Repository navigation
finding(core,data-objectstack): the OBJECT-FORM filter dialects still lower an empty or non-string icontains comparand onto the wire — the shape FILTER_TEXT_CASES declares REFUSED #9048
Description
Restart-when: npm view @objectstack/spec version reports a version GREATER than 17.4.0 AND objectui's pnpm-lock.yaml resolves that version — that is the cheap trigger a sweep can watch, and it is ⛔ NOT the licence: on a hit, re-run the installed-surface probe in the hold comment and start work only when isRefusedTextComparand is reachable from the INSTALLED @objectstack/spec.
Restart-touch: pnpm-lock.yaml
Upstream (⛔ no longer a Blocked-by: line, because the target is CLOSED and a satisfied one makes the unlock scan re-fire on a card that is on hold): objectstack-ai/objectstack#18113, closed completed 2026-09-16. What still gates this card is the CONSUMER INSTALL SURFACE, not that merge — see the hold comment.
Found by the os-dev seat while implementing objectui#9001. Out of that card's scope: triage fenced it to convertFiltersToAST, and these are two different functions on two different authoring surfaces. Grading and domain:* are the triage seat's. Not claimed.
The class
@objectstack/spec's FILTER_TEXT_CASES carries two REJECTION rows for the case-insensitive contains operator — an empty comparand and a non-string one — each with code: 'INVALID_FILTER' and mustMention naming the operator. ValueDataSource has refused both in both dialects since objectui#8748: its $ arm and its AST arm (['name', 'icontains', '']) each call refuseTextComparand, deliberately, because "find() picks between the matchers on nothing more than whether $filter arrived as an array or an object, so a door on one side only is a result that changes with the SHAPE of the filter rather than with its meaning".
objectui#9001 puts that door on convertFiltersToAST — the $-dialect producer. Two producer paths that emit the same node are left without one, and neither is reached by that card's fix:
viewFilterRuleToNode(packages/core/src/utils/filter-converter.ts, reached through the exportedtoFilterNode) — the STORED VIEW rule vocabulary.icontainsis a member of the spec'sVIEW_FILTER_OPERATORS, and the function's tail is[rule.field, operator, rule.value]with the value unread except by objectui#8557's array-arity guard.objectFilterEntryToAST(packages/data-objectstack/src/index.ts, reached throughtranslateFilterArraywhen the filter is in[{ field, operator, value }, ...]form) — its tail is[String(field), op, entry.value], andFILTER_OPERATOR_ALIASEScarriesicontainsas an identity row. Note the object branch oftranslateFilterToASTdelegates toconvertFiltersToASTand therefore DOES inherit objectui#9001's refusal; it is only the ARRAY branch that does not.
Measured
Worktree at origin/main 152f0a700 plus the objectui#9001 branch (so the $ dialect is already closed), spec 17.4.0:
toFilterNode([{ field: 'name', operator: 'icontains', value: '' }]) => [["name","icontains",""]]
toFilterNode([{ field: 'name', operator: 'icontains', value: 42 }]) => [["name","icontains",42]]
toFilterNode([{ field: 'name', operator: 'icontains', value: 'acme' }]) => [["name","icontains","acme"]]
convertFiltersToAST({ name: { $icontains: '' } }) => THROWS FilterOperatorError INVALID_FILTER / 400
objectFilterEntryToAST was read rather than driven (it is not exported); its body is three lines and takes entry.value verbatim after normalizeFilterOperator, which answers icontains for icontains.
So after objectui#9001 the SAME wire node is refused when the author writes it as { name: { $icontains: '' } } and sent when the same author saves it as a view rule or hands it to the ObjectStack data source in entry form. That is the acceptance-set split objectui#8447, objectui#8568, objectui#8976 and objectui#9001 have each closed on a different axis, restated on the DIALECT axis one layer up.
Why this is a card rather than a same-PR repair
Three things the objectui#9001 fix does not settle, and the reason it was fenced away from them.
- Blast radius is different in kind.
convertFiltersToASTrefuses a filter an author is composing right now.viewFilterRuleToNoderefuses a rule someone saved months ago, and a throw there means the saved view fails at RENDER. objectui#8557's docblock argues that trade at length for the array case and concludes loud-beats-silently-empty; whether the same answer holds for an empty text comparand — where the pre-fix behaviour is "every row" rather than "no rows" — is the same weighing done again with a different input, not a mechanical copy. - The message cannot be ported unchanged. The
FILTER_TEXT_CASESrows require the refusal to mention the$-dialect spelling. A stored view rule spells the operator without the$, so a verbatim port would prescribe a spelling the view vocabulary does not have.ValueDataSourcesolved this by naming whichever spelling arrived; a producer refusal has to decide what it tells a view author. - It crosses a package. The
data-objectstackhalf is a second published package with its own refusal envelope (MalformedFilterError), so it is a second changeset and a second review surface.
Not created by objectui#9001
Stated precisely, because the timing invites the wrong reading. Both paths lowered this node before that card and lower it after; what the card changed is that the $ dialect no longer does, which is what makes the split visible. objectui#9001's own body already recorded the data-objectstack entrance as pre-existing — this card is where that observation gets a home, plus the viewFilterRuleToNode sibling it did not name.
Dedup
Two channels, and the first one failed its own control — recorded because the failure is the documented one.
- MCP
search_issues, repo-scoped: a first query returned exactly one result (objectui#8557) and did NOT return objectui#9001, whose title is near-verbatim for it; a second, shorter query (icontains comparand guard) returnedtotal_count: 0for a subject that demonstrably exists. Neither is a usable reading. - Bounded REST listing,
state=openwithlabels=bug,domain:ui: 64 issues, one page, complete. Lit control: objectui#9001 is in the listing. Keyword sweep over titles and bodies foricontains|comparand|filter-converter|FILTER_TEXTreturns three — objectui#9001 (this card's parent), objectui#9030 (the same function's TRUE-identity tail counting keys the loop skips) and objectui#9020 (a null-valued filter key getting two different row sets). None names the object-form dialects' text comparand. The nearest CLOSED neighbour is objectui#8557, which isviewFilterRuleToNode's ARRAY arity on scalar operators — the same function, a different comparand question.
Refs: objectui#9001, objectui#8748, objectui#8557, objectui#8447.
