Skip to content

fork-seam-check: follow-ups from #436 round 2 (future-dated announcement, expired-announce message, docs, tests) #446

Closed

Description

@scuttletoby

Follow-ups from the #436 round-2 lenses (both SAFE; merged as c1698e3). Trust: #436 (comment) State: #436 (comment)

  1. minor (trust 1): an announcement dated in the future stays usable in PR runs (fork-seam-check.cjs:153 isExpired checks age only; the "dated ahead" refusal at :463 runs only from upstream). Repro: main announces with since 2099-01-01, a PR adding the line exits 0, the run from upstream exits 1. Needs a step-1 merge past its own warning. Fix: also refuse in PR runs when since is more than one day after the run's date.
  2. nit (trust 2): a PR run takes its date from base (:756) but its list from configRef (:731); in fork-checks pull_request.base.sha can be older than DEFAULT_SHA. Date by configRef (matters once item 1 is fixed).
  3. minor (state 1): when fork-follow refuses an expired announcement, the message (:461, footer :830, docs/fork/updates.md:123) offers "add them (step 2)", but by then main's own date is past the window, so step 2 is refused too. Say: re-announce (set since to today) or drop the announcement with a list-only PR, then step 2.
  4. nit (state 2): a late step 2 still prints the generic "A new seam line needs its own entry" hint after the re-announce line (:818).
  5. nit (state 3, docs): "pending takes no new file or higher ceiling" (updates.md:166) contradicts the operator hint to raise a ceiling after a rebase; runbook step 2 does not mention raising a count.
  6. nit (state 4, docs): one since per file: say whether a second announcement in the same file renews or inherits the 14 days.
  7. nit (tests): printable C1-range escape untested (:67, trust 3); mutant M22 (upstream-added check reads base instead of head) survives (state 5).
  8. still open by choice from round 1: a \n inside a listed line allows a hidden second line; bidi characters printed raw; revert-order docs line. Decide or fix.
  9. info: in PR runs the 14 days count main's commit dates, not the calendar (trust 4); new list rules also apply to main's own list, so a merge race producing a bad list would crash every run including the fix PR (state 6; could not be built because the JSON comma edits conflict).

State question: the checker stores nothing; re-runs give the same answer for the same commits.

Activity

  1. added a commit that references this issue on Nov 18, 2026
  2. added 4 commits that reference this issue on Nov 18, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions