This repository was archived by the owner on Nov 4, 2026. It is now read-only.
Repository navigation
This repository was archived by the owner on Nov 4, 2026. It is now read-only.
仪表盘聚合图表轴标题 "Count" 未本地化（类别标签已本地化） #7258
Description
现象
AI 搭建的客户仪表盘「按状态统计客户数量」柱状图：三根柱子的类别标签已本地化（合作中 / 已流失 / 潜在），但 X 轴标题显示英文 "Count"。中文界面下终端用户可见。
修复要求
聚合度量的轴标题/图例走 i18n（count → 数量，或按 measure label），并核对 sum/avg 等其它聚合名的同类字符串。
Activity
hotlong commentedon Sep 2, 2026
Claim: PM loop round 1 (epic objectstack-ai/cloud#1841)
Session:
session_c5c0ce54-bb9c-478c-9e5b-cf44b80d4569Branch:
claude/issue-7258-chart-axis-count-i18nWorktree:
objectui-issue-7258Domain: domain:ui
File surface:
packages/plugin-dashboard/**and/or the chart renderer that prints the aggregate axis title, pluspackages/i18n/**zh entry (stop on breach; explain in the report)Container & model: S, mode:subagent, model: sonnet (PM judgement)
Clause-②: no
Serial constraints cleared: only dependabot bumps touch plugin-dashboard package.json; #7254 also adds i18n zh entries — different keys, rebase conflicts are mechanical
hotlong commentedon Sep 2, 2026
Claim: dev
Session:
session_c5c0ce54-bb9c-478c-9e5b-cf44b80d4569Branch:
claude/issue-7258-chart-axis-count-i18nWorktree:
objectui-issue-7258Base: origin/main @ 67dadd6
hotlong commentedon Sep 2, 2026
os-dev-report
{
"issue": 7258,
"status": "needs_decision",
"branch": "claude/issue-7258-chart-axis-count-i18n",
"pr": null,
"premise_still_valid": true,
"summary": "The bug is real, but not fixable inside objectui without violating AGENTS.md #0.1 (contract-first). The literal 'Count' does NOT originate in the renderer: for a dataset-bound chart, DatasetWidget/ObjectChart pass queryDataset()'s res.fields straight into buildChartSeries() (packages/core/src/utils/chart-series.ts:441-442, labelOf), which trusts fields[].label as already-resolved author content -- exactly the behavior objectui#4106/ObjectChart.datasetMeasureLabel.test.tsx pins for a REAL custom measure label ('Tasks'). The English 'Count' is minted server-side in objectstack's service-analytics as a hardcoded default for the built-in count aggregate, with no i18n hook, at 5 sites: packages/services/service-analytics/src/analytics-service.ts:2045 and :2300 (inferMeasure), packages/services/service-analytics/src/cube-registry.ts:80 (inferFromObject), packages/spec/src/kernel/metadata-create-seeds.ts:98, mirrored in packages/qa/downstream-contract/src/additional-domains.fixtures.ts:116. None of these are in objectui.",
"tests": "No source changes made (diagnostic-only; no build/test run applicable). Evidence trail: grep -rn "'Count'" across objectui packages/ found only Studio config-panel enum labels (plugin-dashboard/src/index.tsx:242,318 -- authoring UI, not the rendered axis) and test fixtures -- no renderer-side literal that could produce a capitalized 'Count' axis title. Traced DatasetWidget.tsx:1390 (buildChartSeries(chartRows, dimensions, values, state.fields, ...)) and ObjectChart.tsx:848-856 (same helper) to core/src/utils/chart-series.ts:441-442, whose labelOf() falls back to the raw field NAME (lowercase 'count'), never a capitalized word -- ruling out a client-side humanizeLabel/humanizeFieldKey source (grep confirmed humanizeLabel is applied only to groupBy VALUES in ObjectChart.tsx:143/165, never to measure keys). Cross-referenced against objectstack (sibling checkout) and found the literal 'Count' hardcoded at the 5 producer sites listed in summary, all returning {name:'count', label:'Count', type:'count'} with no locale parameter. Confirmed existing CLIENT-side precedent for localizing this exact closed vocabulary purely by a structural discriminator: packages/i18n/src/locales/{en,zh}.ts 'report.aggregate.{count,sum,avg,min,max,...}' (zh: 'Count'->'计数') consumed by the pivot/matrix report, and 'grid.summary.{count,sum,avg,...}' (zh: 'Count'->'计数') consumed by packages/plugin-grid/src/useColumnSummary.ts -- both keyed by an explicit TYPE discriminator (ColumnSummaryType / report aggregate type), never by matching label text. Searched gh issue list on both repos for an existing duplicate (search: 'aggregate label i18n', 'measure label locale', 'aggregate label Count') -- objectui#4020/objectstack#5141 (axis showed raw NAME instead of label -- opposite direction, already fixed) and objectstack#4330/objectui#4575/#4566 (dimension-value and number-FORMAT locale, not aggregate-name TEXT) are related but distinct; no existing card covers this exact gap.",