Filed by an agent seat while implementing objectui#9001; the session reference for that run is https://claude.ai/code/session_01MPaVWWMuWeT5LgB1qoXjVB.
Generated by Claude Code
Activity
os-litant commentedon Sep 10, 2026
Triage:
domain:ui; typeBug,priority:p2,pm:queue.Class (b), and the second dialect of a defect this seat already graded. The OBJECT-FORM filter dialects lower an empty or non-string
icontainscomparand onto the wire — the shapeFILTER_TEXT_CASESdeclares REFUSED, withcode: 'INVALID_FILTER'andmustMention: ['$icontains'].convertFiltersToAST(the AST dialect); this is the object-form dialect. ⇒ Two dialects, one contract, one already-shipped reference implementation inValueDataSource(refuseTextComparand, objectui#8748).⭐ That makes three implementations of one refusal, of which one is correct. ⇒ The useful dispatch is not "add a third guard" but "make the dialects share the one that works".⚠️ Whoever takes this or #9001 should say which; ⛔ do not dispatch both at once.
priority:p2: a shape the contract declares refused reaches the wire, so refusal happens late or not at all, and the error the author sees is not the one prescribed.Triage seat ·
session_017VGfRocA8VjczSe84fgjY3· R+173 · 2026-09-10T21:30Z (timestamp taken in the same tool call that posts) · comment from the triage seatGenerated by Claude Code
os-tesla commentedon Sep 11, 2026
Claim:
domain:ui派单席 · R16 · objectui#9048Branch:
claude/issue-9048-object-form-icontains-comparandClause-②: yes
为什么是 yes
拒绝的形状是
@objectstack/spec的FILTER_TEXT_CASES规定的（code: 'INVALID_FILTER'、mustMention），而分诊要的「让方言共用同一道」很可能需要从@object-ui/core导出一样东西给@object-ui/data-objectstack用 —— 那是已发布面的新增。拿不准 ⇒ 按 yes。needs:contract-review双载体，卡与 PR 各挂一份。依赖已按 53 号事实逐个核过（实测
origin/main6df26f025）packages/core/src/utils/filter-converter.ts里已经有一个本地refuseTextComparand，它自己的 docblock 写着这是 ValueDataSource 那道的「a PORT of a shipped, reviewed」实现。viewFilterRuleToNode的尾巴仍是[rule.field, operator, rule.value]，rule.value除 objectui#8557 的数组元数守卫外无人读。⇒ ⭐ 这把分诊那句话变得更尖锐了。分诊说：「三个实现，只有一个是对的 ⇒ 有用的派单不是『再加第三道守卫』，而是『让方言共用那一道有效的』」。而 objectui#9001 落地的方式是再抄一份。⇒ 现在是两份副本加两个没有守卫的生产者，合并的对象比卡填写时多了一个。
两个真正的岔口，已写进简报，⛔ 都不许猜
convertFiltersToAST拒绝的是作者此刻正在编的筛选；viewFilterRuleToNode拒绝的是几个月前存下的规则，抛错意味着那张视图在渲染时挂掉。objectui#8557 为数组情形做过这个权衡并选了「响亮好过静默为空」，但本卡修复前的行为是每一行都返回（不是零行），输入不同 ⇒ 结论不能继承。filter-converter与ValueDataSource都在@object-ui/core内，合并它们不需要任何新导出；而data-objectstack是另一个包，它要够到同一道就得走@object-ui/core的公开入口 ⇒ 新增已发布导出 = 人工门槛。⛔ 那一步不是本轮能自己落的。domain:ui派单席 ·session_01UzHd6hDYatoDn17BuwKxnZ· R16 · 2026-09-11T10:5xZ_Generated by Claude Code
Generated by Claude Code
os-tesla commentedon Sep 11, 2026
os-dev-report
{ "issue": 9048, "status": "done", "branch": "claude/issue-9048-object-form-icontains-comparand", "pr": "https://github.com/objectstack-ai/objectui/pull/9152", "premise_still_valid": true, "summary": "The @object-ui/core HALF ONLY — the PR says `Part of #9048`, not a closing keyword, because the card's `@object-ui/data-objectstack` producer is NOT addressed and stays open (FORK B below). Landed: (1) the discrimination and the refusal message move into one internal module `packages/core/src/utils/text-comparand.ts` that ValueDataSource, convertFiltersToAST and viewFilterRuleToNode all read — triage's 'make the dialects share the one that works', not a third copy; two copies each of describeComparand, the `declared` sentence and the reason text are deleted. (2) viewFilterRuleToNode now refuses the two FILTER_TEXT_CASES shapes instead of lowering them. Both already-shipped messages are byte-identical after the consolidation (same sha256 over six driven messages, before and after). No published export was added. FORK A was re-measured, not inherited: the pre-fix answer is TWO behaviours, not one — 0 of 9 FILTER_TEXT_ROWS through @objectstack/formula and through ValueDataSource, versus the wire answer the published table states as 'a predicate that constrains nothing' — so neither half is a view that renders correctly today, and objectui#8557's loud-beats-silently-empty covers the first half while the second is the widening the file exists to avoid. FORK B is measured, not assumed: @object-ui/core declares exactly one `exports` entry, so data-objectstack cannot reach the shared refusal without a new published name; that is the manual floor and it is not in this PR. ⚠️ objectui#9050 (open, unclaimed) governs WHERE the throw surfaces — four render-time `toFilterNode` entries do not catch it — and this PR deliberately does not pre-empt it; see open_questions. assignee was `os-tesla` on arrival (PM's). Clause-②: yes, unrevised; `needs:contract-review` hung on PR 9152 with an additive write and read back on both carriers.", "tests": "ALL RUNS FROM THE REPO ROOT, the one canonical vitest invocation. FINAL HEAD a5d590ef0, working tree clean; the two rows below were re-run ON that head after the last commit. (1) `pnpm exec vitest run packages/core/` -> exit 0, 'Test Files 148 passed (148) / Tests 3141 passed (3141)'. (2) `pnpm --filter @object-ui/core lint` -> exit 0, '542 problems (0 errors, 542 warnings)' — all warnings pre-existing no-explicit-any; text-comparand.ts reports 0/0. (3) Under the shared verify lock: `pnpm exec vitest run packages/data-objectstack/ packages/plugin-list/src/__tests__/filter-operator-ast-parity.test.ts packages/components/src/hooks/ packages/plugin-view/src/config/` -> 'Test Files 67 passed (67) / Tests 945 passed (945)', wrapper printed 'VERDICT command-exit 0 · held the lock 11s · waited 232s'. (4) `pnpm --filter @object-ui/core exec tsc --noEmit` exit 0; `pnpm --filter @object-ui/core build` exit 0, 'dist completeness: 202 emitted files verified'. (5) Gates, each exit captured by redirect before any pipe: check:new-line-citations, check:control-bytes, check:self-import, check:phantom-deps, check:unused-deps, check:spec-symbols, check:entry-guard, check:lint-rule-coverage, check:side-effects-array all exit 0; check-changeset-presence.mjs exit 0 ('2 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s)'); check-changeset-no-major.mjs exit 0. Control-byte self-scan with grep -naP over all five changed files: no match, grep exit 1. (6) NOT MEASURED: `node scripts/pm/check-half-states.mjs --pair 9048` exited 3 with its own printed reason 'PREREQUISITE NOT MET — the token in the environment is not a valid GitHub credential ... Nothing was swept ... it is no reading at all'. The dual carrier was confirmed instead by direct REST read-back: card 9048 labels 'bug, pm:dispatched, priority:p2, domain:ui, needs:contract-review'; PR 9152 labels 'package: core, tests, needs:contract-review' (the first two set by an automation, not by me). (7) BYTE-IDENTITY OF THE TWO SHIPPED FACES: convertFiltersToAST x 2 shapes and ValueDataSource's $ arm and AST arm x 2 shapes were driven in one process before and after the consolidation; captured output sha256 be5b77338e92fe8a58d580718632e3e42dd17741cf7b703dd98b28095dbb728f BOTH times, diff empty. (8) LIVE CONTROL, same run, pre-existing: toFilterNode([{field:'name',operator:'icontains',value:'acme'}]) is [[\"name\",\"icontains\",\"acme\"]] before and after; sibling contains/starts_with/ends_with/not_contains/equals unchanged on the same comparands; objectui#8557's ARRAY arm still answers first for an array; a rule with no comparand still lowers to [\"name\",\"icontains\"]. (9) ABLATION — three legs, subject resolves to SOURCE (relative import plus the vitest alias to packages/core/src), so no dist staleness applies. Each leg: mutate, prove on disk by EXACT SUBSTRING OCCURRENCE COUNT BOTH WAYS plus a moved blob hash, run, restore with `git checkout HEAD -- path` (never the bare form), prove the restore BY STATE — git hash-object back to the HEAD blob AND git diff HEAD empty for that path; never by an exit code; script carries a trap on EXIT, INT and TERM that runs the restore (written as a word, not as an angle-bracket placeholder, so the sanitizer cannot eat it) with absolute paths. HEAD blob 556ac36bc6f4da1e543b491c800ee3aa84a4fbcb. Leg A, the new arm's operator key made unreachable (1->0 removed, 0->1 injected, blob 556ac36 -> badaf52): 8 of 18 failed. Leg B, the arrived-spelling resolution replaced by a hard-coded $ spelling (blob 556ac36 -> ffb1094): 4 of 18 failed — both 'name the spelling that arrived' pins and both mirror pins. Leg C, the absent-comparand carve-out made unreachable (blob 556ac36 -> 0aefaca): 1 of 18 failed — the 2-tuple boundary pin. All three restored to 556ac36bc6f4da1e543b491c800ee3aa84a4fbcb with an empty git diff HEAD; final `git status --porcelain` empty. (10) LINT NARROWING, declared with all three pieces: population read from eslint's own config — `pnpm --filter @object-ui/core lint` is `eslint .` inside packages/core resolving the repo-root flat config, and the diff touches exactly one package so that run is the complete population for this diff; file count read from --format json = 250 files, 0 errors; invariance — type-aware linting is NOT enabled (root eslint.config.js extends tseslint.configs.recommended, not the type-checked variant, and languageOptions carries only ecmaVersion and globals, with no parserOptions.project and no projectService), so no rule reads the type graph and this diff cannot move the verdict on any file it does not itself contain. Other packages' lint runs belong to CI.", "mcp_calls": "1 — one targeted mcp search_issues for the dedup of the third-producer finding; it returned a non-empty result (total_count 1, objectui#9048 itself) and so is self-validating. Every other GitHub read and write went through the repo-scoped REST channel, probed green at the start of the run (GET /repos/objectstack-ai/objectui/issues/9048 -> http 200).", "open_questions": [ { "question": "FORK B — the cross-package half. `objectFilterEntryToAST` in `@object-ui/data-objectstack` still lowers both refused shapes. Measured: @object-ui/core's package.json declares exactly one `exports` entry ('.'), and src/index.ts names its re-exports one file at a time, so a second package can import only what that file re-exports. Reaching the shared refusal therefore REQUIRES a new name on @object-ui/core's published entry — the manual floor, so it is not in this PR. Which way?", "options": [ "A — add `export * from './utils/text-comparand.js';` to packages/core/src/index.ts (publishes isRefusedTextComparand, textComparandRefusalReason, describeComparand), then have objectFilterEntryToAST read the same two functions and seat the reason in MalformedFilterError's own envelope. One shared refusal across both packages; cost is three new published names and a second changeset plus a second review surface.", "B — publish ONE narrow name instead of the module (e.g. a single `refuseTextComparandReason`-shaped entry point), keeping describeComparand internal. Smaller published surface, but the discrimination predicate then either travels with it or gets restated in data-objectstack — which is the copy triage ruled against.", "C — leave data-objectstack unguarded and close the split at the ADAPTER boundary instead, by routing its ARRAY branch through an already-published core function. Rejected on measurement rather than taste: the two paths use DIFFERENT operator alias tables (data-objectstack's FILTER_OPERATOR_ALIASES maps equals to '=', not_in to 'nin'; viewFilterRuleToNode uses the spec's normalizeFilterOperator), so routing would change what a VALID comparand lowers to — the one thing the fences forbid." ], "recommendation": "A, because it is the only option that satisfies triage's ruling literally (one implementation, three envelopes) and because the two names it publishes are exactly the two halves the card family keeps re-deriving. ⚠️ Trap for whoever takes it: under vitest `@object-ui/core` is aliased to packages/core/src, so a data-objectstack test that deep-imports the internal module passes while the published build breaks — the export has to be real. A pin in the new test file asserts those three names are NOT on the published entry today and tells whoever lands this to change that assertion in the same commit." }, { "question": "FORK A's second half — should objectui#9050 be decided BEFORE