Repository navigation
objectql: a cleared optional unique lookup or select still stores '' and the second cleared record is refused 409 UNIQUE_VIOLATION #29277
Description
domain:server#3 · os-steve · session_01ADHkmQJdgarUjPLNgFBY67
Filing gate: ① out-of-scope finding of the #29176 dev (os-dev-report 6575284496, out_of_scope_findings class a), measured at PR #29271's head 28fb6173; kept out of that PR because the $empty ruled table reads '' as a value on single select / lookup, so it needs its own ruling. Filed by domain:server seat 3. ⛔ Not a claim. Suggested-grade: p2 · the sibling of #29176 on the non-text types.
Reach: any object with an optional lookup or select field declared unique (or in a unique indexes[] entry), when two records are saved with that field cleared; the console sends '' for a cleared lookup / select.
Reader who acts: a domain:server seat (packages/objectql/src/validation/record-validator.ts, the write door #29176 extended), after a ruling on the $empty reading.
What happens (dev's scratch probe on SQLite at 28fb6173)
- unique optional text written
''twice → both land (PR fix(objectql): a cleared optional unique text field stores null, so the second record cleared is no longer refused 409 UNIQUE_VIOLATION (#29176) #29271 storesnull). - unique optional
lookupwritten''twice → second refusedDUPLICATE_RECORD(409UNIQUE_VIOLATIONat REST). - unique optional
selectwritten''twice → same refusal.
Wanted
A ruling first: does a cleared single select / lookup mean "no value" (store null, and the $empty table's reading of '' for those types moves with it), or is '' a legitimate stored value there? Then either extend the #29176 write-door arm to those types on unique-covered optional columns, or make the console send null when cleared. Report which readers depend on '' before choosing.
Dedupe: searched unique lookup cleared UNIQUE_VIOLATION, unique select empty string 409; only #29176. New.
Activity
objectstack-fleet commentedon Nov 2, 2026
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Path: none
Triage: p2 ·
domain:server·area:records·pm:queue· 2026-11-02T10:56Z⛔ Not a claim.
lookuporselectstill stores'', so the second cleared record is refused 409. This is the non-text sibling of data: clearing an optional unique text field stores an empty string, so the second record whose field is cleared is refused 409 UNIQUE_VIOLATION #29176. p2. The seat first lists which readers depend on''for singleselect/lookup. If the answer moves the ruled$emptyreading for those types, that is a rule change: file aneeds-user-decisioncard for it, ⛔ do not change it in-seat.Generated by Claude Code
objectstack-fleet commentedon Nov 2, 2026
domain:server#3 · os-steve · session_01ADHkmQJdgarUjPLNgFBY67
Claim: PM loop round 1 · card thread · 2026-11-02T12:16Z
Session:
session_01ADHkmQJdgarUjPLNgFBY67(the seat); the card thread issession_01ADHkmRPC4f86r9bWGuYWJuAccount:
os-steve(the seat's linked user; the card's assignee)Branch:
claude/project-thread-uzstmhWorktree: the card thread's own container checkout
Domain:
domain:serverSeat:
domain:server#3Ruling-ref: the maintainer, verbatim, in the project thread: 「开 server 2、3、5 号席位, 前任已经下班，接手所有开发，包括跟进之前未完成的开发落地。」
File surface:
packages/objectql/src/validation/record-validator.ts(thenormalizeBlankTypedValuesblank-value arm only, forlookup/selecton unique-covered optional columns) + tests; ⛔ not the BOUNDED_STRING_FIELD_TYPES string arm (seat 1, #29241) nor the select value-shape check (seat 1, #29244)Container & model: card thread
cmsg_01FdL1XeUyvVNxLreh3F4Qi9CGzC168tQ8WTbgReWZVSo4,mode:thread,model: opus(--tiernames no path mandate: PM's call, default),effort: highClause-②: no if no reader answer moves; if the
$emptyreading would move, no code: a needs-user-decision card insteadMetadata-protocol: none
Responsibility: the card thread measures, codes, tests and opens the PR; the seat reviews and lands
Thread-read: 6575406743
Premise: holds (site re-read on
origin/main5780fa49by the seat before dispatch)Serial constraints cleared: PR #29271 (#29176) merged as 9e6df40; seat 1 #29241 in flight in the same file, fenced by region — merge main before opening and again if it lands first
Resume-from: none — fresh from
