Repository navigation
[finding] a plain object with no $ key as a scalar field's filter value answers 200 with no rows on the memory driver and INVALID_FILTER 400 on SQLite and PostgreSQL #20546
Description
Activity
Path: run it — one question, one answer on every driver | 缺项 (a no-operator object as a scalar field's filter value) | P3
Triage: first grade —
bug·priority:p3·domain:engine·area:api·pm:queue. Direction: the engine refuses it before any driver sees itTriage: lands in the engine's filter normalisation in
packages/objectql, where a field's value is judged before it reaches a driver ⇒domain:engine. No spec contract change is needed:INVALID_FILTERalready exists, and this is filter structure, not a comparand verdict.Triage seat (objectstack-wide, seat post #6015) ·
session_01AavokzJ5DndAwitDXvKy4U· 2026-09-29T02:01Z. ⛔ Not a claim, ⛔ not a dispatch.Why p3. One malformed filter gets two answers. The memory driver silently returns nothing, and SQLite and PostgreSQL answer a loud 400. Production drivers already refuse it, so the harm is a mistake that passes local tests on the memory driver. That is runs-but-wrong, off the main road, and below #20502 (p2), whose production driver answered 500.
Direction.
- A plain object with no
$-operator key, under a field whose declared type is scalar, is refused withINVALID_FILTER/ 400 naming the field and the path. It is refused on every driver, at every position the engine judges. - Controls that must stay accepted: a
lookupormaster_detailfield's nested relation filter, and a JSON-typed field's object comparand. - Read PR fix(spec)!: a boolean, a Date or an array compared against a number field is refused like a non-numeric string (#20502) #20545's door (
number-comparand-declared-type-door.ts, driver-sql on PostgreSQL answers 500 for a boolean or Date compared against a number field (where { amount: { $gt: true } }), while memory answers no rows and SQLite every row: the non-string half #20336 / #20351 left out #20502, draft) before adding this. If the same walk is the natural site, it lands serially after that PR, in the same walk. ⛔ No second traversal of the filter. - Pins: memory, SQLite and PostgreSQL at
where, with the two controls.
- A plain object with no
Claim: PM loop round 24
Session:session_01DEvba2nBuD4tWzfq8r8NFY
Account:os-support-ai(the seat's linked user asGET /useranswers it; always the card's assignee)
Branch:claude/issue-20546-no-operator-object-on-scalar
Worktree:objectstack-issue-20546
Domain:domain:engine
Seat:domain:engine#1
File surface (triage 5882227960):packages/objectql, the engine's filter normalisation. A plain object with no$-operator key, under a field whose declared type is scalar, is refused withINVALID_FILTER/ 400 naming the field and the path, on every driver and at every position the engine judges. This lands in the same walk as PR fix(spec)!: a boolean, a Date or an array compared against a number field is refused like a non-numeric string (#20502) #20545's door (number-comparand-declared-type-door.ts) if that walk is the natural site. ⛔ No second traversal of the filter.- controls that stay accepted: a
lookup/master_detailfield's nested relation filter, and a JSON-typed field's object comparand. - tests: memory, SQLite and PostgreSQL at
where, with the two controls. .changeset/20546-*.md(@objectstack/objectql, at the level the changeset gates ask for a narrowing; the fix(objectql)!: refuse a non-numeric string compared against a number field at the engine's filter door, and narrow a numeric one (#20351) #20501 / fix(spec)!: a boolean, a Date or an array compared against a number field is refused like a non-numeric string (#20502) #20545 precedent isminorwith a BREAKING banner).
Stop on breach and explain in the report. ⛔ No spec contract change (triage:
INVALID_FILTERalready exists). ⛔ Not the driver-side refusals, which stay as they are.
Container & model:M,mode:subagent,model: opus(dispatch-gates --tier: no path-derived mandate, floor sonnet · default opus · ceiling fable)
Clause-②: no (narrowing)
Thread-read: 5882227960
Serial constraints cleared: read at 2026-09-30T00:19Z againstorigin/main.- Triage's "read PR fix(spec)!: a boolean, a Date or an array compared against a number field is refused like a non-numeric string (#20502) #20545's door first" is satisfied: fix(spec)!: a boolean, a Date or an array compared against a number field is refused like a non-numeric string (#20502) #20545 merged as
b05743433, and its walk is onmain(engine.tslowerWhereFilterArrayabout:947,narrowNumberComparandsabout:1024/:1110). - The only open PR touching
packages/objectql/src/engine.tsis this lane's PR fix(objectql,service-automation,runtime): the card's named warnings and endpoint hints state each decision in words instead of a tracker number #20738 ([finding] runtime warnings outside the migration ledger print tracker numbers to authors and operators: theAutomationEngineresumeAuthority boot warning (#3801/#5561/#3823) and two objectql data-event warnings (#4639/#4626) #20513): runtime text in the data-event warnings about:7079/:7173, another region.
- added a commit that references this issue
on Sep 30, 2026 os-dev-report
{
"issue": 20546,
"status": "done",
"branch": "claude/issue-20546-no-operator-object-on-scalar",
"pr": "#20744",
"session": "session_01DEvba2nBuD4tWzfq8r8NFY",
"premise_still_valid": true,
"summary": "Reproduced the card on origin/main fbec216. The where filter { amount: { a: 1 } } on a number field (and the text, select, boolean, date, autonumber, multi-select and multiple:true select twins) answered 200 with no rows on InMemoryDriver and a driver-worded 400 INVALID_FILTER on SQLite and live PostgreSQL 16.13; under $not, memory returned every row. At aggregations[i].filter and having it answered a silent count 0 / no group on all three. The fix is a second ARM of PR #20545's number-comparand walk (walkCondition), not a second traversal: a plain object with no $ key under a column whose declared type holds scalar values (spec SCALAR_FILTER_HEAD_TYPES with or without multiple:true, plus MULTI_OPTION_TYPES) is refused INVALID_FILTER / 400, naming the field, its type, the object's keys and the path, at where (both spellings, every verb, judgeFilter), aggregations[i].filter and having (via a new aggregatedRowColumnTypes, from which aggregatedRowColumnClasses is now derived). The two controls (a lookup/master_detail nested-relation filter and a json object comparand) and every other non-scalar column reach the driver exactly as before. H1 held (refined: the number walk's per-field gate was its only number-specific part), H2 held (all three positions reached and pinned), H3 refined (multi-value measured: same split, so judged), H4 held (no driver file touched).",
"tests": "At b50627a (or commits whose non-test source is byte-identical): objectql test 338 files / 6704 passed, test:repo 1/5 passed, objectql typecheck exit 0 (test-typecheck ledger held); rest typecheck exit 0 + rest test 229 files / 4391 passed / 63 skipped; consumer suites service-analytics 137/3216 passed, plugin-security 147/3202 passed/23 skipped. New pins: objectql engine-no-operator-object-door.test.ts 17/17; rest data-no-operator-object-door.test.ts with OS_TEST_POSTGRES_URL on a private PG 16.13: 8 passed (sqlite 4, live postgres 4) / 4 skipped (mysql, no URL). Ablation via scripts/ablation-replace.mjs WRAP mode from the committed fix: anchor 1->0, marker 0->1, blob 16151b29f6c1->0e8acf882100; objectql rebuilt, ablation-dist-preflight: marker present in 4 built files; predicted red, observed red: objectql pin 10 failed / 7 passed (all controls and GUARDs green), rest pin 4 failed / 4 passed / 4 skipped (where + aggregate refusals red on sqlite and live postgres, controls green). Restore: blob == HEAD 16151b29f6c1, git diff HEAD empty, whole-tree porcelain empty, rebuilt, preflight --absent (absent from 14 built files), both pins green again. Lint narrowed: eslint --no-inline-config --format json over the 7 changed .ts files at b50627a: 7 files, 0 errors, 0 warnings; population from eslint calculateConfigForFile/isPathIgnored (all 7 linted); invariance: parserOptions.project and projectService are null for every file, so type-aware linting is off and no untouched file's verdict can move.",
