Repository navigation
metadata designer: a report's or dataset's Name drops every underscore while it is typed (qavrf_rev reads qavrfrev), so a namespaced report or dataset cannot be named by typing #12746
Description
Filing gate ①: a user-visible authoring defect with a named landing point. reach: a public entry, the console's metadata designer in create mode, measured in Chromium against a live backend.
Reader who acts: objectui triage. The landing is the Name field of packages/app-shell/src/views/metadata-admin/inspectors/ReportDefaultInspector.tsx and DatasetDefaultInspector.tsx. Filed by domain:ui seat 3 (session_01Vk15bz8xHtKGUfXfCm16h4) from the os-dev report on objectui#12680 (comment 6221063367, PR #12745). That PR does not change this. ⛔ Not a claim.
What happens
In the report and dataset designers' create mode, an author cannot type an underscore into Name. Typing qavrf_rev key by key leaves qavrfrev, because each keystroke's value is normalized and a trailing _ is stripped before the next key arrives. So a namespaced report or dataset cannot get its namespace prefix (qavrf_…) by typing. The object designer's Name, measured the same way, keeps qavrf_rev.
Measured
- Backend: objectstack
66d456d01examples/app-crm,objectstack dev --seed-admin --fresh. The console was the objectuiclaude/issue-12680-object-name-survives-labelbranch on Vite, the same Name handlers asmaina2f38ddfor these two inspectors. Chromium, real keystrokes. - A writable package created with
POST /api/v1/packages {"manifest":{"id":"com.qa.vrf",…,"namespace":"qavrf"}}→ 201. /apps/crm_app/metadata/report/new?package=com.qa.vrfand/apps/crm_app/metadata/dataset/new?package=com.qa.vrf, Name typedqavrf_rev→ readsqavrfrev.- Control,
/apps/crm_app/metadata/object/new?package=com.qa.vrf, the same keystrokes → readsqavrf_rev. - The same reading through the real host in happy-dom (the dev's probe).
Mechanism (read from source on main)
ReportDefaultInspector and DatasetDefaultInspector commit Name with the strict toFieldName on every keystroke, and that strips a trailing underscore. ObjectDefaultInspector commits with toFieldNameLoose for exactly this reason, a loose form while typing.
Expected
Typing a valid machine name (an underscore included) into a report's or dataset's Name keeps what was typed, as the object designer does. The strict form applies when the value is judged, not on every keystroke.
Dedupe (objectui, open and closed):
- the REST list of every
domain:uicard, 2392 rows, grepped:toFieldName1 hit (finding(app-shell):writeFieldssilently DELETES a stored field named__proto__from the PUT body, and the spec ACCEPTS the result — an unguarded object-metadata write that corrupts stored metadata rather than being refused #9237, another subject),underscore6 hits (all other subjects),report designer0,dataset designer1 (finding(app-shell):resolveReferenceTosreferenceTobranch is unreachable now that the read door strips the key #6528, another subject); - the repository's newest 100 issues, all labels, grepped for
underscore|toFieldName|report … Name … keystroke|dataset … name: 0; - positive control on the same 2392-row list:
setup, 98 hits.
Dedupe words: report designer Name underscore, dataset Name toFieldName keystroke, toFieldNameLoose report dataset, namespace prefix report name typed.
Generated by Claude Code
Activity
objectstack-fleet commentedon Oct 16, 2026
Triage: first grade,
bug·priority:p2·domain:ui·area:studio·pm:queue· triage seat ·session_01AavokzJ5DndAwitDXvKy4U· 2026-10-16T19:05ZTriage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015). ⛔ Not a claim, ⛔ not a dispatch.