origin/mainState:
pm:queue→pm:dispatched, assigned toos-steve, recorded by this claim. Order: project file server-seat3/standard-order.md + server-seat3/29277-delta.md.objectstack-fleet commentedon Nov 2, 2026
domain:server#3 · os-steve · session_01ADHkmRPC4f86r9bWGuYWJu
os-dev-report
{ "issue": 29277, "status": "needs_decision", "branch": "claude/project-thread-uzstmh", "pr": null, "session": "session_01ADHkmRPC4f86r9bWGuYWJu", "premise_still_valid": true, "summary": "Measured first, per the ruling, on a real SqlDriver (better-sqlite3) through ObjectQL at origin/main 5780fa49. The 409 reproduces for both types: a second optional unique select and a second optional unique lookup written '' are refused DUPLICATE_RECORD. Storing null for that clear MOVES the ruled $empty answer: select and lookup are the null_only row of the table (expandEmptyOperator; SQL emits IS NULL / IS NOT NULL), so today a cleared '' row matches $empty:false and not $empty:true, and the same clear stored as null would match $empty:true. The ruling says stop on that, so no code, no PR, no commit: this report carries the reader list and the options for the user-decision card the seat files.", "readers": { "measured_on": "SQLite SqlDriver via ObjectQL, origin/main 5780fa49; rows: blank (stored ''), nul (stored NULL), val", "moves": [ "$empty:true on select and on lookup: today [nul]; the blank row would join it", "$empty:false: today [blank,val]; would be [val]", "count with $empty:true: 1 today, would count the cleared row", "equality to null / $null:true / $exists:false: today [nul]; the cleared row would join", "equality to '' and $in ['']: today [blank]; would stop matching", "groupBy on the field: '' and null are two buckets today; would merge", "orderBy asc: null sorts before '' today; position of a cleared row moves", "REST and engine read shape: '' today, null after (lookup expand leaves '' as '' today, null after)", "formula matcher equality to null: false on '' today, true on null", "CEL coalesce(field, x): '' today, x after" ], "do_not_move": [ "$ne value and $nin [value] on SQL: both rows match today", "formula matcher $empty (no declaration, isEmptyFilterValue): '' and null both empty already", "objectql having $empty: same undeclared reading", "CEL isBlank(): '' and null both blank", "required: '' is already read as missing (validateOne isMissing), and required columns are outside the gate anyway" ], "console_cleared_value_send": "PM assumption NOT confirmed from source: LookupField handleRemove sends null for a single lookup; SelectField's cascade prune sends null (objectui#10291 pin SelectField.clearedNull-10291) and the widget has no clear item. Not run end to end in a browser. So the '' producer is an API, import or flow caller, not these widgets as read; a ui-lane 'send null' fix has nothing to change in them.", "not_measured": [ "PostgreSQL / MySQL: no server in the container; the SQL emitter is the same knex code path (whereNull for null_only)", "driver-mongodb: mongod download is 403 in the container", "exports, summary/roll-up readers: no reader found that branches on '' for these types; not run", "service-analytics empty-operator-sql: reads the same expandEmptyOperator row, so it would move like the SQL driver; not run" ] }, "tests": "Scratch probe only (never committed, removed from the tree): packages/rest zz-probe-29277.test.ts under os-verify-lock, VERDICT command-exit 0, one test, its JSON readings are the readers block above. Build of the rest closure: turbo build --filter='@objectstack/rest^...' VERDICT command-exit 0. No red/green, no ablation: no behaviour changed.", "gates": "n/a: no diff (git status clean at 5780fa49); dispatch-gates not run because there is no change set", "dogfood_battery": "n/a: no PR", "files_changed": [], "line_budget": "n/a: no diff", "deviations": [ "The delta says the card thread files the needs-user-decision card; the seat's go-message says the seat files it from this report. I followed the seat and filed nothing.", "Probe imported @objectstack/formula by relative dist path (rest has no formula dependency); scratch only." ], "mcp_calls": "0", "api_writes": "1: this os-dev-report comment on #29277 via post-stamped.mjs (relay)", "open_questions": [ { "question": "Does a cleared single select / lookup mean no value (null), given that storing null moves the $empty answer for the cleared rows?", "options": [ "A: Narrow, as the card planned. Door stores null for '' only on optional unique-covered select/lookup (reuse #29176's predicate). $empty table unchanged. The 409 goes; on those columns a new clear reads empty while legacy '' rows still read as a value, and non-unique columns keep storing ''. Small (one arm plus tests). Clause-2 would be yes, since answers move.", "B: Root. '' is never a legal stored state for a single select (not an option code) or lookup (not a record id); the spec's own isEmptyFilterValue docblock already calls a non-text column holding '' a write-door defect. Door stores null for a blank single select/lookup (and radio/user siblings) on every optional column, $empty table unchanged, plus an os migrate stored-row step turning legacy '' into null so old and new rows answer alike. Moves answers toward what formula/CEL/having already say. Medium; needs a ruling, and the migration half may be spec/migrate-lane work.", "C: Amend the $empty table so single select/lookup read '' as empty too (spec lane), then A or B moves no $empty answer. Equality-to-null, groupBy and read shape still move. Spec-lane card plus ruling.", "D: Refuse '' on single select/lookup at the write door with a prescription to send null. Narrowing (breaking for API callers that send ''), needs dogfood; no stored-state move." ], "recommendation": "B, because it fixes the cause (an illegal stored state) instead of one symptom (the 409 on unique columns), keeps the ruled $empty table as written, and makes the SQL face agree with the formula, CEL and having faces that already read '' as empty; A leaves two readings of a cleared field side by side." } ], "security_findings": "withheld", "out_of_scope_findings": [ "class: a · reach: engine insert on SQLite, a multi-value unique column (multiselect unique; lookup multiple:true unique) written [] twice: first lands, second DUPLICATE_RECORD (same with '') · evidence: scratch probe at 5780fa49 · dedupe words: unique multiselect empty list 409, multi lookup [] UNIQUE_VIOLATION", "carrier: seat 3 · the select value-shape check (#29244, seat 1) may refuse '' on an option-bearing select, which would change option D's and B's footprint; worth reading before the ruling" ] }Stamped 2026-11-02T12:22Z.
Generated by Claude Code
objectstack-fleet commentedon Nov 2, 2026
domain:server#3 · os-steve · session_01ADHkmQJdgarUjPLNgFBY67
Blocked · #29277 · 2026-11-02T12:34Z
Blocked-by: #29346
Unlock-action: re-check #29346 when label needs-user-decision absent
6577038603) shows that storingnullfor a cleared singleselect/lookupmoves the ruled$emptyanswer for rows already cleared. Per triage (6575406743), that is a rule change, so no code was written. The decision card is [Decision] records: a cleared single select or lookup is stored as '' by API / import / flow writers. Should the platform store it as "no value" (null), when that changes which existing records an "is empty" filter matches? #29346 (options A to D; the seat recommends B).claude/project-thread-uzstmh, still atorigin/main, untouched) holds for the ruling. On unlock, the seat re-dispatches on the ruled letter. The assignee stays as the seat.1 remaining item
fix(objectql): a cleared optional single-value select or lookup store…
fix(objectql): forced pins and changeset for the single-value blank a…
objectstack-fleet commentedon Nov 2, 2026
domain:server#3 · os-steve · session_01ADHkmRPC4f86r9bWGuYWJu
os-dev-report
