finding(fields): BooleanField's readonly branch reads the value by truthiness — the string 'false' and {} say "Yes" #8593

Activity

1. os-justin

   added

   bugSomething isn't workingSomething isn't working

   pm:dispatched

   priority:p2

   domain:uiobjectui ui stream: fix lands on the published library or apps — objectui execution seatobjectui ui stream: fix lands on the published library or apps — objectui execution seat

   package: fields

   on Sep 8, 2026
2. os-justin

   self-assigned this

   on Sep 8, 2026
3. os-justin commented on Sep 8, 2026

   

   os-justin

   on Sep 8, 2026 – with Claude

   CollaboratorAuthor

   More actions

   Claim: session session_01YBWFb5YgMU5dw8p2VKj16S · branch claude/issue-8593-booleanfield-readonly-truthiness

   Claimed by the domain:ui PM seat on behalf of the dev it is dispatching. The dev inherits this claim and the assignee; it posts no second claim and does not touch the assignee field.

   ⛔ The direction is already ruled — do NOT re-derive it

   This is objectui#8582's defect one surface over: BooleanField's readonly branch reads the value by truthiness, so 'false', '0', {} and [] all render Yes.

   objectui#8582 / PR #8594 established, from @objectstack/spec on four independent instruments, that a boolean field holds a JS boolean and nothing else:

   • valueSchemaFor's boolean / toggle arm is a bare z.boolean() (data/field-value.zod.ts)
   • BOOLEAN_VALUE_TYPES — "a JS boolean on the wire (driver read-coercion repairs SQL 0/1)"
   • NON_TEXT_STORED_VALUE_TYPES — the stored value is "never text on any backend"
   • the truth table already lives at the producer boundaries: objectql's coerceBooleanFields plus its invalid_boolean write refusal, and rest's parseBooleanCell over the spec's IMPORT_BOOLEAN_TRUE_TOKENS / IMPORT_BOOLEAN_FALSE_TOKENS

   ⇒ A non-boolean is no value. A renderer- or widget-side truth table would be the AGENTS.md #0.1 second dialect. Read PR #8594 and follow it; do not open the question again.

   What is actually yours to decide

   Only one thing: what a readonly BooleanField draws for no value. The cell renderer's answer is the shared EmptyValue, which is this file's own convention for an unreadable value (cf. DateTimeCellRenderer). Whether a widget in readonly mode should match that, or whether widgets have their own established empty affordance, is a real question — look at what the sibling widgets in packages/fields/src/widgets/ do for an absent value and follow the local convention rather than importing the cell renderer's by default. Say which you found.

   ⚠️ Check the second direction too

   objectui#8582 found the card's own description incomplete: besides fabricated truth ('false' → Yes), there is fabricated falsehood — 0 and '' rendered a negative the record never made, and on a status-named field that was a destructive-styled badge, which reads louder than the positive. Check whether the readonly widget has an equivalent, and sweep both directions or say why not.

   ⭐ Evidence bar

   The caricature is a widget that renders "no value" for anything but a literal true — it satisfies "'false' no longer says Yes" while destroying the populated cases. Your load-bearing assertions are the populated ones: a real true still says Yes, a real false still says No.

   Run the caricature, do not predict it. Read failure modes, not counts. ⚠️ Confirm each discriminating assertion actually runs in the leg it was written for.

   ⭐ Steal objectui#8582's LEG D. That dev deliberately reproduced objectui#8490's accident — an empty source file — as a leg, and measured 0 AssertionError lines against 37 harness-death signatures, proving its counter could tell the two apart. Do the same: prove your instrument distinguishes "the harness died" from "the assertions failed" before you rely on any red count. Prove every mutation on disk in both directions plus a line-total gate.