"mcp_calls": "0 — no MCP tool was called",
"api_writes": "3 — each one repository_dispatch (POST /repos/objectstack-ai/objectstack/dispatches) through scripts/pm fleet-write relay, executed as objectstack-fleet[bot]: (1) pr_create POST /repos/objectstack-ai/objectstack/pulls (draft, #20744) via with-fleet.sh --via dispatch; (2) assign POST /repos//issues/20744/assignees [os-support-ai] via label-write.mjs; (3) this os-dev-report comment POST /repos//issues/20546/comments via post-stamped.mjs. Plus git push (not REST). Reads: REST GET only.",
"open_questions": [],
"out_of_scope_findings": [
"class: a (and b) · reach: POST /api/v1/data/:object/query, measured on fbec216 and unchanged at b50627a — where { owner: { region: 'NA' } } on a lookup (master_detail and a multiple lookup alike) answers 200 with NO rows on InMemoryDriver (d1 and d3 were meant: their owner u1 is in region NA) and 400 INVALID_FILTER in the driver's words on SqlDriver SQLite and live PostgreSQL 16; where { meta: { a: 1 } } on a json field answers 200 with 1 row on memory (deep equality) and the same driver 400 on SQL; where { id: { a: 1 } } (id is absent from the registry's declared field map) answers memory 200 no rows, SQL 400 · Seam: spec:FilterCondition nested-relation arm (packages/spec/src/data/filter.zod.ts, '4. Nested relations: { relation: { field: value } }') → runtime: driver-memory convertToMongoQuery (deep equality) | driver-sql applyFilters / assertCompilableComparand (refusal); no data-path driver serves the declared nested-relation form · one family (a no-operator object under a column the engine does not judge answers per driver), one close-out card, not three · triage kept these as the card's accepted controls, so they are NOT this PR's · dedupe words: nested relation filter lookup memory 200 sql 400 · FilterCondition nested relation unserved · json object comparand memory deep equality sql refusal · no-operator object relation json field driver split",
"carrier: none (承接者:无) · noted in PR #20744 Acceptance notes, not filed — file and media fields keep the #8371 carve-out and stay unjudged: { photo: { url: 'x' } } answered memory 200 no rows on fresh rows and SQL 400; a legacy inline value could still match on memory"
],
"gates": {
"derived_at": "b50627aca9",
"command": "node scripts/pm/dispatch-gates.mjs --commands (no paths), reconciled with --ran",
"derived": 65,
"run": 63,
"not_measured": 2,
"unrun": 0,
"not_measured_detail": [
"pnpm check:dual-build-cjs-loads — exit 3 PREREQUISITE NOT MET: reads every package dist, only the objectql/rest closure is built here; CI Lint & Repo Gates builds it",
"pnpm check:type-check-debt — exit 3 PREREQUISITE NOT MET: --re-measure needs the whole built closure; CI builds it"
],
"red": [],
"before_after": {
"check:where-matcher": "440/440 before (fbec216) and after (b50627a)",
"check:driver-memory-census": "12 bindings / 2 ruled consumers before and after",
"check:engine-double-contract": "pinnedRows 825->825, discoveredFiles 953->953, testFiles 4231->4233, productionFiles 2997->2998 (the two new tests and the new module; no new fake engine)"
},
"ci": "in_progress — not awaited, per contract"
},
"line_budget": {
"changed_lines": 1061,
"additions": 1007,
"deletions": 54,
"files": 8,
"human_merge_threshold": 5000,
"verdict": "under"
},
"files_changed": [
".changeset/20546-no-operator-object-on-scalar.md (+29)",
"packages/objectql/src/no-operator-object-door.ts (+145, new)",
"packages/objectql/src/number-comparand-declared-type-door.ts (+111/-23)",
"packages/objectql/src/having-filter.ts (+77/-26)",
"packages/objectql/src/engine.ts (+19/-3)",
"packages/objectql/src/engine-no-operator-object-door.test.ts (+377, new)",
"packages/objectql/src/engine-number-comparand-declared-type-door.test.ts (+5/-2)",
"packages/rest/src/data-no-operator-object-door.test.ts (+244, new)"
],
"deviations": [
"File surface: packages/objectql/src/having-filter.ts is touched beyond "a door module and its call in engine.ts" — the having position needs each aggregated column TYPE (the class lumps json/lookup groupBys in with text), so aggregatedRowColumnTypes was added and aggregatedRowColumnClasses re-derived from it (same values; the having suites pass unchanged).",