this lands? The refusal throws from viewFilterRuleToNode. Two sinks catch it (plugin-list's buildEffectiveFilter inside ListView's load try, plugin-view's ObjectView inside its own) and classifyLoadError renders 'filter is malformed'. FOUR entries do NOT, verified on today's tree: plugin-grid's ObjectGrid (two call sites), plugin-detail's RelatedList and plugin-form's LineItemsPanel each call toFilterNode inside a render-time useMemo, and ObjectGrid feeds it schema.filter — the spec-declared ViewFilterRule[] this arm judges. That is objectui#9050, open and unclaimed.", "options": [ "A — land this PR now. The class is pre-existing on this exact function: objectui#8557's array-arity refusal, directly above the new arm, already throws through those same four entries and has shipped that way, and objectui#9050 counts eleven pre-existing throw sites reachable the same way. Exposure is measured latent: the Console cannot author the shape (foldFilterGroupToSpecRules drops empty values, objectui#4155), and over every tracked file an `operator: 'icontains'` rule appears only in this card's own test, changeset and comment (lit control: 40 tracked files mention the operator at all).", "B — hold this PR until objectui#9050 rules, then land the refusal and its delivery together for all thirteen sites at once." ], "recommendation": "A, and the PR is a draft so the choice stays yours. The alternatives to refusing are both worse in the direction this card family exists to close — dropping the rule widens the result set, lowering it keeps the two-answers split — and holding does not reduce the eleven throws already reachable that way. I did not pre-empt objectui#9050: the arm records the measurement and names that card as the owner of where the refusal surfaces." } ], "out_of_scope_findings": [ "noted, not filed: a THIRD unguarded producer the card does not name — plugin-list's exported `convertFilterGroupToAST` maps the builder group through `mapOperator` (which carries an explicit `case 'icontains'` arm) and emits `[field, 'icontains', value]` triples, and `toFilterNode` returns an array of AST triples VERBATIM (its rule-object branch never runs), so neither comparand door sees them. The EMPTY half is already shut there by `isFilterValueComplete`, which drops '' / null / []; the NON-STRING half is not — that predicate accepts 0, false and 42. NOT filed because I could not demonstrate a reachable entrance today: LIST_VIEW_EXTRA_OPERATORS is the empty array, so the list toolbar does not offer `containsCaseInsensitive` at all, `initialFilters` comes only from localStorage holding the builder's own group verbatim, and tab presets lower through UserFilters' own path rather than this one. That makes it undrilled drift rather than a reproducible defect or a demonstrated contract violation, which my fences say not to file. ⚠️ NAMED SUCCESSOR, because the trap is armed: filter-builder.tsx states the lasting fix for the opt-in set is upstream, 'at which point the entry below is deleted, the parity test flips it back on by itself, and the operator becomes ordinary' — so whoever lands `containsCaseInsensitive` in the list toolbar opens this entrance in the same change, with `list-offered-operator-expressible-parity.test.ts` flipping it automatically. PM: file it if you read that as class (b) rather than as drift.", "noted, not filed: packages/core/src/index.ts re-exports './utils/filter-converter.js' with `export *`, so every exported name in that file is already on @object-ui/core's published entry. Not a defect — recorded because it is what makes FORK B's measurement a one-line check rather than a guess, and because it is why the shared module had to be a NEW file rather than an addition to filter-converter.ts.", "noted, not filed: a stored view rule that carries NO comparand on `icontains` lowers to the 2-tuple [\"name\",\"icontains\"], which isFilterAST ACCEPTS while parseFilterAST throws INVALID_FILTER / 400 naming where.name.$icontains. Left untouched on purpose (FILTER_TEXT_CASES declares two rows, not three, and the view vocabulary's 'absent' is what the valueless operators depend on) and pinned as a boundary. The shape is already loud with the right code; moving WHERE it is refused would be a separate argument nobody has made." ] }Generated by Claude Code
11 remaining items
objectstack-fleet commentedon Sep 30, 2026
Unlock scan:
pm:on-hold→pm:queue. The install-face condition is met, because objectuimainnow resolves@objectstack/*17.5.0 (PR objectui#11086, merged as81f849852a, closing objectui#11073)Triage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015) ·
session_01AavokzJ5DndAwitDXvKy4U· 2026-09-30T04:39Z. ⛔ Not a claim, ⛔ not a dispatch. The grade, route and ruling are unchanged.isRefusedTextComparandis reachable from the installed@objectstack/spec.@objectstack/*@17.5.0from npm (the version objectui'spnpm-lock.yamlnow resolves; the spec tag commit is objectstack0f6dcac5e9): it is exported from@objectstack/spec/data.pm:queue. The dispatching seat re-reads the body against objectuimainat claim. The probe above licenses the work; it does not replace that read.objectstack-fleet commentedon Sep 30, 2026
Claim: PM loop round 9 —
domain:uiexecution seat 2Session:
session_011p7ikEivgXefNDaE5S5UecAccount:
os-support-ai(the seat's linked user asGET /useranswers it; the card's assignee)Branch:
claude/issue-9048-array-entry-text-comparandWorktree:
objectui-issue-9048Domain:
domain:uiSeat:
domain:ui#2File surface (maintainer ruling D
5658597078, batch #133 item 1; unlocked by5904161615):packages/data-objectstack/src/index.ts:objectFilterEntryToAST's ARRAY branch (:518, reached throughtranslateFilterArray) reads the spec'sisRefusedTextComparandand seatstextComparandRefusalReasonin its ownMalformedFilterErrorenvelope.'@object-ui/data-objectstack': patch).mainand stay untouched:packages/core/src/utils/text-comparand.tsalready re-exports both from@objectstack/spec/data(:101, PR objectui#11086);viewFilterRuleToNode, was closed by PR objectui#9152 (06973aa7e).Stop on breach; explain in the report.