   Hazards

   • ⚠️ packages/fields is busy: PR fix(fields): only a real boolean is a value of a boolean cell — a non-boolean scalar drew a checked box #8594 is armed, objectui#8580 is in flight on src/index.tsx (markdown/html), and PR fix(fields): an empty array is not a cell value, and eight renderers fabricated one #8586 landed as 21529629c. Your file is src/widgets/BooleanField.tsx, which none of them touches — but bring main in rather than rebasing: a pushed branch must not be force-pushed, and objectui#8582 hit exactly this.
   • ⚠️ Cite commit shas, not PR numbers, when referring to history.
   • Run from the repo root with paths, nothing after a bare --. pnpm --filter PKG exec vitest run FILE is refused by the guard (objectui#3378). Never --no-inline-config.
   • ⚠️ queryByText throws on multiple matches too — use queryAllByText(...).length or scope first.
   • ⚠️ type-check only after building the closure, both programs.
   • ⚠️ A non-zero exit can mean a wrong script name: pnpm run check:changeset-fixed → ERR_PNPM_NO_SCRIPT.
   • ⚠️ skip-changeset is a phantom label. Write a real changeset.
   • ⚠️ check:control-bytes does not cover zero-width spaces inside string literals; use a Python scanner with a lit control. (My grep -naP '[\x{200b}…]' warning is single-sourced and unconfirmed since — the Python scanner is cheaper than settling it.)
   • ⚠️ The zero-quota single-card payload channel delivers issue bodies fine, but its embedded timeline is capped at ~15 items — reading a specific comment on a commented card costs one MCP call. Budget for it.
   • Restore by state after every ablation with a trap on EXIT INT TERM and absolute paths.

   ---

   Generated by Claude Code
4. os-justin commented on Sep 8, 2026

   

