Files
2026-10-12 19:13:13 -05:00

2.6 KiB
Raw Permalink Blame History

Findings remediation plan

This queue groups the open issues in FINDINGS.md into reviewable implementation batches. The batches follow package ownership and shared behavior; a finding that spans packages is completed only when every named part is fixed. Completed batches are removed; their public behavior changes are in the 1.0 release notes.

Work from top to bottom within each priority. Independent batches can proceed separately. A batch with an entry in the Decision column needs its compatibility or configuration choice resolved before changing the affected behavior. Its reproduction, tests, and design can proceed while that choice is pending. The related larger designs are recorded in ROADMAP.md.

No open remediation batches. B33–B37 are complete; their behavior changes and migration guidance are recorded in the 1.0 release notes.

Execution rules

  1. Confirm each finding against the current code and add a regression test for its observable behavior. Preserve its ID until all portions of a multi-package finding are complete.
  2. For rows with a Decision, record the selected behavior and any migration note before implementation. An open compatibility decision must be resolved explicitly; this plan does not grant approval.
  3. Keep each code change reviewable. A multi-package batch may be delivered as linked package-scoped changes, with the same batch ID and a single final completion check. Update the codemap when entry points, ownership, invariants, configuration, or dependency directions change.
  4. Run targeted package tests and affected consumers. For shared-state fixes, use a test with overlapping reads and writes and run make test RACE=true with Docker available for the Redis-backed tests. Benchmark performance fixes against the same scenario before and after; run race checks separately from timing.
  5. Obtain an independent code review and address its comments; record the review in PR_REVIEW.md. Run make testshort and make lint; run the full make test suite with Docker for the final integrated change. Keep the total coverage above the 90% CI gate (MIN_TESTCOV) and every package with statements above 90%. Record any unavailable fixture or incomplete check.
  6. When a batch is complete, remove its fixed findings from FINDINGS.md, remove the batch from this queue, and note public behavior changes in the next release notes. If only part of a finding is fixed, leave the ID and record the remaining portion here. Never reuse an ID.