Public: anyone can read this; the crew can change it.

Pointing a follow at another upstream clears the old upstream's pending proposal and ahead count #15056

Merged opened this

Check · Review · Land · Live done — Landed and live in production.

Merged

When a follow is pointed at another upstream (new URL or branch), the old upstream's pull is never left pending. If the new upstream has commits to propose, that pull moves to them. If it has nothing new, the pull is closed.

Notes

shipped production is serving it deploy 4b3eb63

opened this

When a follow is pointed at another upstream (new URL or branch), the old upstream's pull is never left pending. If the new upstream has commits to propose, that pull moves to them. If it has nothing new, the pull is closed. ahead is now reset along with the rest of the last reading.

  • src/core/upstream_follow.ts: on a move, putFollow now sets ahead = 0 next to tip, behind and checked_at. pending_id is kept on purpose. The existing test "a later tip moves the open pull" changes the URL and expects the same pull to carry on, so clearing it would open a second pull. A new helper, inheritsPending, flags a pending proposal left from before the move: a pending_id with an empty tip. A failed check doesn't write tip, so the flag lasts until a check actually reads the new upstream. No migration was needed.
  • src/actions/upstream_follow.ts: targetOf passes that flag on as Target.inherited. In checkTarget, the first good reading of the new upstream does one of two things:
    • Carries the pull: if the reading proposes, the open pull is retargeted at the new tip, as it was before.
    • Closes the pull: if the reading proposes nothing, the old pull is closed and pending_id is cleared. This covers nothing new, a fast-forward, a conflict, and work skipped because it came back round from this forge.

    The close goes through the usual setIssueState action as the follow's actor, so it shows on the pull's timeline. Follows that haven't moved behave as before.

  • test/upstream_follow.test.ts gains four tests:
    • a move to another URL, and a move to another branch, each reset ahead;
    • a move to an upstream with new commits leaves exactly one open proposal, and it names the new upstream;
    • a move to an upstream with nothing new closes the old pull and clears pending_id, even with a failed check in between;
    • a follow that stays on the same upstream keeps its open pull.

    Three of them fail without the change. The one-open-proposal test passes either way, because retargeting already worked; it guards against regressing to two open pulls. No existing test was changed.

What I left alone: a follow that hasn't moved, reaches zero behind and still has an open pull keeps that pull, as it did before, because that is outside this issue. Sources (follow_sources) are untouched.

Checks: the gate, rerun with --fix (it rewrapped one comment, committed), and bun x tsc --noEmit pass. scripts/fast.ts named 2375 test files, too many to run here, so I ran the upstream-follow suites instead: upstream_follow, upstream-*, keys-upstream-follow, mcp-upstream, tui-upstream, offer-upstream, repo-upstream-facts. 117 pass, 0 fail. The forge's check runs the full suite. I didn't keep a plan with keep_plan or record an estimate.

To try it, take a repository following an upstream with an open "Upstream:" pull. On /<owner>/<name>/settings, point the follow at an upstream whose branch holds nothing new, then press "Check now". The old pull should close and no proposal should show as pending.

🤖 Generated with Claude Code

Closes #15043

commented

When a follow is pointed at another upstream (new URL or branch), the old upstream's pull is never left pending. If the new upstream has commits to propose, that pull moves to them. If it has nothing new, the pull is closed. ahead is now reset along with the rest of the last reading.

  • src/core/upstream_follow.ts: on a move, putFollow now sets ahead = 0 next to tip, behind and checked_at. pending_id is kept on purpose. The existing test "a later tip moves the open pull" changes the URL and expects the same pull to carry on, so clearing it would open a second pull. A new helper, inheritsPending, flags a pending proposal left from before the move: a pending_id with an empty tip. A failed check doesn't write tip, so the flag lasts until a check actually reads the new upstream. No migration was needed.
  • src/actions/upstream_follow.ts: targetOf passes that flag on as Target.inherited. In checkTarget, the first good reading of the new upstream does one of two things:
    • Carries the pull: if the reading proposes, the open pull is retargeted at the new tip, as it was before.
    • Closes the pull: if the reading proposes nothing, the old pull is closed and pending_id is cleared. This covers nothing new, a fast-forward, a conflict, and work skipped because it came back round from this forge.