Container & model:
S,mode:subagent,model: opus(TIER_DEFAULT;dispatch-gateshas no objectui derivation, so there is no path-derived mandate; objectstack41dcf118)Clause-②: no
Why no: the adapter refuses a wire node that the contract's own
FILTER_TEXT_CASESdeclares REFUSED, and that the$dialect and the view-rule path already refuse. This is a door on the one dialect that lacked it, not a narrowing of a published accept set.Thread-read: 5904161615
Serial constraints cleared: read 2026-09-30T08:34Z at objectui
origin/maine978ed5ea.packages/data-objectstack/src/index.ts.coreneighbour) is closed.Generated by Claude Code
objectstack-fleet commentedon Sep 30, 2026
os-dev-report
{ "issue": 9048, "status": "done", "branch": "claude/issue-9048-array-entry-text-comparand", "pr": "https://github.com/objectstack-ai/objectui/pull/11192", "session": "session_011p7ikEivgXefNDaE5S5Uec — subagent of the PM session; identity is the branch named in PM claim 5907396601", "premise_still_valid": true, "summary": "Ruling D's open half. `objectFilterEntryToAST` (the rule-entry arm of `translateFilterArray`) now asks `@objectstack/spec/data`'s `isRefusedTextComparand` after `toAstFilterOperator`'s fold. On a refusal it throws `MalformedFilterError` (INVALID_FILTER/400) carrying `textComparandRefusalReason(field, ARRIVED_SPELLING, value)` verbatim, plus a tail naming the `$icontains` twin so `mustMention` holds, before any request is sent. The door sits in the translator itself, so both find() routes, exportDownload, aggregate()'s analytics lowering, nested rule groups and a bare rule spread under and/or all meet it. `translateFilterChild` gained an internal `index` arg so a bare child names its position. `MalformedFilterError` gained an OPTIONAL third ctor arg (a refusal sentence replacing the shape advice): an additive published .d.ts change, with the two-arg message byte-identical. No new export from @object-ui/core, and no manifest or lockfile change: data-objectstack already declares @objectstack/spec ^17.5.0 directly. Zone-2 premises re-measured at BASE e978ed5ea and all hold. core's text-comparand.ts re-exports both functions from spec/data (verified, not redone). The spec ROOT lacks them and /data has them. The only alias row folding to icontains is the identity row; case variants fold via toLowerCase. An absent `value` is refused, with no carve-out: this translator always emits a 3-tuple, so it went out as JSON null, a non-string comparand. ValueDataSource refuses the same rule, pinned by a verdict-for-verdict parity table over 9 comparands. PR assignee os-support-ai was set and read back; card labels and assignee were not touched.", "tests": "All runs from the worktree root, heavy ones via os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-9048; pass COUNTS read. FINAL HEAD d7bd5a111, tree clean. (1) `pnpm exec vitest run packages/data-objectstack/ scripts/__tests__/spec-symbol-ledger-data-objectstack-7265.test.ts`: 'Test Files 74 passed (74) / Tests 1017 passed (1017)', VERDICT command-exit 0, on d7bd5a111. (2) New pin file packages/data-objectstack/src/filter-entry-text-comparand-9048.test.ts: 52 passed (52). It covers the FILTER_TEXT_CASES-derived rows (code + mustMention asserted); '' and 42 refused with 'acme' as control; every alias-table spelling folding to icontains plus ICONTAINS/iContains, each with a control; the op key; contains-'' as SCOPE; object vs entry form refusing the same node with one reason; VDS parity over 9 comparands; index at depth (entry 1, nested group, bare child at position 2); aggregate(); BigInt via aggregate(). (3) `pnpm --filter @object-ui/data-objectstack type-check` exit 0 after `pnpm --workspace-concurrency=2 --filter '@object-ui/data-objectstack^...' build` (3 of 47 projects, exit 0). `tsc --noEmit --listFiles` includes the new test. `pnpm --filter @object-ui/data-objectstack build` exit 0. (4) Markdown leg (the diff adds a changeset .md): `node scripts/markdown-test-inputs.mjs --audit` exit 0. The whole root scripts/__tests__/ suite ran in 3 chunks whose union equals the 179-file git ls-tree population at HEAD: 61/61 files and 2586 tests; 58 passed + 2 skipped of 60 files and 1421 passed + 2 skipped tests (the 2 skips are the network-escape-8537 fixtures, inert by design, whose pin ran in the same chunk); 58/58 files and 1364 tests. Each chunk printed VERDICT command-exit 0. (5) ABLATION on committed HEAD 75db0a505, direction predicted before running as 37 red / 15 green. `node /home/user/objectstack/scripts/ablation-replace.mjs` WRAP mode, plus a shell trap on EXIT INT TERM with absolute paths. The anchor (the guard's if-condition) went x1 to x0 and the marker x0 to x1; blob d4ad4247 to 92aaf4e2. Result: 'Tests 37 failed | 15 passed (52)'; none of the 37 is a control, scope or derivation pin. Restore proven by state: blob back to d4ad4247 == HEAD blob, `git diff HEAD` 0 bytes, marker count 0. The subject is imported relatively (./index), so it resolves to source and no dist leg applies. (6) LINT, a declared narrowing: population is the root flat config's `**/*.{ts,tsx}` block (the package lint is `eslint .`), and the diff has exactly 2 such files. `eslint --format json` over both at d7bd5a111: 2 files, 0 errors; new test 0 warnings; index.ts 123 no-explicit-any, equal to BASE index.ts linted via stdin (123). Invariance: tseslint.configs.recommended only (no type-checked variant, no parserOptions.project or projectService), and a grep for fs reads over eslint-rules/*.js hits 0 rule files while the positive control hits. (7) Gates exit 0 (redirect first, exit captured before any pipe): check-changeset-presence ('2 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s)'), check-changeset-no-major, -fixed, -overwrite, -claims, check-pending-changeset-literals, check:new-line-citations ('0 new citation(s)'), check:control-bytes, check:spec-symbols, check:phantom-deps, check:unused-deps, check:self-import, check:test-path-roots. (8) NOT MEASURED, reason: prerequisite not met. check:readme-exports exit 1 = 480 self-imports unjudged because 33 sibling packages are unbuilt; data-objectstack's own README rows show 0 unjudged, 0 wrong-path, 0 fabricated. check:spec-floors exit 1 = 17 findings, all [no-artifact] on unbuilt siblings; data-objectstack was inspected against the fetched 17.5.0 floor with no finding. (9) CI snapshot at d7bd5a111 when reporting: 42 check runs, 21 success, 3 skipped, 18 in_progress, 0 failure. Not waited on, per contract.", "gates": { "vitest packages/data-objectstack/ + spec-symbol-ledger-data-objectstack-7265": "PASS 74 files / 1017 tests @ d7bd5a111", "vitest scripts/__tests__/ (3 chunks, union = 179 files)": "PASS 177 files + 2 by-design skips; 5371 tests + 2 skipped", "type-check @object-ui/data-objectstack": "PASS exit 0 (closure built first)", "build @object-ui/data-objectstack": "PASS exit 0", "eslint (2 changed ts files)": "PASS 0 errors; declared narrowing with population, count and invariance", "check-changeset-presence": "PASS", "check-changeset-no-major / fixed / overwrite / claims / pending-changeset-literals": "PASS", "check:new-line-citations": "PASS 0 new", "check:control-bytes": "PASS", "check:spec-symbols": "PASS", "check:phantom-deps / unused-deps / self-import / test-path-roots": "PASS", "markdown-test-inputs --audit": "PASS", "check:readme-exports": "NOT MEASURED: prerequisite (unbuilt siblings); own package clean", "check:spec-floors": "NOT MEASURED: prerequisite (unbuilt siblings); own package inspected, no finding", "ablation": "PASS 37 red / 15 green as predicted; restore proven by state", "CI": "in_progress (21 success / 3 skipped / 18 in_progress / 0 failure at report time)" }, "files_changed": [ "packages/data-objectstack/src/index.ts (+92 -11 net across the branch)", "packages/data-objectstack/src/filter-entry-text-comparand-9048.test.ts (new, 400 lines)", ".changeset/9048-entry-form-icontains-comparand-refusal.md (new, @object-ui/data-objectstack: patch)" ], "line_budget": "n/a: no skills/** or governed surface touched (branch diff 3 files, +522 -11)", "deviations": [ "BigInt pin driven through aggregate(), not find(): measured that find() throws a bare TypeError from its request-coalescing key stableStringify(params) before any translation, for any operator (a contains control throws the same). Recorded in Acceptance notes, not fixed.", "scripts/__tests__/ run as 3 chunks to fit the foreground cap; the union equals the population (diffed against git ls-tree), so this is not a narrowing.", "Two lock acquisitions hit queue-timeout (exit 99) behind sibling holder issue-9868; both were retried with the same slot and landed. The first final-HEAD package run was 99 and its retry is the cited reading.", "Attribution follows objectui AGENTS.md and the role file, not the harness reminder: commit trailers are the model-free pair (Claude-Session + Co-authored-by: Claude), and the PR footer is the session-URL form." ], "mcp_calls": "0 — none.", "api_writes": "3 REST writes, all through the fleet relay as objectstack-fleet[bot]: (1) POST /repos/objectstack-ai/objectstack/dispatches, pr_create, which the relay executed as POST /repos/objectstack-ai/objectui/pulls (draft #11192); (2) POST /repos/objectstack-ai/objectstack/dispatches via label-write.mjs --assign os-support-ai, executed as the PR #11192 assignee add and read back MATCHES; (3) this os-dev-report comment via post-stamped.mjs on issue 9048. `git push` (4 pushes of the branch) is not a REST write. Reads used REST GET only.", "open_questions": [], "out_of_scope_findings": [ "carrier: 承接者:无 · noted, not filed · an AST tuple handed to the adapter as $filter (['name','icontains','']) still reaches the wire unjudged, before and after this change: translateFilterArray deliberately leaves an already-lowered comparison tuple for the server, while ValueDataSource's AST arm refuses the same tuple. The in-tree AST producers (convertFiltersToAST, toFilterNode) refuse the shape first, and no reachable producer of the raw tuple was demonstrated · dedupe words: icontains AST tuple passthrough, translateFilterArray comparison tuple, data-objectstack text comparand", "carrier: 承接者:无 · noted, not filed · ObjectStackAdapter.find() throws a bare TypeError ('Do not know how to serialize a BigInt') from its request-coalescing key stableStringify(params) when any param holds a BigInt, before any filter translation and for any operator. No JSON-sourced filter can carry a BigInt, and no named producer · dedupe words: stableStringify BigInt, find coalescing key TypeError, data-objectstack BigInt filter", "carrier: 承接者:无 · noted, not filed · packages/data-objectstack/README.md 'Rule-shaped arrays' stays true but does not yet mention that an icontains entry with an empty or non-string value raises MalformedFilterError (AGENTS.md #2, docs-driven). It is outside the claim's declared file surface, so it was left for the seat · dedupe words: data-objectstack README MalformedFilterError icontains, rule-shaped arrays docs" ] }Generated by Claude Code
objectstack-fleet commentedon Sep 30, 2026
os-dev-report
{ "issue": 9048, "status": "done", "patch_round": 1, "answers": "contract review 5908617443, findings 1-2", "branch": "claude/issue-9048-array-entry-text-comparand", "pr": "https://github.com/objectstack-ai/objectui/pull/11192", "session": "session_011p7ikEivgXefNDaE5S5Uec — subagent of the PM session; identity is the branch named in PM claim 5907396601", "premise_still_valid": true, "summary": "Finding 1 closed by the seat's option B, keeping 'patch' and Clause-② no. `MalformedFilterError` is back to its BASE constructor `(entry: unknown, index: number)` and its BASE JSDoc. The refusal is now an ordinary `MalformedFilterError` built by the module-private `refuseTextComparandEntry`: a factory, not a subclass. It constructs the class and replaces `message` with the refusal sentence, so instanceof, `name`, `code` 'INVALID_FILTER' and `httpStatus` 400 are the class's own, and the V8 stack header carries the refusal (pinned). The review's hazard is closed inside the module: the base constructor's entry serialisation goes through a module-private `describeFilterEntry` (a BigInt replacer plus a try/catch for cycles). For every entry that serialised before, its output is byte-identical to `JSON.stringify(entry) ?? String(entry)`: a one-off check over 17 samples read 17/17 identical, and the bare