Make Mergecraft's publisher survive a moving base and keep PR text receipts current #286

Activity

1. nisavid

   added

   needs-triageMaintainer needs to evaluate this issueMaintainer needs to evaluate this issue

   on Sep 29, 2026
2. nisavid

   marked this as blocked on Sep 29, 2026

   • Encode, evaluate, and land the seed policies in Mergecraft #210
3. nisavid commented on Sep 29, 2026

   

   nisavid

   on Sep 29, 2026

   OwnerAuthor

   More actions

   Evidence from nisavid/arch-pkgs, 2026-09-26 and 27.

   3a. Creation while the base moved. The PR is nisavid/arch-pkgs#115.

   • The delegate checked origin/main at 7aeab8b. Before create_reviewable_pr.py finished, main moved to 9c4c5b2: docs(utilyze): point the READMEs at the utilyze lane issue (#84) arch-pkgs#111 squash-merged at 2026-09-27T00:45:17Z, touching only packages/utilyze.
   • The helper created draft fix(versionkeeping): reject HTTPS profile drift before credential use #115 with its placeholder body, then exited 1: created PR does not have the exact nonce-tagged transport state. It neither retried nor rolled back.
   • It could not resume: the nonce is random, and a second run refuses because an open PR already exists for the branch.
   • Recovery was a body-only update_reviewable_pr.py text, with a manifest bound to base 9c4c5b2, installing the body the template would have rendered. The merge base, and so the Diff numbers, were unchanged.
   • Result: the first receipt is a text-update receipt (sequence 1, canonical, not-required), and there is no creation receipt. Readiness must now use the numbered manifest; the token manifest can't be used.

   3b. Stale PR text and an unreceipted edit. The PR is nisavid/arch-pkgs#94.

   • Its latest receipt was sequence 14, at head 5e0ae2e. The head had since moved 13 commits, to fe43f42.
   • The body's change navigation still described 5e0ae2e: 10 files and IMPL +5108, where the real figures were 13 files and IMPL +5934.
   • The body had also been edited at 2026-09-26T09:42:34Z with no receipt in the canonical store.
   • The pre-update audit returned drift ("latest receipt belongs to an older PR identity epoch").
   • Nothing flagged the stale navigation until a later bullet-only text update tried to publish. The validator then failed all four Diff checks, and the update had to widen to regenerate the navigation. That was receipt 15.

   Related ergonomics. Two delegates first ran the publisher helpers from the wrong directory. They failed at validation before any write, because the helpers validate against the current directory's repository, which must be the checkout at the PR head.
4. nisavid

   removed

   needs-triageMaintainer needs to evaluate this issueMaintainer needs to evaluate this issue

   on Oct 23, 2026