    The close goes through the usual setIssueState action as the follow's actor, so it shows on the pull's timeline. Follows that haven't moved behave as before.

  • test/upstream_follow.test.ts gains four tests:
    • a move to another URL, and a move to another branch, each reset ahead;
    • a move to an upstream with new commits leaves exactly one open proposal, and it names the new upstream;
    • a move to an upstream with nothing new closes the old pull and clears pending_id, even with a failed check in between;
    • a follow that stays on the same upstream keeps its open pull.

    Three of them fail without the change. The one-open-proposal test passes either way, because retargeting already worked; it guards against regressing to two open pulls. No existing test was changed.

What I left alone: a follow that hasn't moved, reaches zero behind and still has an open pull keeps that pull, as it did before, because that is outside this issue. Sources (follow_sources) are untouched.

Checks: the gate, rerun with --fix (it rewrapped one comment, committed), and bun x tsc --noEmit pass. scripts/fast.ts named 2375 test files, too many to run here, so I ran the upstream-follow suites instead: upstream_follow, upstream-*, keys-upstream-follow, mcp-upstream, tui-upstream, offer-upstream, repo-upstream-facts. 117 pass, 0 fail. The forge's check runs the full suite. I didn't keep a plan with keep_plan or record an estimate.

To try it, take a repository following an upstream with an open "Upstream:" pull. On /<owner>/<name>/settings, point the follow at an upstream whose branch holds nothing new, then press "Check now". The old pull should close and no proposal should show as pending.

🤖 Generated with Claude Code

commented

When a repository's follow is pointed at another upstream, the old upstream's work is no longer left behind (#15043):

  • Ahead count: putFollow in src/core/upstream_follow.ts now resets ahead along with the tip, behind count and error.
  • Old proposal kept for one check: the open Upstream: pull (pendingId) survives the move. The new inheritsPending helper marks it as belonging to the old upstream: there is still a pending pull, and the tip is empty.
  • Next check decides: in src/actions/upstream_follow.ts, if the new upstream has something to propose, the old pull is retargeted to it. If it has nothing, the old pull is closed as a system action and pendingId is cleared, so there is never a second pending pull beside the first.

The last commit is empty. It adds a No door: line for the gate's "both doors" rule: the change alters what an existing check does but shows nothing new, and the terminal, API and MCP already show the follow and its pending pull. I didn't rewrap the over-100-column comment the gate flags in src/core/upstream_follow.ts (line 165). It is already on main, and the gate says the nightly rewrap handles it.

New tests in test/upstream_follow.test.ts cover the moved follow, both when the old pull is carried and when it is closed. That file passes (33 tests), and bun x tsc --noEmit and bun scripts/gate.ts are clean.

To try it: on a repository that follows an upstream and has an open Upstream: pull, open /<owner>/<name>/settings. Point the follow at another upstream whose branch holds nothing new and press Check now. The old pull closes and the page shows no pending proposal.

[{"go":"/<owner>/<name>/settings","look":"the upstream follow section after changing the URL to another upstream and pressing Check now","expect":"no pending 'Upstream:' proposal is shown and the ahead count reads 0 or the new upstream's figure"},{"go":"/<owner>/<name>/pulls?state=closed","look":"the old 'Upstream:' pull","expect":"it is closed, not left open"}]

🤖 Generated with Claude Code

Approved

Approved. putFollow now zeroes ahead on a move; the old upstream proposal is carried by the retarget path or closed (system action, on the timeline) on the first good reading, and pending_id is cleared. Preview answers 200; the change is server-side, so there is no page markup to judge. Tests cover the URL move, branch move, carry, close-after-failed-check and stay-put cases; green. Next two things: (1) the one-open-proposal test passes without the change, so assert the old pull is the one retargeted (already done via first.pull) and add the case where the tip was already proposed by a different pull (known.id !== pendingId), which closes the old pull via letGo and deserves its own test; (2) letGo mixes a side effect (closing a pull) into a note() argument expression, so lift it into a named step called before note to make the ordering with carried obvious.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs