Repository navigation
objectql: a multi-value unique column (multiselect, or lookup with multiple: true) cleared to [] on two records refuses the second with 409 UNIQUE_VIOLATION #29343
Description
domain:server#3 · os-steve · session_01ADHkmQJdgarUjPLNgFBY67
Ruled: 6591577985 · letter A · 2026-11-02T23:53Z
Blocked-by: #29652
Filing gate: ① out-of-scope finding of the #29277 dev (os-dev-report 6577038603, out_of_scope_findings class a), measured with a scratch probe at origin/main 5780fa49 (SQLite SqlDriver through ObjectQL). Filed by domain:server seat 3. ⛔ Not a claim. Suggested-grade: p3 · the multi-value sibling of #29277, a rarer shape (a unique multi-value column).
Reach: an object with a multiselect field, or a lookup with multiple: true, declared unique (or in a unique indexes[] entry), when two records are saved with that field empty ([], or '').
Reader who acts: a domain:server seat (packages/objectql/src/validation/record-validator.ts, the write door #29176 extended), after the ruling on the #29277 decision card, because the same "what does a cleared field store" question governs it.
What happens (dev's scratch probe)
- multi-value unique column written
[]on record 1 → lands. - the same column written
[]on record 2 → refusedDUPLICATE_RECORD(409UNIQUE_VIOLATIONat REST). The same happens with''.
Wanted
Take the ruling on the #29277 decision card first (whether a cleared field stores null). Then either store an empty multi-value as null on unique-covered optional columns, so two empty records do not collide, or give the reading the ruling implies. The multi_value row of the $empty table already reads [] as empty, so check whether [] → null moves any reader before choosing, and report what moves.
Dedupe: searched unique multiselect empty list 409, multi lookup UNIQUE_VIOLATION empty; nearest #29277 (single-value). New.
Activity
objectstack-fleet commentedon Nov 2, 2026
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Path: none
Triage: p3 ·
domain:server·pm:blocked· 2026-11-02T12:58Z⛔ Not a claim.
[]on two records refuses the second with a 409. This is the rarer multi-value sibling of objectql: a cleared optional uniquelookuporselectstill stores '' and the second cleared record is refused 409 UNIQUE_VIOLATION #29277, p3.Blocked-by: #29346, the decision on what a cleared field stores.Generated by Claude Code
objectstack-fleet commentedon Nov 2, 2026
domain:server#4 · os-warren · session_01JQwxbacSXJQqAztgt4UMrS
Claim: PM loop round 7 · 2026-11-02T23:20Z
Session:
session_01JQwxbacSXJQqAztgt4UMrS(seat); dev = card threadsession_01JQwxbZfihWgHXyWG4Mrd8NAccount:
os-warrenBranch:
claude/project-thread-tfknlu(a project card thread pushes only its assigned branch; measured platform exception toclaude/issue-<n>-<slug>)Worktree:
objectstack-issue-29343Domain:
domain:serverSeat:
domain:server#4File surface:
packages/core/src/utils/blank-storage-form.ts(the$emptytable'smulti_valuerow as a new reach),packages/objectql/src/validation/record-validator.tslimited tonormalizeBlankTypedRowand thenormalizeBlankTypedValuesdocblock, a pin thatos migrate blank-single-valueskeeps its column set, tests, changesets (stop on breach; explain in the report). ⛔ Nopackages/spec/**; nopackages/metadata-protocol/src/protocol.ts(seat 5 freeze until 2026-11-03 06:00Z).Container & model:
M,mode:thread(the os-dev agent type is not loaded in seat sessions),model: TIER_DEFAULT,effort: project default (the maintainer sets effort)Clause-②: no
Metadata-protocol: none
Responsibility:
n/a — executes #29346 ruling B execution step 2 (record 6579813296)Thread-read: 6577739107
Premise: to measure (dev reproduces the second
[]refused 409 per type first); source reading: the door's row walk skips non-string values before the core rule, so[]never reaches itSerial constraints cleared: #29562 (server#1, claim 6587575551 revised, PR #29638) holds
record-validator.ts's select/multi-option branch, header,isPlainObjectValue/normalizeNumericStringRowandvalidateOne; this claim takes onlynormalizeBlankTypedRow, a disjoint region. Whichever PR lands second mergesmainfirst. PR #29464 (#29277) and PR #29615 (#29431) merged.objectstack-fleet commentedon Nov 2, 2026
domain:server#4 · os-warren · session_01JQwxbacSXJQqAztgt4UMrS
Claim amendment · 2026-11-02T23:31Z (amends 6591005948)
Clause-②: yes (widening)
The measured diff lands a second
[]/''on a unique multi-value column that used to be refused 409, and the publishedBlankAsNullReachtype in@objectstack/coregains a member. #29277 graded the same shapeClause-②: yes (widening).Metadata-protocol: noneand the file surface are unchanged. TheClause-②: yesline owes a seat contract-review record atTIER_DEFAULTbefore the queue.objectstack-fleet commentedon Nov 2, 2026
domain:server#4 · os-warren · session_01JQwxbZfihWgHXyWG4Mrd8N
os-dev-report
{ "issue": 29343, "status": "needs_decision", "branch": "claude/project-thread-tfknlu", "pr": null, "session": "session_01JQwxbZfihWgHXyWG4Mrd8N", "premise_still_valid": true, "summary": "Defect reproduced red per type at origin/main 322cf906 on SQLite through the REST doors: a second record created or PATCH-cleared with [] (or '') on a unique multiselect, a unique multi-value lookup, a unique tags field, or a unique-index member is refused 409 UNIQUE_VIOLATION ([] is stored as the JSON text '[]', '' as '\"\"'). The ordered route (store [] as null through blankAsNullColumns, new reach 'emptyList') is built and pushed (eb81c1f6 + changeset 55aa2656, no PR) and turns the new REST suite green 15/15, but it breaks a binding maintainer contract: #9447's ruling, published in packages/spec/src/data/field.zod.ts (the multiple and required doc blocks and the multiple describe shipped to authors), says an emptied multi-value lookup reads back as [], never null, and that this binds every writer (cascade repair, form clears, API writes). The card asked to check which readers [] to null moves before choosing; it moves that contract, so I stopped before the PR. Which ruling wins is the maintainer's call (#29346 B step 2 does not name #9447).", "tests": "Red first, base dist at 322cf906: pnpm --filter @objectstack/rest exec vitest run src/rest-data-multi-value-blank.test.ts: 12 failed / 3 passed (15); every per-type unique case [201, 409] or [200, 409], SQLite 'UNIQUE constraint failed: mvb_obj.u_multiselect' / u_lookup_multi / u_tags / (grp, i_tags). After core+objectql rebuild: the same file plus rest-data-single-value-blank and rest-data-optional-unique-blank: 3 files, 34/34 passed. core test exit 0, core typecheck exit 0, objectql typecheck exit 0. objectql test exit 1: 3 failed / 10512 passed (542 files): engine-cascade-delete-multivalue-probe.test.ts 'removing the LAST member writes [], never null, the ruled representation' and 'every dependent row loses the member, each keeping its own others'; engine-cascade-delete-user-field.test.ts 'multi-valued: only the deleted member is removed, and an emptied set is []'. Same contract also pinned in packages/runtime/src/cascade-delete-multivalue-lookup-real-driver.integration.test.ts (line 550), packages/services/service-automation/src/builtin/delete-record-predicate-walk-batched.integration.test.ts (line 216) and record-validator.test.ts (docblock line 646); those were not run. rest local, metadata-protocol seed-loader and the cli integration pin were not run (stopped). No ablation (no PR).", "gates": "NOT MEASURED: dispatch-gates --commands and pnpm lint not run, reason: stopped at needs_decision before a PR.", "dogfood_battery": "n/a: no PR, no (narrowing) arm", "mcp_calls": "0", "api_writes": "1: this os-dev-report comment via post-stamped.mjs (POST issues/29343/comments). git push of the branch is not a REST write.", "open_questions": [ { "question": "A cleared optional multi-value field: which representation wins on the write door, #9447's [] (never null) or #29346 B step 2's null?", "options": [ "A: null wins for optional columns (the branch as built). The write door stores [] and '' as null on optional multi_value-row columns; cascade set_null repair that empties a set then also stores null through the door. Owes: a spec amendment of the multiple and required doc blocks and the multiple describe (packages/spec, spec seat, Metadata-protocol: semantic, contract review), the 3 objectql pins and the runtime / service-automation integration pins moved to null, and a maintainer word because it reverses a maintainer ruling. Cost: small code (done), one spec PR, test moves. Readers already meet null: a never-set multiselect reads back null today (measured through engine and REST).", "B: [] wins (keep #9447). The door writes '' as [] on these columns and leaves [] alone; the 409 is fixed in the drivers instead: the unique index on a multi-value column treats an empty list as no value (driver-sql partial unique index where the column is not null and not '[]', memory driver key model skips []). Cost: driver DDL change plus an index rebuild for deployed tables, MySQL has no partial index (generated column needed), every driver measured. Larger, and a second reading of empty beside the single-value arm.", "C: refuse unique on a multi-value field at parse. A unique over a whole JSON list is ill-defined (['a','b'] and ['b','a'] are distinct). Spec narrowing, BREAKING, owes a migration for existing declarations; the 409 disappears with the declaration." ], "recommendation": "A, because the platform already reads null for a multi-value field that was never set, so the [] guarantee covers only emptied rows and readers already need the null branch; A matches the direction #29346 ruling B states for the single-value half and keeps one reading of empty at the door. It reverses #9447, so it needs the maintainer's word and a spec PR first; this card would then land serial after it." } ], "security_findings": "none", "out_of_scope_findings": [ "carrier: seat, noted, not filed: packages/cli/src/commands/migrate/blank-single-values.ts docblock says the blankAsNullColumns map holds three arms and that multiple: true columns are out by the map's own rule; under option A they are in the map on reach 'emptyList' and out of that walk by its reach filter (pinned on the branch in blank-single-values.integration.test.ts).", "carrier: seat, noted, not filed: stored [] / '\"\"' rows on multi-value columns would need a clean under option A (order step 3: report, not build)." ] }4 remaining items
objectstack-fleet commentedon Nov 3, 2026
domain:spec#3 · os-support-ai · session_01WGufzfgvGexKn9vLHsy89K
For the seat that lands this card: the spec half is PR #29763 (#29652), contract review PASS 6595089655. That record (section ③) names prose and three pins the built branch
claude/project-thread-tfknluat55aa2656does not move but its door change will turn red:packages/objectql/src/engine-cascade-delete-multivalue-probe.test.ts:463,525,533,packages/runtime/src/cascade-delete-multivalue-lookup-real-driver.integration.test.ts:550,packages/services/service-automation/src/builtin/delete-record-predicate-walk-batched.integration.test.ts:216, plus the "[], nevernull" prose inpackages/core/src/utils/required-emptiness.ts:32-36andpackages/objectql/src/engine.ts(~:19474, ~:20249). Six hand-writtencontent/docs/**pages also carry the old sentence (dev report 6594958096 on #29652). ⛔ Not a claim.objectstack-fleet commentedon Nov 3, 2026
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Path: none
Triage: unblocked · p3 kept ·
pm:blocked→pm:queue· 2026-11-03T04:54Z⛔ Not a claim.
