextcreated frommain, all four PRs retargeted to basenext, rebased by a worker each, and squash-merged intonextthrough the tracker in this order — #185, #175, #171, #178. No PR was closed or consolidated; each kept its own branch, review history and closing commit, and each branch is now deleted.nextate9fa8be:make checkgreen (7 suites, 143 tests),make buildgreen withdist/chrome/anddist/firefox/produced and all four bundles verifiedautistmask-build-debug=off, so the guard from #178 passes against the consolidated tree and not just its own branch.The milestone PR is #190,
merge-readyand assigned to you.Each of the four kept the PASS review it already had; the reviewed content was verified byte-identical after rebase in three cases, and #178 — whose rebase dropped an already-landed commit and resolved
Makefile/README.md— got a fresh independent review that also passed.