The MongoDB lookup-index divergence is pinned only in an opt-in suite that runs on no CI lane — the pin cannot fail, in either direction

[os-zhuang]

Reported by the #12252 dev as a judgement call beyond its instruction, and filed here by the domain:engine PM seat. Small, mechanical, and independent of #13222's disposition.

The gap

PR #13224 inverts the assertion in packages/drivers/driver-mongodb/src/mongodb-driver.test.ts to pin the divergence #13222 owns: a canonically-spelled lookup gets no idx_<field>_lookup on MongoDB.

⚠️ That suite is describe.skipIf(!sharedMongod) — gated on OS_TEST_MONGODB_MEMORY_SERVER_ENABLED=1, which is opt-in because of the ~123 MB binary download #5517 retired from default runs. It runs on no ordinary CI lane.

So the pin is inert in both directions:

• if driver-mongodb indexes lookup joins off reference_to — a key the spec REFUSES — so no authored lookup field has ever been indexed on MongoDB #13222 is fixed and lookups start getting indexed, nothing goes red — the pin's own "flips back to toContain deliberately" instruction has no mechanism to remind anyone;
• if the divergence widens, nothing goes red either.

A pin that cannot fail records a fact for human readers but enforces nothing. That is materially weaker than what the inverted-in-place convention assumes: its whole justification is that from main, a pin's disappearance and the divergence silently reopening must not look identical — which only holds if something runs it.

The fix

packages/drivers/driver-mongodb/src/mongodb-schema-declared-indexes.test.ts tests the same function, is driven by a fake Db recorder with no server at all, and does run on every lane. Its own header states the function under test runs verbatim.

Add the equivalent assertion there: a { type: 'lookup', reference: '…' } field produces no idx_<field>_lookup, with the type: 'user' positive control alongside it so the day the arm is removed entirely, the case fails loudly rather than passing vacuously.

⚠️ The positive control is the load-bearing half. Without it, not.toContain passes just as happily against a function that produces no indexes at all, a renamed index, or a harness that never called the function — the exact vacuity that makes a negative assertion worthless.

Cross-reference the two pins so a future editor of either finds the other.

Scope

One test file, additive. ⛔ Do not change the driver — that is #13222. ⛔ Do not un-skip or re-enable the opt-in suite; #5517's merge-queue hazard is the reason it is opt-in and this card does not reopen it.

Why it is worth its own card

It holds whichever way #13222 goes, and it is what makes #13222's eventual outcome observable: if the lookup arm is ever taught to read reference, this is the assertion that turns red and tells someone to retire the divergence note rather than leaving a stale "disposition open" comment behind.

Related

#13224 (the PR that creates the pin) · #13222 (owns the divergence) · #12252 · #5517 (why the covering suite is opt-in)

[claude]

Claim — dev seat starting on #13225.

• Session ID: session_01LZbWd2jNV1FErXTPSS4Dry
• Branch: claude/issue-13225-lookup-index-fake-db-pin
• Worktree: ../objectstack-13225 off origin/main

Scope as dispatched: additive assertion in packages/drivers/driver-mongodb/src/mongodb-schema-declared-indexes.test.ts only. Driver untouched (#13222 owns it); mongodb-driver.test.ts untouched (#13224 owns it); the opt-in real-mongod gating is not reopened (#5517).

---

Generated by Claude Code

[claude]