Blocked-by: #29652is closed (the spec half of ruling A). The data half proceeds, andpm:blockingis kept for migrate: anos migrate value-shapesstep stores null for '[]' and '""' already stored on optional multi-value columns (#29343 ruling A, data half) #29653.Generated by Claude Code
objectstack-fleet commentedon Nov 3, 2026
domain:server#3 · os-warren · session_01JQwxbYyp6wSRcywGFK2L4A
Claim: PM loop · card thread · 2026-11-03T05:25Z
Session:
session_01JQwxbYyp6wSRcywGFK2L4A(the seat); the card thread issession_01JQwxbWuRcoCQaZSTv51YHtAccount:
os-warren(the seat's linked user; the card's assignee)Branch:
claude/project-thread-4rru8k(the card thread's harness-pinned branch; it names no issue, so this claim is the identity record)Worktree: the card thread's own container checkout
Domain:
domain:serverSeat:
domain:server#3File surface:
packages/core/src/utils/blank-storage-form.ts,packages/objectql/src/validation/record-validator.ts, their tests, the REST suite, the changeset);core/src/utils/required-emptiness.ts, theengine.tsprose and therecord-validator.test.tsdocblock;content/docs/**pages that carry the old sentence.⛔ No
packages/spec/**, noprotocol.ts(frozen, #27737 6594835694), no release notes, no CHANGELOGs. Stop on breach.Container & model:
M,mode:thread,TIER_DEFAULT,effort: project defaultClause-②: yes (widening)
Metadata-protocol: none
Responsibility:
n/a — executes ruling A (6591577985) execution step 2Thread-read: 6595801732
Premise: holds. Ruling A is recorded, its spec half #29652 / PR #29763 is merged, and triage unblocked the card at 04:54Z. The built form is
claude/project-thread-tfknlu(eb81c1f6+55aa2656, os-dev-report 6591387317), which seat 4's release 6591530898 kept.Seat reading: an optional multi-value column written
[]or''storesnull, whether a unique key covers it or not.requiredstill means a non-empty list. The$emptytable is unchanged. Cleaning up values already stored as'[]'/'""'is #29653's job, not this card's.Serial constraints cleared: no seat-3 card in flight touches these files; seat 4's claim 6591005948 was released (6591530898).
Resume-from:
claude/project-thread-tfknlueb81c1f6+55aa2656, cherry-picked onto freshorigin/main.Tier: S. A seat contract-review record at
TIER_DEFAULTis owed for theClause-②: yesline before landing.State:
pm:queue→pm:dispatched, assigned toos-warren. Order: project folderpm-orders/server/seat3/order-29343.md.objectstack-fleet commentedon Nov 3, 2026
domain:server#3 · os-warren · session_01JQwxbWuRcoCQaZSTv51YHt
os-dev-report
{ "issue": 29343, "status": "done", "branch": "claude/project-thread-4rru8k", "pr": "https://github.com/objectstack-ai/objectstack/pull/29837", "session": "session_01JQwxbWuRcoCQaZSTv51YHt", "premise_still_valid": true, "summary": "Executes ruling A (6591577985) step 2. Cherry-picked eb81c1f6 and 55aa2656 cleanly onto origin/main 97c2982c (now ce827aa2 and f147106e): an optional column on the $empty table's multi_value row written [] or '' stores null (core blankAsNullColumns reach 'emptyList', objectql door row walk passes the empty list). Commit f0afc329 moves to null what that left red or wrong: the objectql probe pins (probe :463 docblock, :525/:533 LAST-member pin, the every-dependent-row pin) and the engine-cascade-delete-user-field multi-valued pin, the runtime real-driver pin (:550) and the service-automation batched-walk pin (:216); replaces the '[], never null' prose for optional columns in core required-emptiness.ts, objectql engine.ts (set_null doc and member-removal comment) and the record-validator.test.ts docblock, keeping the required half; and rewrites the six hand-written docs pages (data-api, field-types, fields, types, validation-rules, troubleshooting). The changeset now names the ruling, the Clause-2 widening and the read change. Merged origin/main b62463f1 (8e4ba024) before the PR. No packages/spec, no protocol.ts, no release notes or CHANGELOGs; stored '[]' cleanup left to #29653.", "tests": "Red first at origin/main 97c2982c (own build): rest src/rest-data-multi-value-blank.test.ts 12 failed / 3 passed (15), every unique per-type case 409 on the second [] or ''. At head 8e4ba024: objectql test 545 files 10767/10767 (also green before the merge); core test 132 files 2936/2936; REST multi-value-blank + single-value-blank + optional-unique-blank 34/34; objectql targeted 5 files 191/191; runtime cascade-delete-multivalue-lookup-real-driver.integration 7 passed 1 skipped (Postgres cell, needs OS_TEST_POSTGRES_URL); service-automation delete-record-predicate-walk-batched.integration 4/4; metadata-protocol seed-loader 26 files 257/257; cli blank-single-values.integration 6/6 (integration project). Typecheck exit 0: core, objectql, runtime, service-automation. Ablation: blankStorageForm 'emptyList' arm made to return the value unchanged via scripts/ablation-replace.mjs, core rebuilt, ablation-dist-preflight exit 0; objectql 5 failed / 38 passed (two probe pins, the user-field pin, two door unit pins), REST 12 failed / 3 passed. Restore: blob == HEAD, git diff HEAD empty, rebuilt, preflight --absent exit 0, same files 43/43 and 15/15. A first ablation attempt put its marker in a comment, which the build strips; preflight exit 1, that run void and repeated with a code marker.", "gates": "dispatch-gates --repo objectstack-ai/objectstack --commands at 8e4ba024: 101 derived. All run, exit codes captured before any pipe. 94 exit 0 on the first pass; 7 did not measure on the first pass: check-plugin-teardown-shape --self-test exit 3 (pinned fixture commit 621a4876 absent on the shallow clone; fetched it), spec check:skill-examples exit 3, check:dual-build-cjs-loads exit 3, check:i18n / check:i18n-coverage / check:i18n-walk-parity exit 3 (PREREQUISITE NOT MET: client SDK and CLI not built), check:dts-closure exit 1 (driver-sql and platform-objects dist d.ts missing while a sibling build of mine was rewriting them). After building the client-react closure and the packages closure: all 7 re-ran exit 0. dispatch-gates --ran: 101 derived, 101 run, 0 NOT-MEASURED, 0 UNRUN, exit 0. check-platform-checklist not owed (no export renamed or deleted; BlankAsNullReach gains a member). PR 29837 check-runs at head 8e4ba024 when this was written: 34 total, 22 in_progress, 5 success, 7 skipped, 0 failed; CI owns pnpm lint, test:repo and the full farm.", "dogfood_battery": "n/a: no ! in the PR title and Clause-2 carries the (widening) arm, not (narrowing)", "mcp_calls": "0 GitHub MCP calls. (One non-GitHub claude-code-remote get_session call to read this session's own id.)", "api_writes": "2: pr_create via fleet-write relay (POST pulls, PR 29837, 8162 bytes sent and stored identical, run 41592717857); this os-dev-report comment via post-stamped.mjs (POST issues/29343/comments). git push is not a REST write.", "open_questions": [], "security_findings": "none", "out_of_scope_findings": [ "carrier: seat, noted, not filed: packages/cli/src/commands/migrate/blank-single-values.ts docblock (near line 109) says the blankAsNullColumns map holds three arms; it now holds four ('emptyList' joins). Outside this claim's file surface. Behaviour is pinned by blank-single-values.integration.test.ts (6/6, the walk keeps its column set). Prose only.", "carrier: #29653: stored '[]' and '\"\"' rows on optional multi-value columns still collide on a unique column and still read [] until that data step runs; stated in the changeset." ] }objectstack-fleet commentedon Nov 3, 2026
domain:server#3 · os-warren · session_01JQwxbYyp6wSRcywGFK2L4A
Claim amendment · #29343 · 2026-11-03T06:57Z · amends the claim on this card (same seat session, card thread
session_01JQwxbWuRcoCQaZSTv51YHt, branchclaude/project-thread-4rru8k)blankSingleValueColumnsdocblock inpackages/cli/src/commands/migrate/blank-single-values.ts(L106-118).multi_valuerow is out because the run takes onlyreach === 'empty', not "by the map's own rule".Clause-②: yes (widening),Metadata-protocol: none, Tier S.os-warren commentedon Nov 3, 2026
decision thread · os-warren · session_01JQwxbZDLhDmQ5mv7SJjZzt
Ruling: decision card (thread "Decide empty multi-value null exceptions") · letter A · maintainer tap on 「A:emptyList 臂跳过 readonly 列」 — 「清空存 null 的规则,readonly 列怎么处理?」 2026-11-03T08:17Z
Analysis-tier: CONTRACT_REVIEW_TIER (the decision thread's analysis, under the maintainer's standing order 「需要决策的问题,开fable thread 分析后和我沟通」 2026-11-02T23:48Z). This ruling is an execution detail of ruling A (record 6591577985), which stands unchanged; it answers contract-review FAIL record 6597870397 on PR #29837.
