Skip to content

refactor(quota): reuse heartbeat receipt identity owner - #9208

Open
cocolord wants to merge 1 commit into
mainfrom
codex/mechanism-caller-cutover-20261121
Open

cocolord wants to merge 1 commit into
mainfrom
codex/mechanism-caller-cutover-20261121

Conversation

@cocolord

@cocolord cocolord commented Nov 21, 2026 •

Copy link
Copy Markdown
Member

Python heartbeat receipt readers could accept Todo bindings that the existing TypeScript settlement owner rejected. Reuse that owner for Python binding and effective-receipt selection, returning the original event by index so unknown details and exact integers stay intact. Locked writers reuse the binding returned by the same batch selection.

The intentional behavior change is rejection of noncanonical Todo IDs with the settlement owner's existing diagnostic and no log mutation. File/GoalRef filtering, locks, admission, lease authority and settlement keep their existing boundaries. The visible-Goal control reuses the proof handling already merged in #9203 and adds File/SQLite coverage plus an explicit missing-proof rejection check. The paired mechanism RFC checkpoint keeps broader caller cutover, installed/native adoption and outcome/full-cost qualification open.

Validation so far:

  • Current source: TypeScript typecheck and configured 17-file kernel Mypy pass; bilingual RFC diff and whitespace checks pass.
  • Initial candidate: 19 receipt tests, 40 typed identity tests and 63 typed recovery/fence tests passed. Four illegal-ID cases fail on immutable baseline b422ace0e and pass on the candidate; the unknown-detail and exact-integer control passes on both.
  • Real File/SQLite public CLI controls retain selection, work obligations, original effect, state bytes and zero extra spend across three reads per backend. These bounded observations do not establish a performance gain.
  • The surrounding five-file Python run stopped with 164 passed and one failed. The same failure reproduces on the immutable baseline: the old visible-Goal test omitted the returned lease proof. Main fix(tests): pass lease_proof to refresh-state in hard-lease spend test #9203 supplies it; this proposal preserves that fix and checks refusal without proof.
  • Exact head 3c4ddcef: the final Python selection expanded to 73 parameterized nodes, exceeding the agreed 36-node execution scope. It was safely interrupted with 35 passed, no test failures and exit 2; this is partial evidence. Those 35 include receipt 21, File/SQLite proof 2, interrupted/resume 5 and journal 7. The source, original Python interpreter and supervision helpers remained unchanged, and owned test runtimes were shut down. Logs and the interrupted namespace are retained. Static enumeration identifies 38 pending nodes (journal 11 and Codex completion 27); verify their collection in the next authorized serial slot and run only those remaining nodes.
  • Final typed identity/recovery controls, real CLI readback and premerge remain queued. CI currently includes successful core, static, build and real PostgreSQL checks, failed browser (2/3), browser (3/3) and Windows PowerShell checks, and pending jobs. The captured browser failures are the same static ContextHelp assertion; Windows fails canonical fixture initialization when a child stdout pipe is selected with Python DefaultSelector (WinError 10038). The affected fixture, backend and presentation sources are byte-identical at this PR base and head. These diagnoses preserve the failed CI state and do not replace the pending checks of this change. This proposal is ready for review; approval and merge qualification remain held pending the necessary evidence.

Signed-off-by: lusendong.6789 <lusendong.6789@bytedance.com>
Co-authored-by: TRAE CLI <traecli@bytedance.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant