Repository navigation
fix(ci): record the HTTPServer bind site that merged before the ratchet - #17906
iamwhatever wants to merge 1 commit into
Conversation
The ratchet counted 15 sites on a base without the opencode MCP-deny test's loopback fake, which merged first; the tree has 16.
First Principles Review (Opus 5.5) — ⏭️ skippedRevision |
UX Review (Opus 5.5) — ⏭️ skippedRevision |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing to check. [DESIGN-REVIEWED] ca41e1d Description read[DESCRIPTION-READ] 46543ca00393da79d8981ce00062fcda020aaa5f21138d27316dd2a8499b92d7 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
|
Main has two reds:
We are backing #17909, which carries both fixes and keeps the ratchet at 15. This PR stays open until #17909 merges. |
Problem / Motivation
Goal: Turn main's Backend Tests green again by recording the HTTPServer bind count the tree actually has.
test/test_http_server_bind_ratchet.py::test_bare_http_server_constructions_match_the_recorded_countfails on main: 16 sites, recorded 15.Why it matters
The red lands on every PR based on main (for example #17875 and #17881).
Root cause
A merge race between two PRs that both merged on 2026-10-07:
896b87c6c(#16728, per-tool MCP deny for opencode)test/test_per_tool_mcp_deny_single_binary.py:627ThreadingHTTPServer(("127.0.0.1", 0), Handler)781294256(#17450, docs audit)_BASELINE_SITES = 15, counted on a base without #16728git log -Son each of the 16 scanned sites dates all of them; #16728's site is the only one newer than #17450's base.What changed
test/test_http_server_bind_ratchet.py_BASELINE_SITES15 → 16Reason the 16th site is recorded rather than converted: it is a loopback (
127.0.0.1, port 0) fake OpenAI endpoint in a test, the same shape as the 13 other test sites already counted. The tree has no shared test helper that skipsgetfqdn(the twoserver_bindoverrides, inplumb_cpu.pyandtest_decisions_local_runtime_real_spawn.py, are private to their files), so there is nothing to reuse; converting it would mean a new helper, which is out of scope for a main-red fix. The ratchet still fails on a 17th site.Not a goal
test/test_session_control_doc.py::test_every_tabulated_refusal_code_is_one_the_source_raises, caused by docs: 2026-10 full audit -- docs, skills, rules and CI guards #17450 tabulatingremote_target_unsupportedafterc6f9fed4a(feat(remote-crew): delete the turn relay; old relay chats are read-only #17662) removed it). That fix is already open as docs(session-control): drop refusal row removed in #17662 #17903 and is not duplicated here. Until docs(session-control): drop refusal row removed in #17662 #17903 merges, this PR's Backend Tests still show that one main-side red.Backwards compatibility
Compatible: test-only constant.
Tests
python3 -m pytest -q -p no:cacheprovider test/test_http_server_bind_ratchet.py test/test_session_control_doc.py: ratchet tests pass; the one remaining failure is the doc test fixed by #17903.Manual verification
N/A — test-only change.
Related Issues
Part of #17900
Pattern harvest
Not generalizable here: an exact-count ratchet added by a long-lived PR can miss a site that merges in between. The ratchet's own message already names the fix.
Checklist