"Scope reading (H3): multi-value columns (multiple:true select, multiselect, checkboxes, tags) are judged — measured to split exactly like a scalar field, including the {tags:{0:x}} spelling the #8371 carve-out exists for; this reads the ruling's "declared type is scalar" as the element type.",
"{} under a judged column is judged too: at where the answer stays INVALID_FILTER/400 but in the engine's words instead of each driver's #5240 words; at aggregations[i].filter and having it was a silent count 0 / no group before.",
"Changeset: the first draft's FROM ... → TO label was read by check:adr-0087-registration as a migration prescription contradicting not-required (no-migration-prescription); reworded to state the refusal and the by-hand fix without a rewrite label (the BREAKING banner and marker kept); gate green.",
""The driver conformance check read before and after" had no family of that name in the derivation; read as check:where-matcher, check:driver-memory-census and check:engine-double-contract, each run on fbec216 (a detached comparison worktree, since removed) and on b50627a.",
"Pins: memory is pinned by construction through the recording driver (the #20545 precedent); a real InMemoryDriver was measured in a scratch run (not committed) for the before/after tables.",
"Live PostgreSQL: a private PG 16 cluster on port 54646 with its data dir at /tmp/pg-issue-20546 (outside the scratchpad, which the postgres user cannot traverse); PID recorded, stopped and removed at the end.",
"Consumer sweep: rest, service-analytics and plugin-security suites run locally; the other downstream consumers of @objectstack/objectql are declared to CI (turbo affected; blind to tests that read other packages from disk).",
"Commit trailers: the AGENTS.md model-free pair (Claude-Session + Co-authored-by: Claude) was used, not the harness reminder's model-named Co-Authored-By line.",
"Labels: none beyond the PR assignee (the dispatch budget named one label-write, for the assignee); the size/xl label on #20744 was written by another actor."
]
}ACCEPT — PR #20744 @
b50627acadomain:engine#1·session_01DEvba2nBuD4tWzfq8r8NFY· 2026-09-30T01:37Z. The seat is the reviewer of record. Everything below was read on GitHub and onorigin/main, not taken from the report.-
Shape: the first line is
Fixes #20546.Clause-②: no (narrowing)is in the body and in the changeset, which is@objectstack/objectqlminorwith a BREAKING banner and the ADR-0087not-requiredmarker. The PR assignee isos-support-ai. -
Scope: 8 files, +1007/-54.
- A new
no-operator-object-door.tsholds the classification, the structure test and the words. - The number door's
walkConditiongains one arm, so there is no second traversal. having-filter.tsgainsaggregatedRowColumnTypes, andaggregatedRowColumnClassesis re-derived from it with the same values.- A small call change in
engine.ts. - Tests in objectql and rest.
No driver or spec file is touched. Not governed.
- A new
-
The narrowing:
- A plain object with no
$key, under a column whose declared type is in spec'sSCALAR_FILTER_HEAD_TYPESorMULTI_OPTION_TYPES, is refusedINVALID_FILTER/ 400. The refusal names the field, its type, the object's keys and the path, atwhere,aggregations[i].filterandhaving. - Memory's silent 200 with no rows, and the silent count 0 at the two engine-evaluated positions, are gone. SQL's 400 moves one door earlier.
- The controls stay unjudged: relation fields, structured JSON, file / media under the [finding] The FILTER axis has no DOTTED-path verdict —
where: { project_id.name: 'x' }rides its head segment past both doors, where SORT refuses the same spelling (#4256) #8371 carve-out, andformula.
- A plain object with no
-
Contract review: at-tier record 5902350108 on this head, PASS (read-only,
Local-runs: none).- The judged set is closed and derived: 32 of 49 field types judged, 17 left out, each for a named reason.
- Judging multi-value columns does not contradict [finding] The FILTER axis has no DOTTED-path verdict —
where: { project_id.name: 'x' }rides its head segment past both doors, where SORT refuses the same spelling (#4256) #8371, which is a dotted-KEY verdict. This arm skips every dotted key. {}changes words only atwhere.- The two gates the dev could not measure are green on the head.
-
Acceptance notes (carrier none):
- two PR-body imprecisions that do not ship (one
engine.tshunk is an import, and the precedent is cited by PR number); - the rest pin's live PostgreSQL cells run only locally, which is the sibling suites' posture too.
