upstream-rebase: send open questions and refused pushes to review

[kaniini]

Two cases where the nightly rebase currently does the wrong thing, both seen in yesterday's runs.

1. A clean rebase with an open question would auto-push

Run 37816859023 passed every mechanical check: zero drift and all three builds passing. Claude's report still flagged that upstream 0822a2e890 (unaligned vPCI accesses by the hardware domain) now reaches downstream Arm dom0 vPCI, where nothing on the Arm config path can carry out an unaligned access. Under the current rule that result would have been force-pushed with the question unread. It wasn't pushed only because the push happened to be refused.

Now the push also requires that the report asks for no decision. .github/scripts/report-needs-decision.sh says a decision is needed when either:

• the report's Needs a decision section contains anything other than None., or
• any line is marked UNSURE.

It errs towards a pull request: a stray UNSURE costs one review, while a missed question costs an unreviewed hypervisor change. The skill now tells Claude that raising a question stops the push, and to reserve UNSURE for that.

2. A refused push dropped the result

When the force-push was refused by a branch rule (run #4) or would be refused because edera/4.22 moved under the lease, publish just failed. The result was lost, and the next night repeated the same rebase, paid for it again, and failed the same way. A refused push now falls back to a pull request, stating the refusal as the reason, e.g. the push to `edera/4.22` was refused (`error: GH013: Repository rule violations found …`). The backup branch made for the refused attempt is deleted. If the tip moved, the land workflow already refuses to land a stale result, and the next night supersedes it.

Testing

• report-needs-decision.test.sh, 9 cases:
  • no decision: no section; None.; an empty section.
  • decision: an item; UNSURE outside the section; heading case and trailing space; a level-3 heading inside the section; the real run Fix static plugin linking in oxenstored #4 report.
  • error: an unreadable file.
• The suite passes under gawk and busybox awk. The script avoids regex interval braces, which older mawk (the Ubuntu runner's awk) does not support.
• Checked the reason extraction against the real GH013 rejection text from run Fix static plugin linking in oxenstored #4 and against a real --force-with-lease "stale info" rejection.
• upstream-rebase-selftest.yml runs the new tests and shellchecks the new scripts.
• actionlint and shellcheck are clean, and the existing checker suite still passes.
• Not yet exercised end to end in Actions. The next run that hits either case will.

[github-actions]

PR Review

Reviewed at ed4ae58.

This check did not finish, so it has nothing to say about the diff. The run log has the reason.

Test Coverage

Reviewed at ed4ae58.

This check did not finish, so it has nothing to say about the diff. The run log has the reason.