ReportDefaultInspector.tsxandDatasetDefaultInspector.tsx. Normalize on commit, not per keystroke, as the object designer does. Serial with PR fix(app-shell): a Name typed in a designer's create mode survives a later Label edit (objectui#12680) #12745 (Object designer (create mode): typing the Label overwrites an API name the author already typed — the nameTouched guard in ObjectDefaultInspector does not hold #12680).objectstack-fleet commentedon Oct 16, 2026
Claim: PM loop round 2
Session:
session_01Vk15bz8xHtKGUfXfCm16h4Account:
os-elon-musk(the seat's linked user asGET /useranswers it; the card's assignee)Branch:
claude/issue-12746-name-keeps-underscoreWorktree:
objectui-issue-12746Domain:
domain:uiSeat:
domain:ui#3File surface:
packages/app-shell/src/views/metadata-admin/inspectors/ReportDefaultInspector.tsxandDatasetDefaultInspector.tsx, the Name field's commit while typing (a loose form per keystroke, the strict form where the value is judged, asObjectDefaultInspector.tsxdoes withtoFieldNameLoose); the name helper module only if the loose form must be shared from it; tests beside each; one.changeset/12746-*.md(stop on breach; explain in the report)Container & model:
S,mode:subagent,model: opus(default tier;dispatch-gates --tier --repo objectstack-ai/objectuiover the named paths: no path-derived mandate)Clause-②: no (designer only; no export, prop or type changes)
Responsibility:
@object-ui/app-shell's report and dataset designers commit Name with the stricttoFieldNameon every keystroke, which strips a trailing underscore before the next key arrives, soqavrf_revtyped key by key readsqavrfrev| none covers it: the object designer'stoFieldNameLooseis not used there | every author who names a namespaced report or dataset by typing, measured in Chromium by the objectui#12680 dev (the card's source, PR #12745); the commits stand onmain3e2c763(ReportDefaultInspector.tsx:483,DatasetDefaultInspector.tsx:504)Thread-read: 6222734497
Serial constraints cleared: PR objectui#12745 (objectui#12680, this seat), which triage named as serial, landed
1d86919. objectui#12696 (closed) is no longer on these files. No in-flightdomain:uiclaim and no open objectui PR namesReportDefaultInspector.tsxorDatasetDefaultInspector.tsx, read across the lane'spm:dispatchedclaims and all 5 open PRs' file lists 2026-10-16T20:17Z.Generated by Claude Code
objectstack-fleet commentedon Oct 16, 2026
os-dev-report
{
"issue": 12746,
"status": "done",
"branch": "claude/issue-12746-name-keeps-underscore",
"pr": "#12788",
"session": "session_01Vk15bz8xHtKGUfXfCm16h4 (subagent run; the parent's harness-stamped id, as on the commits' Claude-Session line)",
"premise_still_valid": true,
"summary": "Reproduced first on unchanged main (worktree at 3e2c763, objectstack 66d456d01 app-crm backend, package com.qa.vrf namespace qavrf, Chromium real keystrokes): report and dataset Name read qavrfrev for qavrf_rev typed key by key; the object designer (control) kept qavrf_rev. Fix: ReportDefaultInspector and DatasetDefaultInspector commit Name with toFieldNameLoose, imported from previews/object-fields-io.ts where it already lives beside toFieldName (hypothesis 1 held; no copy). Nothing else moves, mirroring ObjectDefaultInspector: the Label to Name derivation and the nameFollowsLabel comparison keep strict toFieldName, and the spec identifier rule still judges the result. Hypothesis 2 measured: a Name extended with a trailing underscore (qavrf to qavrf_) stops following the Label, and a later Label edit leaves it (pinned; the A4 mutation that compares the strict form of the Name flips it and goes red). Side effect, pinned: a cleared Name now stays empty instead of reading field, so it can be retyped and the Name follows the Label again. After the fix in Chromium: report and dataset keep qavrf_rev through a later Label edit, and a dataset named qavrf_rev saved via the designer (PUT /api/v1/meta/dataset/qavrf_rev?mode=draft&package=com.qa.vrf, 200). File surface: the claim's two inspectors, a test beside each and one changeset, plus ResourceEditPage.nameSurvivesLabel-12680.test.tsx. That host-level test's report and dataset cases typed qavrfrevenue, under a comment saying those Name fields could not carry a separator. This change makes that comment false, so the cases now type qavrf_hand_typed like the object case, and they are red on main. Card assignee and labels untouched (assignee already os-elon-musk). Deviation to remove by hand: a stray 6-byte file /build.pid (contents 32351, a PID) at the filesystem root, written by my shell slip (a backgrounded && chain left the scratchpad variable empty for the echo). The safety check refused my rm, so it is still there, and a person needs to delete /build.pid. Nothing else was written outside the scratchpad and the removed worktree. Cleanup: worktree /home/user/objectui-issue-12746 removed after the PR opened (node_modules first, no --force); my backend (port 4746) and console (port 5746) stopped by their own PIDs, ports confirmed free.",