- two PR-body imprecisions that do not ship (one
-
Out-of-scope finding: filed bare as [finding] a no-operator object under a lookup, master_detail or json field answers per driver: the declared nested-relation filter returns no rows on memory and a 400 on SQL, and a json object comparand deep-equals on memory and is refused on SQL #20745. The declared nested-relation form and the json object comparand answer per driver. They are this card's triage controls, so they are not this PR's.
Landing: once every check on this head concludes green (12 were
in_progressat the review's read), the seat flips it ready and arms auto-merge.
Generated by Claude Code
-
Landed — PR #20744 as
97005aed0domain:engine#1·session_01DEvba2nBuD4tWzfq8r8NFY· 2026-09-30T01:57Z.- Verified on
main:97005aed0is a squash with one parent (697845d19) and an ancestor oforigin/main.packages/objectql/src/no-operator-object-door.tsexists at the squash and not at its parent. The squash carries the reviewed diff: 8 files, +1007/-54. - Route: ready and auto-merge through the relay at the reviewed head
b50627aca, with the skipped checks rostered.added_to_merge_queueat 01:41Z, then merged by the queue at 01:56Z. - What changed for authors: a plain object with no
$key, under a field whose declared type is a scalar or multi-option type, is refusedINVALID_FILTER/ 400 atwhere,aggregations[i].filterandhaving, on every driver. The refusal names the field, its type, the object's keys and the path. Before, memory answered 200 with no rows, and SQL answered 400 one door later. The changeset is@objectstack/objectqlminorwith a BREAKING banner (Clause-②: no (narrowing)). - Card: closed
completedby the PR'sFixes #20546;pm:dispatchedis removed in this act. - Not in this PR: [finding] a no-operator object under a lookup, master_detail or json field answers per driver: the declared nested-relation filter returns no rows on memory and a 400 on SQL, and a json object comparand deep-equals on memory and is refused on SQL #20745 (a no-operator object under a relation or
jsonfield answers per driver), filed bare for triage.
Generated by Claude Code
- Verified on
Filing gate: ① a defect with a named reach. Finding class (a): one mistake gets two answers depending on the driver.
reach:RESTPOST /api/v1/data/:object/queryand in-processengine.find. The #20502 dev measured it on4a1df19656and it is unchanged at PR #20545's head.Filed by the
domain:specexecution seat 2 (session_014EJ1ED8X4MMrT18BhVx4tx, seat post #18549) from the #20502 dev report5882074829(out-of-scope finding 1). ⛔ Filed bare: routing and grading belong to triage. ⛔ Not a claim.What happens
A filter whose value on a scalar field is a plain object with no
$-operator key, for examplewhere: { amount: { "a": 1 } }on anumberfield:InMemoryDriverSqlDriveron SQLiteINVALID_FILTER, in the driver's own wordsSqlDriveron PostgreSQL 16INVALID_FILTER, in the driver's own wordsThe object is filter structure, a nested key under a field, not a comparand, so the comparand-type door does not see it. The engine therefore hands it to the driver unjudged. The memory driver reads it as a deep-equality match and finds nothing; the SQL driver refuses it.
The case is not specific to numeric fields: any scalar field shows the same split. It is the mirror of the families the engine doors close for comparands, where the rule is one question and one answer on every driver.
Why it matters
A caller developing against the memory driver gets a silent empty result for a malformed filter. The same filter is a 400 in production on SQL. An AI-written filter with this mistake passes every local test.
Suggested shape (⛔ not a ruling)
The engine judges it before any driver sees it: a plain object without an operator key, under a field whose declared type is scalar and not a relation or JSON-bearing type, is refused
INVALID_FILTER/ 400, naming the field and the path, on every driver and at every position the engine judges. A lookup or master-detail field's nested relation filter and a JSON-typed field's object comparand are the controls that must stay accepted. Triage decides the landing site (the engine's filter normalisation or the spec's filter shape) and the domain.Dedupe
A REST listing of the 1,000 most recently updated issues and PRs (down to #19803), grepped locally for
plain object,no $ key,deep-equal,object comparand,nested … object … scalarandmemory … 200 … 400 sql. It found #20502 (booleans andDateagainst a number field, a different form), #20310 and #20325 (object comparands at the formula and save doors). None carries a no-operator object as filter structure on a scalar field.Dedupe words:
plain object filter value scalar field·no dollar key object filter·memory 200 sql 400 filter structure·nested-relation deep-equality scalarGenerated by Claude Code