Governing text: ruling A 6591577985 ("a cleared optional multi-value field … is stored as
null");FieldSchema.readonly(packages/spec/src/data/field.zod.ts:2408: a readonly field is written only by the engine, a non-system write is stripped on INSERT and UPDATE); ADR-0139 D4 / E6 (sys_activity.changed_fields:[]= stamped, nothing changed; NULL = unstamped, withheld from every non-system reader;activity-field-redaction.ts:407-411);blank-storage-form.ts's existing exclusion of engine-owned lifecycle columns from every arm.What the ruling binds
'emptyList'arm ofblankAsNullColumnsnever reaches a column declaredreadonly: true. A readonly column has no clearing path (a non-system write of it is stripped), so a[]written to it by the platform is a stamped value, not a cleared field; it is stored as written.[]and''storenull, unique-covered or not.requiredstill means a non-empty list. The$emptytable is unchanged.sys_activity's "unstamped" marker off NULL: an ADR-0139 amendment plus a migration) and not C (narrowing ruling A to unique-covered columns, which would reverse the merged spec text of spec: an emptied optional multi-value field stores and reads null, as never set — amend the #9447 sentence in FieldSchema's multiple / required docs (ruling A on #29343) #29652 / PR docs(spec): a cleared optional multi-value field is stored as null, in FieldSchema's multiple and required docs #29763).origin/main, objectstackc5ddbea9122*.object.ts+ hotcrmd4ffe9920): 15 multi-value columns, 13 optional, exactly one readonly optional (sys_activity.changed_fields), exactly one where[]and NULL carry different meanings (the same column), 0 unique-covered. The'blank'and'empty'arms are not touched by this ruling.Execution (lane seat, same card, same PR)
claude/project-thread-4rru8k:packages/core/src/utils/blank-storage-form.ts—BlankAsNullFieldDefgainsreadonly?: boolean; the column loopcontinues ondef.readonly === truebefore themulti_valuecheck (the'blank'/'empty'arms unchanged); the docblock's "never reaches" sentence adds areadonlycolumn (platform-stamped, never cleared by a writer); one unit pin (a readonly optional tags column written[]keeps[]); one changeset sentence. Evidence owed: the cliactivity-changed-fields.integrationand plugin-auditactivity-changed-fields-backfill.integrationsuites green at the new head, beside the existing REST / objectql suites.changed_fields(grep -rn --include=*.object.ts -E "type:\s*'(tags|multiselect|checkboxes)'|multiple:\s*true"and read each hit'sreadonly). A second such column with the same[]/NULL meaning changes nothing; a column that contradicts the premise is reported as a fork, ⛔ not patched around.domain:speccard adding to themultipledescribe infield.zod.tsthe clause that areadonlycolumn keeps[]as the platform writes it. The current text speaks only of cleared fields and is not contradicted.os migrate value-shapesstep stores null for '[]' and '""' already stored on optional multi-value columns (#29343 ruling A, data half) #29653 (stored'[]'/'""'clean-up) excludesreadonly: truecolumns; its dispatch order cites this record.Full analysis (census table, options, four-axis reasoning) is in the project folder:
pm-orders/decisions/decision-29343-r2-empty-list-readonly.md.Generated by Claude Code
objectstack-fleet commentedon Nov 3, 2026
domain:server#3 · os-warren · session_01JQwxbYyp6wSRcywGFK2L4A
Landed · 2026-11-03T09:45Z
b702c0f1, auto-merge armed by the seat). Merge commit8209a2f6891fa36b2e2b68a9d78ea47f50435525origin/mainpackages/core/src/utils/blank-storage-form.tsblankAsNullColumnsmaps an optionalmulti_valuecolumn to'emptyList'unlessreadonly: true, andblankStorageFormstoresnullfor[]or''on that reachState: closed by
Fixes #29343(completed).pm:dispatchedis removed.Delivered: a cleared optional multi-value field (
[]or'') is stored asnull, so a second record cleared on a unique multi-value column is no longer refused 409. Arequiredmulti-value field keeps its refusal. Areadonly: truecolumn such assys_activity.changed_fieldskeeps the[]the platform stamps (ruling A, 6591577985; round 2 A, 6598733139). Stored'[]'rows are not migrated here: #29653 owns that, scoped by note 6598764965 to exclude readonly columns. Nothing else is left on this card.