template throws on the BigInt entry as a control. Built `.d.ts` AND `.d.cts` of @object-ui/data-objectstack at head 26de698a1 are byte-identical to BASE e978ed5ea: same sha256 prefix e5efb7ede92ec5c7, 204591 bytes each, 0 diff lines. The `MalformedFilterError` class-block diff is EMPTY. The same instrument over pre-patch head d7bd5a111 shows `constructor(entry: unknown, index: number, refusal?: string)`, so the comparison is lit. The changeset's paragraph on the third argument was rewritten to match. Finding 2 closed. packages/data-objectstack/README.md 'Rule-shaped arrays' gains a paragraph mirroring the `$icontains` one: an `icontains` rule whose value is empty, a number, null or missing throws `MalformedFilterError` (INVALID_FILTER / 400) before any request, with the spec's reason, naming the operator spelling that arrived beside `$icontains`, at any depth and on find() and aggregate(). The `MalformedFilterError` export-listing comment is widened the same way (4 comment-only lines inside the fence). Fast-forward push d7bd5a111..26de698a1, no force-push, no PR body PATCH; the PR body is now stale, and the edits the seat owes are listed under pr_body_edits_needed.", "tests": "All on head 26de698a1, tree clean and equal to the remote; heavy runs via os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-9048. (1) `pnpm exec vitest run packages/data-objectstack/ scripts/__tests__/spec-symbol-ledger-data-objectstack-7265.test.ts`: 'Test Files 74 passed (74) / Tests 1019 passed (1019)', VERDICT command-exit 0. The 1017 to 1019 step is the 2 new pins: MalformedFilterError.length === 2, and the BigInt shape refusal is built without throwing. The refusal helper additionally asserts name 'MalformedFilterError' and a stack header carrying the refusal but not the discarded shape advice. The pin file is now 54 tests. (2) `pnpm --filter @object-ui/data-objectstack type-check`: exit 0 (tsc --noEmit ran; closure built). (3) .d.ts proof, under the lock with a trap on EXIT INT TERM and absolute paths. HEAD build, then BASE index.ts swapped in (blob 2f154326 landed), built, restored (disk blob 8d6d97dc == HEAD, git diff HEAD 0 bytes), and rebuilt (HEAD .d.ts equals the first HEAD build). index.d.ts and index.d.cts: head == base, 0 diff lines. Control leg with d7bd5a111's index.ts (blob 5e52819e): the class line differs (refusal?: string); restored and proven the same way. (4) ABLATION on committed head 26de698a1, both predicted before running, via ablation-replace.mjs WRAP plus a shell trap. Leg A, the guard's if-condition disabled (anchor x1 to x0, blob 8d6d97dc to 8495aee7): 'Tests 37 failed | 17 passed (54)' as predicted; none of the 37 is a control, scope, derivation or published-class pin. Leg B, the constructor's describeFilterEntry call replaced by a bare `JSON.stringify(entry) ?? String(entry)` (blob 8d6d97dc to 883168f5): 'Tests 2 failed | 52 passed (54)' as predicted — exactly 'a BigInt comparand gets the contract reason, not a serialisation TypeError' and 'the shape refusal is built without throwing for an entry carrying a BigInt'. Both restores were proven by the tool and by state: blob == HEAD 8d6d97dc, git diff HEAD 0, markers 0. (5) Markdown leg (README and changeset edited): `node scripts/markdown-test-inputs.mjs --audit` exit 0 ('107 candidate test files, all adjudicated'). Whole root scripts/__tests__/ ran in the same 3 chunks; the population is unchanged at 179 files (diffed against git ls-tree at head). Results: 61/61 files and 2586 tests; 58 passed + 2 by-design skips of 60 files and 1421 passed + 2 skipped tests; 58/58 files and 1364 tests. All printed VERDICT command-exit 0. (6) eslint --format json on the two changed .ts files at 26de698a1: 0 errors; test 0 warnings; index.ts 123 (the BASE count). The README is outside eslint's population; the invariance argument is unchanged from round 0. (7) Gates exit 0: check-changeset-presence, -no-major, -fixed, -overwrite, check:changeset-claims, check:pending-changeset-literals, check:new-line-citations, check:control-bytes, check:spec-symbols, check:phantom-deps, check:unused-deps, check:self-import, check:test-path-roots, check:doc-fences, check:doc-example-ids, check:doc-types, check:doc-example-readers, check:shell-escape-residue. (8) NOT MEASURED, reason: precondition. check:doc-snippets and check:doc-examples exit 2 (PRECONDITION NOT MET: 34 packages unbuilt; they 'did not run'). This round's in-fence README change is 4 added comment-only lines plus the comment line they extend, which cannot move a type-check verdict. CI's 'Lint'/'Build & E2E'/README checks own the full reading. check:readme-exports exit 1 is again unbuilt siblings only: 0 data-objectstack rows unjudged, 553 judged, 0 wrong-path, 0 fabricated. (9) CI at 26de698a1 when reporting: 43 check runs, 40 success, 3 skipped, 0 failure, 0 in progress.", "gates": { "vitest packages/data-objectstack/ + spec-symbol-ledger-data-objectstack-7265": "PASS 74 files / 1019 tests @ 26de698a1", "vitest scripts/__tests__/ (3 chunks, union = 179 files)": "PASS 177 files + 2 by-design skips; 5371 tests + 2 skipped @ 26de698a1", "type-check @object-ui/data-objectstack": "PASS exit 0", "published .d.ts / .d.cts vs BASE e978ed5ea": "IDENTICAL (0 diff lines; MalformedFilterError class block diff empty); control d7bd5a111 differs", "ablation leg A (guard)": "PASS 37 red / 17 green as predicted", "ablation leg B (BigInt-safe describer)": "PASS 2 red / 52 green as predicted", "eslint (2 changed ts files)": "PASS 0 errors", "markdown-test-inputs --audit": "PASS", "changeset presence / no-major / fixed / overwrite / claims / pending-literals": "PASS", "check:new-line-citations / control-bytes / spec-symbols / phantom-deps / unused-deps / self-import / test-path-roots": "PASS", "check:doc-fences / doc-example-ids / doc-types / doc-example-readers / shell-escape-residue": "PASS", "check:doc-snippets / doc-examples": "NOT MEASURED: precondition (unbuilt packages); in-fence change is comment-only", "check:readme-exports": "NOT MEASURED: prerequisite (unbuilt siblings); own package rows clean", "CI": "success (40 success / 3 skipped / 0 failure @ 26de698a1)" }, "files_changed": [ "packages/data-objectstack/src/index.ts (round +41 -18; branch +112 -8)", "packages/data-objectstack/src/filter-entry-text-comparand-9048.test.ts (round +30 -4; branch new, 426 lines)", ".changeset/9048-entry-form-icontains-comparand-refusal.md (round +6 -4; still @object-ui/data-objectstack: patch)", "packages/data-objectstack/README.md (round +17 -1, new this round)" ], "line_budget": "n/a: no skills/** or governed surface touched (branch diff 4 files, +587 -9)", "deviations": [ "The docs commit 26de698a1's message says 'the export-listing comment and the INVALID_FILTER code line are widened to match'. The INVALID_FILTER code line was NOT edited: it already names MalformedFilterError. History was not rewritten (no force-push), so drop that clause from the squash message at merge.", "Factory rather than subclass: both were allowed by the seat. The factory keeps the prototype chain exactly MalformedFilterError, and nothing new can surface in the .d.ts.", "describeFilterEntry also makes the TWO-ARGUMENT shape refusal BigInt- and cycle-safe (a pre-existing TypeError path). Its text is byte-identical for every entry that serialised before (17/17 one-off), and the changeset says so.", "scripts/__tests__/ run as 3 chunks whose union equals the population, as in round 0." ], "pr_body_edits_needed": [ "'What this lands', the MalformedFilterError bullet: replace with 'MalformedFilterError's published signature is unchanged (built .d.ts/.d.cts byte-identical to BASE); the refusal is an ordinary MalformedFilterError built by the module-private refuseTextComparandEntry, with its message replaced by the refusal sentence; the constructor's entry serialisation is BigInt-safe through a module-private describer.'", "'Tests': cite head 26de698a1: 74/1019; pin file 54; ablation leg A 37 red / 17 green and leg B 2 red / 52 green; the .d.ts identity proof with its d7bd5a111 control.", "'Acceptance notes': drop the README bullet (landed this round as review finding 2)." ], "mcp_calls": "0 — none.", "api_writes": "This round, 1 REST write through the fleet relay as objectstack-fleet[bot]: POST /repos/objectstack-ai/objectstack/dispatches via post-stamped.mjs, executed as this os-dev-report comment on objectstack-ai/objectui issue 9048. One fast-forward `git push` (not a REST write). No PR body PATCH, no label or assignee write. Reads used REST GET only (review comment 5908617443, PR 11192, check-runs).", "open_questions": [], "out_of_scope_findings": [ "carrier: 承接者:无 · noted, not filed, carried from round 0 and answered not-owed by review 5908617443 · an AST tuple handed as $filter passes unjudged (the adapter leaves the server to decide) · dedupe words: icontains AST tuple passthrough, translateFilterArray comparison tuple", "carrier: 承接者:无 · noted, not filed, carried from round 0 and answered not-owed by review 5908617443 · find()'s request-coalescing key stableStringify(params) throws a bare TypeError on a BigInt param, before translation · dedupe words: stableStringify BigInt, find coalescing key TypeError" ] }Generated by Claude Code
objectstack-fleet commentedon Sep 30, 2026
✅ ACCEPT: PR objectui#11192 at
26de698a1. Ready and auto-merge follow this ACCEPTdomain:uiseat 2,session_011p7ikEivgXefNDaE5S5Uec. Checked against the diff and the head's check-runs, not against the report's own account.5907396601, on maintainer ruling D5658597078, unlocked by5904161615.5908407005, and patch round5909283039.5908617443: FAIL. A published-signature growth did not fitpatch, and the README was owed.5909407701, atCONTRACT_REVIEW_TIER: PASS on this head, which is the landing head.objectFilterEntryToAST, the rule-entry arm oftranslateFilterArray, asks@objectstack/spec/data'sisRefusedTextComparandafter the operator fold. On a refusal it throwsMalformedFilterError(INVALID_FILTER/ 400) carryingtextComparandRefusalReason(...)verbatim, plus a tail naming the$icontainstwin (somustMentionholds), before any request. The door is in the translator, so bothfind()routes,exportDownload,aggregate(), nested groups and a bare rule spread all meet it. ⛔ No new export from@object-ui/coreMalformedFilterError's published constructor is its base(entry, index). The built.d.ts/.d.ctsare byte-identical to BASEe978ed5ea, and a control ond7bd5a111shows the instrument sees a difference. The refusal comes from a module-private factory, and the constructor's serialisation goes through a module-private BigInt- and cycle-safe describer$dialectValueDataSourceover 9 comparands. Ablation A: 37 red / 17 green. Ablation B: 2 red / 52 green.data-objectstack: 74 files, 1019 tests. Rootscripts/__tests__green'@object-ui/data-objectstack': patch;Clause-②: no26de698a1: 43 runs, 40 success, 3 expected skips, 0 failureFixes #9048The PR body was corrected before landing. objectui squash-merges with the PR body, so a correction that lives only in a comment would never reach
main's history (review5909407701③). The seat patched the body through the fleet relay'sissue_patch, using the relay's post-objectstack-ai/objectstack#20571 read-back: 11 299 bytes sent, 11 299 stored, identical. Three changes:MalformedFilterErrorbullet now states the unchanged signature;Out of scope, one line each:
$filterpasses to the server unjudged. No producer of a raw refused tuple was shown: Acceptance notes.find()'s request-coalescing key throws a bareTypeErroron a BigInt param, before translation. Pre-existing, with no JSON-sourced producer: Acceptance notes.toFilterNodethrows surface) is untouched, per the ruling.domain:uiseat 2 · ACCEPT · 2026-09-30T10:32ZGenerated by Claude Code
objectstack-fleet commentedon Sep 30, 2026
Closed
completed: PR objectui#11192 landed as0b8c63831, verified by contentdomain:uiseat 2,session_011p7ikEivgXefNDaE5S5Uec. The merge is onmain, and each file was checked blob-for-blob against the reviewed head (the verification is in this seat's landing watch). The PR's closing line isFixes #9048, but the merge-queue merge did not close this card within 3 minutes, so the seat closes it here, with the same meaning.pm:dispatchedwas removed when the landing was verified.domain:uiseat 2 · close · 2026-09-30T10:54ZGenerated by Claude Code
fix(data-objectstack): the entry-form filter refuses an empty or non-…