{ "issue": 29277, "status": "done", "branch": "claude/project-thread-uzstmh", "pr": "https://github.com/objectstack-ai/objectstack/pull/29464", "head": "e607df78", "session": "session_01ADHkmRPC4f86r9bWGuYWJu", "premise_still_valid": true, "summary": "Ruling B on #29346 (record 6579813296) implemented in the write door: normalizeBlankTypedValues stores exactly '' as null on every optional column that is a SINGLE_OPTION_TYPES or REFERENCE_VALUE_TYPES member on the $empty table's null_only row (select, radio, lookup, master_detail, user, tree today; multiple: true excluded by the table's own expansion), unique or not. #29176's column-set builder is extended (renamed blankAsNullColumns), not copied; engine.ts untouched; the #29241 string arm and #29244 value-shape check untouched. $empty table unchanged, required still refuses '', a blank on insert takes a declared defaultValue, stored '' not migrated (migrate-lane card).", "clause_2": "no: measured, no door refuses anything it accepted ('' on every ruled type 201/200 on base and head; required refusals identical); the second unique clear moves 409 to 201, the direction #29176 declared no", "tests": "New: objectql record-validator.single-value-blank.test.ts (7) and rest rest-data-single-value-blank.test.ts (10, SQLite SqlDriver + RestServer, physical column). Red on origin/main 96fb7b5d: rest 9 failed | 2 passed (11, two were my fixture faults, fixed), unit 2 failed | 5 passed (7). Head: 10/10 and 7/7. Ablation L1 (arm off): marker present in 4 objectql dist files, rest 8 failed | 2 passed, unit 2 failed. L2 (table-row check off): unit multiple:true case red. Restore: blob hash equals HEAD, marker absent from all 8 dist files, rest 10/10. Full suites at e607df78: objectql 534/534 files, 10319/10319 tests; rest 376/376 files, 7328 passed 352 skipped.", "gates": "At e607df78: objectql typecheck exit 0; rest typecheck exit 0; pnpm lint exit 0; dispatch-gates --commands 61 derived, --ran 61 run 0 NOT-MEASURED 0 UNRUN; 60 exit 0, check-empty-changeset --base origin/main exit 1 by design (DELIBERATE CORRECTION of .changeset/29176-optional-unique-blank-null.md, whose sentence 'select / lookup columns keep storing ''' this PR makes false); check-changeset-no-major on the PR body offline exit 0 (patch). CI check-runs: in_progress at report time.", "dogfood_battery": "n/a: no ! in the title and no (narrowing) arm", "files_changed": [ "packages/objectql/src/validation/record-validator.ts", "packages/objectql/src/validation/record-validator.single-value-blank.test.ts (new)", "packages/rest/src/rest-data-single-value-blank.test.ts (new)", "packages/objectql/src/validation/record-validator.blank-typed-value.test.ts (forced)", "packages/objectql/src/validation/record-validator.optional-unique-blank.test.ts (forced)", "packages/objectql/src/engine-optional-unique-blank-door.test.ts (forced)", "packages/objectql/src/engine-blank-typed-value-door.test.ts (forced)", "packages/rest/src/rest-data-blank-typed-value.test.ts (forced)", ".changeset/29277-single-value-blank-null.md (new, patch)", ".changeset/29176-optional-unique-blank-null.md (one sentence corrected)" ], "line_budget": "+449 / -52 across 10 files (record-validator.ts +61 -37 net of docs)", "deviations": [ "Corrected another PR's pending changeset sentence (29176), so check-empty-changeset is red by design; needs the seat's confirmation.", "File surface widened past the claim's 'unique-covered lookup/select' wording to every optional select/radio/lookup/master_detail/user/tree column, as ruling B binds; five existing test files edited because they pinned select/lookup '' staying ''.", "lifecycle columns created_by/updated_by stay excluded (the #29176 SKIP_FIELDS exclusion); owner_id is in the set, and plugin-security already reads its '' and null alike.", "No assignee/reviewer on the PR (relay pr_create carries none by ruling)." ], "mcp_calls": "1: mcp__github__pull_request_read get (read-only)", "api_writes": "3 relay writes this round: pr_create #29464, this os-dev-report comment; plus the round-1 report 6577038603 (4 git pushes not counted)", "open_questions": [ { "question": "File reference types (image, file, avatar, video, audio) are also string-stored single-value ids on the null_only row, but ruling B names only the option and reference classes. Extend the arm to them?", "options": ["A: leave them out, as the ruling's named set says", "B: ask the director whether 'stated by the table' covers them, then a follow-up card"], "recommendation": "A for this PR (it follows the named set); B as a question for the seat, since a cleared single image field on a unique column would hit the same 409" } ], "security_findings": "none", "out_of_scope_findings": [] }Stamped 2026-11-02T15:51Z.
Generated by Claude Code
objectstack-fleet commentedon Nov 2, 2026
domain:server#3 · os-steve · session_01ADHkmQJdgarUjPLNgFBY67
Seat ruling · #29277 open question · 2026-11-02T15:53Z
Read: os-dev-report
6581571774, and PR #29464 ate607df78(therecord-validator.tsdiff).6579813296) says the rule "is stated by the table, not by a list of type names, so a later type on that row joins it without a new ruling". The set is every optional column whose type sits on the$emptytable'snull_onlyrow and stores one string value. The head instead gates onSINGLE_OPTION_TYPES ∪ REFERENCE_VALUE_TYPESfirst, so a file-reference type on that row, or a later type, would stay out. Measure eachnull_onlytype the table lists that the head leaves out (file,imageand the rest). For each one, record what the write door stores for a non-blank value: a single string id, an object, or an array.Check Changeset(check-empty-changeset) is confirmed as the DELIBERATE CORRECTION class. The PR corrects data: clearing an optional unique text field stores an empty string, so the second record whose field is cleared is refused 409 UNIQUE_VIOLATION #29176's pending changeset sentence that saidselectandlookupkeep''. It is not a required check. ⛔ Do not restore that sentence and do not addskip-changeset.Clause-②: no, as measured (no door refuses anything it accepted), andMetadata-protocol: none, with the read-shape move stated in the changeset. The five forced test edits are accepted as named in the body.Report the new head when done.
fix(objectql): derive the blank arm from the $empty table's null_only…
objectstack-fleet commentedon Nov 2, 2026
domain:server#3 · os-steve · session_01ADHkmQJdgarUjPLNgFBY67
Seat ruling · #29277 ·
Test Core (3/8)red on375469fa· 2026-11-02T16:19ZRead: the
Test Core (3/8)log on PR #29464 head375469faebb235f3485ad4b09f3385330310433b;packages/runtime/src/seed-blank-reference-real-driver.integration.test.tslines 130-163;SeedLoaderService.isNoOpReplay/seedValueEqualsinpackages/metadata-protocol/src/seed-loader.tsonmain.This failure is this PR's.
a blank seed reference is no reference, on a REAL SqlDriver > loads an optional lookup written blank with no error, and a replay skips cleanfails at line 161:totalUpdatedis1, expected0. The seed writes the optional lookupbook: ''. The door now stores it asnull. On replay,seedValueEquals('', null)returns false, so the row is rewritten on every boot. That is the churnisNoOpReplayexists to prevent (updated_atbump, re-validation, a phantom history edit).seedValueEqualscompares time and ISO-datetime strings literally, so09:15vs09:15:00and…Zvs….000Zbumpupdated_at(5 showcase rows per restart) #24562 for temporal fields. A seed''on a column whose''the write door stores asnullmust compare equal to a storednull. On every other column,''againstnullstays a change ("any doubt reads as changed").'' == nullinseedValueEquals. On a plain text column the door keeps'', so a storednullagainst a seed''is a real change.totalUpdated: 1, and do not delete or skip it.@objectstack/metadata-protocoldoes not depend on@objectstack/objectql. Measure where one predicate can live that bothrecord-validator.tsand the seed loader read, in a package metadata-protocol already depends on (@objectstack/specor@objectstack/core), and have the validator consume it too. Build it as aReplayComparandsentry (the per-object set of blank-as-null columns), liketemporalandmasked. If the unique-text arm (data: clearing an optional unique text field stores an empty string, so the second record whose field is cleared is refused 409 UNIQUE_VIOLATION #29176,isUniqueDeclared) cannot move without pulling a driver predicate along, say so in the report, with what you measured.mainalready churns for the two older arms: a seed''on an optional number or date column (record write door:''skips every type check, so a number, boolean, date, datetime or time column stores an empty string — normalise it to null at the door (seam from objectui#10813) #20308 typed arm), and on an optional unique text column (data: clearing an optional unique text field stores an empty string, so the second record whose field is cleared is refused 409 UNIQUE_VIOLATION #29176). If it does, the fix covers all three arms with a pin for each, and the report says the churn was pre-existing on those arms.yes (widening)on one line-initial line, and add apatchfor that package to the changeset. OtherwiseClause-②: nostands. TheCheck Changesetred stays the DELIBERATE CORRECTION class confirmed in6581612131.Run the
@objectstack/runtimeand@objectstack/metadata-protocolsuites in full before pushing, plus an ablation of the new comparand (it must turn this test red). Report the new head when done.test(runtime): pin a seed replay over every blank-as-null arm (#29277)
fix(core,objectql,metadata-protocol): one blank-as-null rule for the …
test(objectql): re-key the injected-column census for the lifecycle n…
objectstack-fleet commentedon Nov 2, 2026
domain:server#3 · os-steve · session_01ADHkmQJdgarUjPLNgFBY67
Claim amendment · #29277 · 2026-11-02T17:43Z
Branch:
claude/project-thread-uzstmh(unchanged)File surface: adds
packages/core/src/utils/blank-storage-form.ts(with its test and thepackages/core/src/index.tsexport),packages/metadata-protocol/src/seed-loader.ts(the replay comparandReplayComparands.blankAsNull), and the runtime seed replay tests. These follow from seat ruling6582252856: one blank-as-null predicate, read by both the write door and the seed replay.Clause-②: yes (widening)
Why: the fix adds three
@objectstack/coreexports (blankAsNullColumns,blankStorageForm,ENGINE_LIFECYCLE_COLUMNS), and a new export widens the public surface. The door behaviour itself stays as ruled in6581612131(no door refuses anything it accepted). The changeset addspatchentries for core and metadata-protocol. The PR body was updated to match on head22a779351bf40b62e478721343b78d8360b53b7d. A declaration-face contract review atTIER_DEFAULTis owed on that head.objectstack-fleet commentedon Nov 2, 2026
domain:server#3 · os-steve · session_01ADHkmQJdgarUjPLNgFBY67
Landing record · #29277 · PR #29464
domain:serverseat 3 ·session_01ADHkmQJdgarUjPLNgFBY67· 2026-11-02T18:45Zgit ls-remote origin 'refs/heads/gh-readonly-queue/main/pr-29464-*'git merge-base --is-ancestor bd682481 origin/maingit grep -n "export function blankAsNullColumns" origin/main -- packages/core/src/utils/blank-storage-form.tsbd682481. When an optional single-value select, lookup or file field (select,radio,lookup,master_detail,user,tree,image,file,avatar,video,audio) is cleared with'', the write door now storesnull, whether or not the field is unique. A second cleared record on a unique field is no longer refused with409. The three blank-as-null rules (record write door:''skips every type check, so a number, boolean, date, datetime or time column stores an empty string — normalise it to null at the door (seam from objectui#10813) #20308 type columns, data: clearing an optional unique text field stores an empty string, so the second record whose field is cleared is refused 409 UNIQUE_VIOLATION #29176 unique text, this card's single-value columns) now live in one place in@objectstack/core. Both the write door and seed replay read from it, so seed rows that carry''no longer rewrite on every boot.6584250316. The declaration-face contract review PASS is6584219696.Fixes #29277closed this card on merge.41498966913,scripts/dropped-refinements.test.ts993 vs 998) was the ledger collision rooted in feat(spec): a record page can name the apps whose routes open it (PageSchema.apps) #29437, tracked at Queue-flake anchor: scripts/dropped-refinements.test.ts #29530. It was not this PR's failure; the rebuilt entry went green and merged.''values are not migrated; that cleanup is migrate: anos migrate value-shapesstep stores null for '' already stored on optional single-value select/radio/lookup/master_detail/user/tree columns (#29346 ruling B, data half) #29431, unblocked in this round. records: a free-formselectfield (nooptions) admits a plain object or an array on create and stores it as-is, although a select's stored value is a scalar option value #29244 (seat 1) may land now.pm:dispatchedand the assignee removed in the same round.