   os-justin

   on Sep 8, 2026 – with Claude

   CollaboratorAuthor

   More actions

   os-dev-report

   {
     "issue": 8593,
     "status": "done",
     "branch": "claude/issue-8593-booleanfield-readonly-truthiness",
     "pr": "https://github.com/objectstack-ai/objectui/pull/8597",
     "premise_still_valid": true,
     "summary": "Premise verified on 21529629c: BooleanField.tsx line 54 read the value by truthiness exactly as the card measured, so 'false', '0', {} and [] said Yes, and the second direction is present too: 0, '', null and undefined said No (no louder than the positive: one text span, no field-name badge). The readonly branch now uses the 09fab31d1 guard (typeof value !== 'boolean') and draws EmptyValue for a non-boolean; a real true still says Yes, a real false still says No. EmptyValue was chosen because it is the widgets' own convention, not imported from the cell renderer: 19 sibling readonly branches in packages/fields/src/widgets/ render it for an absent value (NumberField, CurrencyField, PercentField, DateTimeField, EmailField, PhoneField, RadioField, ObjectField, ...) and validation-feedback.test.tsx pins the dash for six of them; the exceptions are domain-shaped (MasterDetailField's 'No related records', PasswordField's mask). Added a 33-case test file (THE DEFECT, THE OTHER DIRECTION, THE RULING, POPULATED across boolean/toggle and switch/checkbox variants, THE HOST PATH through FieldEditWidget), a Read-only Display docs section (prose only), and a patch changeset. Draft PR #8597 whose first line names this card with the closing keyword, not flipped out of draft, no auto-merge, no labels written (skip-changeset is a phantom here). Assignee untouched; the newest Claim: names this branch. origin/main moved one unrelated commit (d02942b0e, packages/types tests) after the cut; no overlap, so main was not merged in and the PR merges clean from 21529629c. PR #8594's boolean.mdx hunk is the Cell Renderer section; mine is a separate section above Use Cases.",
     "tests": "All on c303b681d, heavy runs under os-verify-lock (VERDICT command-exit 0 each). (1) pnpm exec vitest run on the new file + standard-widgets + validation-feedback + readonly-host-plumbing-e2e: 4 files, 201 passed. (2) pnpm exec vitest run packages/fields/: 141 files, 2432 passed. (3) packages/plugin-form/ + packages/plugin-detail/: 228 files, 2129 passed, 1 skipped. (4) pnpm --filter @object-ui/fields type-check after building '@object-ui/fields^...': both programs (tsc --noEmit && tsc -p tsconfig.test.json) green, and tsc -p tsconfig.test.json --listFiles lists the new test (1 hit). (5) eslint narrowed to the two .tsx files, --format json: 2 files, 0 errors, 1 pre-existing no-explicit-any warning at BooleanField.tsx:11; population = eslint.config.js files '**/*.{ts,tsx}' minus ignores over 4397 tracked ts/tsx; invariance = no parserOptions.project/projectService (0 hits, control 'files:' 7) and ratchet rules are per-file, so untouched files' verdicts cannot move. (6) Gates green: check-control-bytes (6783 files), check-changeset-no-major / -presence (2 source files, 1 changeset) / -overwrite / -fixed, check-doc-fence-languages, check-doc-component-types, check-doc-links, governed-queue-guard --test = NOT GOVERNED, Python zero-width/control scan of the 4 changed files with a lit control = 0 hits. NOT MEASURED, declared: check-doc-snippet-types exits 2 (could not run, eight dist/ absent) and the docs diff adds 0 fences (control 4 in file); repo-wide pnpm lint, pnpm check and the other 19 turbo-affected packages are CI's (git grep: no test outside packages/fields pins the readonly Yes/No text). ABLATION on packages/fields/src/widgets/BooleanField.tsx (136 lines, HEAD blob 035cdf27c); source resolves by relative import and vitest aliases @object-ui/components to src (vitest.config.mts:505), so no dist/ on the path and no rebuild between legs; each leg proved on disk (removed-text count, injected-text count, line total, blob hash), restored with git checkout HEAD -- ABSOLUTE_PATH under trap EXIT INT TERM, restore proved by state (git diff HEAD empty AND hash-object == HEAD blob) all 5 times; instrument = per-test classification from vitest's JSON reporter. LEG 0 HEAD: 33 passed. LEG A the 21529629c blob (c8325fff6, 108 lines, hash == BASE blob): 23 failed / 10 passed = 12 'fabricated true' (10 THE DEFECT + 2 HOST PATH), 8 'fabricated false', 3 'coercion the spec assigns', 0 deaths; all 8 POPULATED + 2 HOST PATH false green. LEG B caricature value !== true (guard 0, marker 1, lines 136): 6 failed / 27 passed, all six 'a real false must still say No' (4 POPULATED + 2 HOST PATH); every DEFECT/OTHER/RULING case green, observed not predicted. LEG C bare span: 33 failed = 23 'affordance must be present' + 6 'must still say No' + 4 'must still say Yes', 0 'fabricated' sentences (locators see only the widget's output). LEG D empty file (0 bytes, hash e69de29bb): 33 failed = 29 harness deaths ('Element type is invalid') vs 4 assertions (HOST PATH cases, where FieldEditWidget renders a fallback for an undefined widget). Instrument correction: a first grep -c run over the text reporter with the brief's four death signatures read LEG D as 0 deaths / 4 AssertionErrors, because vitest prints one block per distinct message; that run's counts are not relied on.",
     "mcp_calls": "4 — create_pull_request, pull_request_read (PR body read-back), add_issue_comment (this report), issue_read get_comments (its read-back); all reads of the card, the brief and PR #8594 went through the zero-quota payload channel and git (the 8582 branch fetched by ref); repo-scoped REST probed once and answered 403.",
     "open_questions": [],
     "out_of_scope_findings": [
       "not filed, handed to the PM: BooleanField's EDIT branch still maps a non-boolean onto the control by !!value (the string 'false' renders as an ON switch / checked box; it does write a real boolean back on toggle). The card body already records it as the lesser half; an editable Switch has no 'no value' state, so the right rendering (indeterminate checkbox, refuse, or coerce-and-write-back) is a decision rather than a sweep, and filing it would have spent a write on the rate-limited MCP budget for a card whose author already logged the observation."
     ],
     "contradicts_the_brief": [
       "The brief's evidence recipe counts harness deaths by the signatures 'Cannot read properties of null', 'Unable to find', 'TypeError', 'is not a function': all four count 0 on this repo's LEG D. React's death for an empty module is 'Element type is invalid', and vitest's text reporter prints one block per DISTINCT message, so a grep -c instrument read 29 dead tests as 0 deaths / 4 AssertionErrors. Per-test classification from vitest's JSON reporter is the instrument that told 29 from 4; a LEG D that relies on grep -c over the text reporter can pass while blind.",
       "The brief budgets one MCP call to read the brief comment because the payload timeline is capped at ~15 items: this card had 8 timeline items with hasNextPage false, so the brief (comment 5586553497) was read at zero quota. The cap is real; it did not bite here.",
       "The card's table lists 0 and '' under 'No' and stops; null and undefined went through the same ternary and also said No. Swept in as the absence rows (THE OTHER DIRECTION), since an absent value saying No is the same fabrication."
     ]
   }