os-dev-report

{
 "issue": 13225,
 "status": "done",
 "branch": "claude/issue-13225-lookup-index-fake-db-pin",
 "pr": "https://github.com/objectstack-ai/objectstack/pull/13238",
 "premise_still_valid": true,
 "summary": "Added one additive case to packages/drivers/driver-mongodb/src/mongodb-schema-declared-indexes.test.ts, the fake-Db suite that runs on every lane, pinning that a canonically-spelled { type: 'lookup', reference: '...' } field produces NO idx_company_id_lookup, alongside a type:'user' positive control that DOES get idx_owner_id_lookup, plus a toEqual over the exact recorded index set. Premise verified on origin/main before writing: mongodb-schema.ts line 112-115 gates the join-index arm on (field.type === 'lookup' && field.reference_to) || field.type === 'user', so the canonical 'reference' key falls through and the 'user' disjunct is unconditional. The comment records the divergence as OPEN and owned by #13222, says the assertion is expected to flip to toContain deliberately when that lands, and cross-references the twin pin in mongodb-driver.test.ts (which #13224 owns and I did not touch). Driver untouched, opt-in real-mongod gating untouched, one file, +75 lines. No changeset: test-only, publishes nothing, so the repo's skip-changeset label is the route -- applied additively via REST POST and read back (labels now ['skip-changeset']).",
 "tests": "ALL READINGS ON THE FINAL COMMIT e15a908e8; every exit code captured before any pipe; verdicts quoted from each tool's own printed line. (1) PACKAGE SUITE, deps built first with pnpm --filter '@objectstack/driver-mongodb^...' build, under scripts/pm/os-verify-lock.sh: 'Test Files 21 passed | 5 skipped (26) / Tests 488 passed | 143 skipped (631)' and 'os-verify-lock: VERDICT command-exit 0'. The 5 skipped files are the opt-in real-mongod suites, untouched. (2) REPO-WIDE ESLINT, the whole farm rather than a narrowing: pnpm lint = 'node --stack-size=4000 node_modules/eslint/bin/eslint.js . --no-inline-config', LINT_EXIT=0, no output. (3) TYPECHECK -- GREEN BUT NOT MEASURED OVER MY FILE, reported not claimed: pnpm --filter @objectstack/driver-mongodb typecheck -> TYPECHECK_EXIT=0, but that package's tsconfig.json carries exclude ['node_modules','dist','**/*.test.ts'], so tsc never reads the edited file. Counted with --listFiles rather than assumed: 0 hits for mongodb-schema-declared-indexes.test.ts, 1 hit for the non-test sibling /src/mongodb-schema.ts as control, 387 files in the program. Consequence handled in the code: the lookup field literal is bound through a variable, because the driver's FieldDef declares only reference_to and a fresh literal carrying 'reference' would trip the excess-property check on the very key the case exists to record the driver does not read; a comment says so at the binding. (4) GATE FAMILY re-derived on the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack (it asserted the remote and named commit e15a908e8): 22 families run, all exit=0 -- cross-package-test-inputs, driver-conformance, logger-receiver-detach, objectql-double-limit, page-declaration-shape, published-files, slot-lookup, test-source-alias, type-source-resolution, check-ci-filter-parity, check-comment-mask-adoption, check-cross-package-test-inputs, check-keyed-text-bounds, check-plugin-teardown-shape, check-shard-attestation, check-undeclared-dep-imports, query-options-erasure, type-check-coverage, engine-double-contract, where-matcher, nul-bytes. TWO exit=3 results are NOT MEASURED, not red, in the gates' own words: check:dual-build-cjs-loads printed 'PREREQUISITE NOT MET -- this gate reads built output, and some package has no dist/. ... This is NOT a pass: nothing was measured' (wants a full pnpm build; CI's), and check-test-completeness.mjs printed 'the local reading for this gate is NOT MEASURED. It is not a red, and there is nothing here to fix' (grades a saved turbo run test log only CI produces). check:type-check-debt --re-measure left to CI deliberately: it refuses on an unbuilt worktree and that refusal is NOT MEASURED. Also ran grep -naP over the edited file for control bytes: no matches. (5) NON-VACUITY -- THE ASSERTION WAS SHOWN FAILING THREE WAYS. Implementation was COMMITTED FIRST (e15a908e8) so the restore leg had a real reference point; each mutation ran under trap RESTORE_FN EXIT INT TERM with ABSOLUTE paths; each was PROVEN ON DISK by grep counts on the anchor text (never by the editor's exit code); each restore was proven by comparing git hash-object against the HEAD blob AND by git diff HEAD being empty (never by the checkout's exit code), and restores named HEAD explicitly rather than using a bare git checkout -- path. No build/dist is involved: the test imports './mongodb-schema.js' relatively, so vitest resolves package source directly and there is no dist leg to poison. Baseline: [BASELINE] vitest exit = 0, 'Test Files 1 passed (1) / Tests 7 passed (7)', new case listed green in --reporter=verbose. M1, flip the new not.toContain to toContain in the test (on-disk proof: injected=1 removed=0): [M1] vitest exit = 1, 'AssertionError: expected [ idx_id_unique, ...(3) ] to include idx_company_id_lookup'. M2, ablate the driver's unconditional field.type === 'user' disjunct (on-disk proof: removed_user_disjunct=0 injected=1): [M2] vitest exit = 1, 'AssertionError: expected undefined to be defined' -- the POSITIVE CONTROL going red is what proves the case fails loudly if the arm is removed entirely rather than passing vacuously. M3, teach the arm to read 'reference', simulating what #13222 would do (on-disk proof: injected=1 old_form=0): [M3] vitest exit = 1, 'AssertionError: expected [ idx_id_unique, ...(4) ] to not include idx_company_id_lookup' -- the NEGATIVE going red is what proves the pin fires the day the divergence closes, which is the entire point of the card. All three restored: RESTORE-OK lines for both files with blob hashes a426585d4e23604f9ae98c0338b64a5a0e45e97f (test) and 7373499ec6a4ab686995bc3ed38b3ddaca13a67b (schema) matching HEAD, git diff HEAD empty; final re-run on the restored tree [RESTORED] vitest exit = 0, 7 passed (7); git status empty. mongodb-schema.ts does NOT appear in the PR diff.",
 "mcp_calls": "1 -- a single mcp__github__issue_read (get) on #13225. Everything else (whoami, comment read, assign, claim comment, PR-13224 metadata, PR create, label add, all read-backs) went through repo-scoped REST, which probed HTTP 200 from this container on the first call.",
 "open_questions": [],
 "out_of_scope_findings": []
}

---

Generated by Claude Code