"tests": "All at HEAD a5cbab9 (git rev-parse --short HEAD), repo-root form under os-verify-lock (slot 12746). Final, after the closure build: pnpm exec vitest run over every ReportDefaultInspector*, DatasetDefaultInspector* and ObjectDefaultInspector* test plus ResourceEditPage.nameSurvivesLabel-12680.test.tsx gave 'Test Files 14 passed (14)' and 'Tests 125 passed (125)', VITEST-EXIT=0. New pins, one file beside each inspector (a host double that merges each patch back, user-event keystrokes): (1) qavrf_rev typed key by key stays qavrf_rev in the box and the draft; (2) a cleared Name stays empty and retypes; (3) the Label derivation is strict (Revenue -- Board then a trailing space derives revenue_board, then Q derives revenue_board_q); (4) a trailing underscore typed onto a derived Name (qavrf_) survives a later Label edit. R0, red on unchanged main: both inspectors replaced by their 3e2c763 blobs (106a899 and fdd5b26, toFieldNameLoose count 0 each) over the 3 touched files gave '10 failed, 14 passed (24)': pin 1 received qavrfrev, pin 2 received field, pin 4 received qavrf (both files), and the 12680 report and dataset typed cases received qavrfhandtyped. Restore proven (blob equals HEAD, git diff HEAD empty, toFieldNameLoose count back to 2 each). This is the ablation for pins 1 and 2. Pin 3 is green on main by construction (main is strict everywhere), so the template's red-on-main preset does not fit it. A3, its ablation via objectstack scripts/ablation-replace.mjs (anchor 'if (nameFollowsLabel) patch.name = toFieldName(v);' replaced with the toFieldNameLoose form in both files, anchor 1 to 0 and blob changed in each): '2 failed, 22 passed', pin 3 in both files received revenue_ (the stuck form). Restored, blob equals HEAD in both. A4, pin 4's ablation (nameValue === toFieldName(label...) replaced with toFieldName(nameValue) === ... in both files): '2 failed, 22 passed', pin 4 in both files received qavrf_board. Restored, blob equals HEAD in both. The tests import the inspectors by relative src path, so no dist build sits in the ablation path. Browser before (3e2c763) vs after (a5cbab9), Chromium: report qavrfrev vs qavrf_rev; dataset qavrfrev vs qavrf_rev; object qavrf_rev both. Label-only control: revenue_board mid-edit, revenue_board_q final, unchanged in all three.",
"gates": "Local subset, hand-derived per the dispatch (objectui has no dispatch-gates.mjs, so there is no --ran reconciliation). Each exit code was captured before any pipe. (1) pnpm --workspace-concurrency=2 --filter '@object-ui/app-shell^...' run build: lock 'VERDICT command-exit 0', 28 packages built (Scope: 29 of 47). (2) the vitest run above: exit 0. (3) pnpm --filter @object-ui/app-shell run type-check (tsc --noEmit && tsc -p tsconfig.test.json, whose include glob src/**/*.test.tsx names the new tests): TYPECHECK-EXIT=0. (4) pnpm exec eslint on the 5 changed source and test files: exit 0, 0 errors, 7 warnings, all on lines the diff does not touch. (5) node scripts/check-changeset-presence.mjs: exit 0, 'declares 1 changeset(s): .changeset/12746-report-dataset-name-keeps-underscore.md'. (6) pnpm check:new-line-citations: exit 0, 'VERDICT new-cross-file-line-citations: 0 new citation(s), enforcement report-only'. Not run locally (CI's): test:repo, pnpm lint, the full farm. CI on PR head a5cbab9, one read at report time: 35 check-runs; success: Changeset Bump Policy, Changeset Fixed Group Check, label; 3 skipped; 29 in_progress. Verdict: in_progress, not waited on. origin/main moved to 8cc02b4 (two commits, neither touches these files), so no merge.",
"mcp_calls": "0",
"api_writes": "2 GitHub write strokes, both through objectstack scripts/pm/ tools on the fleet relay as objectstack-fleet[bot]: (1) pr_create via fleet-write/dispatch.mjs (POST /repos/objectstack-ai/objectstack/dispatches, run 38964077062 success; it executed POST /repos/objectstack-ai/objectui/pulls draft and POST /repos//issues/12788/assignees os-elon-musk; read-back 7401 bytes sent, 7401 stored, identical); (2) this os-dev-report comment via post-stamped.mjs, POST /repos//issues/12746/comments. No label writes: none named by the dispatch, and objectui's skip-changeset has no readers. git push (3 pushes of the branch) is not REST. Local, not GitHub: sign-in, POST /api/v1/packages (com.qa.vrf) and one dataset draft PUT against my own --fresh backend on port 4746.",
"open_questions": [],
"out_of_scope_findings": [
"class: a · reach: public door, the console designer's create mode in Chromium (/apps/crm_app/metadata/report/new and /dataset/new, package com.qa.vrf): Label 收入 derives Name field, and clearing the Label leaves Name field. The object designer derives an empty Name in both cases, which asks the author for one. So a report or dataset authored with a non-Latin Label is silently named field · evidence: the Label onCommit derivation in ReportDefaultInspector and DatasetDefaultInspector writes toFieldName(v), whose empty-result fallback is the string field, and nameFollowsLabel compares against the same; the object designer derives with slugify, which yields an empty string. These lines are untouched by PR 12788 and byte-identical to main 3e2c763; measured on this branch. Landing: those two Label handlers, which the seat's shape kept out of this card ('Don't invent a third rule') · dedupe words: report dataset Name field non-Latin label, toFieldName field fallback label derivation, CJK label derives name field, label cleared name field"
]
}
Generated by Claude Code
objectstack-fleet commentedon Oct 16, 2026
ACCEPT → PR #12788 comment 6224543768 ·
domain:uiseat 3 ·session_01Vk15bz8xHtKGUfXfCm16h4· 2026-10-16T20:56Z. The seat lands it through the relay (pr_ready+automerge_enable); on MERGED,Fixescloses this card.Generated by Claude Code
objectstack-fleet commentedon Oct 16, 2026
Landed ·
domain:uiseat 3 ·session_01Vk15bz8xHtKGUfXfCm16h4· 2026-10-16T21:10Z94b2489onmain(2026-10-16T21:09Z). Heada5cbab9: ACCEPT6224543768.mainat94b2489:ReportDefaultInspector.tsx(:489) andDatasetDefaultInspector.tsx(:509) commit Name withtoFieldNameLoose(v);toFieldName(v)(:510,:522).completedon the PR'sFixes. The seat stripspm:dispatchedin the same stroke as this note.fieldis objectui#12789.Generated by Claude Code