   ---

   Generated by Claude Code
5. os-justin commented on Sep 8, 2026

   

   os-justin

   on Sep 8, 2026 – with Claude

   CollaboratorAuthor

   More actions

   Contract review — accepted. PR #8597 is out of draft and armed.

   ⭐ You found that the instrument I prescribed to detect a dead instrument was itself dead

   I have been putting objectui#8582's LEG D into every brief: reproduce the empty-file accident, count harness-death signatures separately, prove your counter can tell "the harness died" from "the assertions failed".

   You ran it and measured:

   All four signatures — Cannot read properties of null, Unable to find, TypeError, is not a function — count 0 on this repo's LEG D. React's death for an empty module is Element type is invalid. And vitest's text reporter prints one block per distinct message, so a grep -c instrument read 29 dead tests as 0 deaths / 4 AssertionErrors.

   A leg that exists to prove your counter works, passing while blind. That is the purest form of the failure this whole week has been about, and it was inside the check I was propagating as the remedy.

   objectui#8452 independently measured the same break with a different death mode — a plain module gives Failed Suites / No test suite found in file / Tests no tests, and my alternation read 0 on all four legs including D, indistinguishable from a clean run.

   Two seats, two modes, one conclusion: the shape of the check transfers; the strings do not. The template now prescribes per-test classification from vitest's JSON reporter, and if an alternation is used at all it must be shown to read non-zero on D and zero on a real red — which is what you did.

   ⭐ The local-convention question, answered properly

   I left you exactly one thing to decide and told you to follow the widgets' own convention rather than importing the cell renderer's. You measured it:

   19 sibling readonly branches in packages/fields/src/widgets/ render EmptyValue for an absent value — NumberField, CurrencyField, PercentField, DateTimeField, EmailField, PhoneField, RadioField, ObjectField… — and validation-feedback.test.tsx pins the dash for six of them. The exceptions are domain-shaped: MasterDetailField's "No related records", PasswordField's mask.

   So EmptyValue is the widgets' convention and the cell renderer's, and you can say which reason you followed. Naming the exceptions as domain-shaped rather than as counterexamples is what makes the reading trustworthy.

   ⚠️ Two corrections against the card and me

   • The card's table stops at 0 and ''. null and undefined go through the same ternary and also said No. An absent value asserting No is the same fabrication as 'false' asserting Yes, and sweeping them in as THE OTHER DIRECTION is right.
   • My channel warning did not bite here — 8 timeline items, hasNextPage: false, brief read at zero quota. That is the second card in a row where the ~15-item cap was real but inapplicable. I state the condition now, not the conclusion.

   Your unfiled observation — agreed, and the reasoning is the useful part

   BooleanField's edit branch still maps a non-boolean onto the control by !!value. You did not file it because an editable Switch has no "no value" state, so the right answer — indeterminate checkbox, refuse, or coerce-and-write-back — is a decision rather than a sweep, and the card's author had already logged the observation.

   That is the correct call on both counts. A card whose deliverable is "make a decision nobody has framed" is worth less than the sentence you just wrote, and spending a rate-limited write to restate an existing observation would have been the wrong trade. It is recorded here where the next person on this file will meet it.

   ---

   Generated by Claude Code
6. os-justin

   closed this as completedin #8597on Sep 8, 2026