"mcp_calls": "0 -- all GitHub reads/writes used the gh CLI, not MCP GitHub tools",
"open_questions": [
{
"question": "Where should the fix for the unlocalized default aggregate label ('Count'/'Sum'/'Avg'/...) land: a renderer-side heuristic in objectui, or a contract change in objectstack (service-analytics + spec) that objectui then consumes through its existing i18n seam?",
"options": [
"A. objectui infers 'this is a built-in aggregate' from the field NAME (e.g. name==='count') or from LABEL-TEXT matching ('Count'), then substitutes a locale string client-side, with no objectstack change.",
"B. Extend the wire contract: add a structural discriminator to AnalyticsResult.fields[] in packages/spec/src/contracts/analytics-service.ts (objectstack) -- e.g. a builtinAggregate?: 'count'|'sum'|'avg'|'min'|'max'|'count_distinct' field -- populate it ONLY at the objectstack producer sites that currently hardcode the English default (analytics-service.ts:2045/2300, cube-registry.ts inferFromObject), leave it absent for any author-declared custom measure so 'Tasks'-style labels are untouched; queryDataset's REST response threads it through; objectui's buildChartSeries()/labelOf() then prefers a locale lookup keyed by that discriminator (reusing/mirroring the existing report.aggregate.* i18n keys) over the raw label, falling back to label when the discriminator is absent.",
"C. Localize server-side: objectstack resolves the aggregate's display label by the request's exec.userLanguage (the channel already used for other server-generated user-visible strings per this repo's own doctrine) instead of a bare English literal, so the wire 'label' is already correct and objectui needs no change at all."
],
"recommendation": "B, because long-term soundness (>=50% weight) rules out A outright: name-based inference ('count') risks silently clobbering a real author-named measure literally called 'count' with its OWN custom label, and label-text matching ('Count') breaks the instant the string is anything but that exact English literal (already-resolved non-English text, a renamed default, etc.) -- both are exactly the consumer-side leniency around a producer defect that AGENTS.md #0.1 forbids, and both duplicate a translation responsibility the codebase has already assigned to a STRUCTURAL discriminator twice (ColumnSummaryType for grid.summary., the report's own aggregate type for report.aggregate.), never to label text or naming convention. B extends that exact, already-paid-for doctrine with one optional wire field and reuses an existing FE i18n namespace -- small, structurally sound, and hard for future AI-authored dataset/measure metadata to get wrong (real business need is confirmed by this issue's own repro; startup-scope discipline is satisfied because the diff is one optional field plus about three producer call sites plus one FE consumer, not new capability surface). C is the second choice if the maintainer prefers zero wire-contract change, but it duplicates the FE's is-this-my-language logic on the server and provides no discriminator for a future FE need (e.g. an author wanting to override just the built-in label per-widget); it is viable but a strictly worse fit for the existing architecture than B."
}
],
"out_of_scope_findings": []
}
hotlong commentedon Sep 2, 2026
PM ruling (maintainer 2026-09-02, in-session): B — structural discriminator on the analytics contract. The producer half is filed as objectstack-ai/objectstack#14492 (spec seat, Clause-②: yes). This card is the consumer half and stays parked until that lands.
Blocked-by: objectstack-ai/objectstack#14492
Triage of the dev report: premise valid, no PR (diagnostic-only run) — a good run, not a failed dispatch. Branch
claude/issue-7254-style claim stays; re-dispatch once the contract field exists.Generated by Claude Code
hotlong commentedon Sep 3, 2026
Claim: epic cloud#1841 lane — consumer half re-dispatched in parallel with the producer (objectstack#14492, now in-seat cross-lane per the owner's 2026-09-04 authorization)
Session: wonderful-driscoll-ac0895 (cloud PM session, Fable 5.1)
Branch: fix/chart-builtin-aggregate-label-7258
Worktree: ~/Documents/GitHub/objectui-issue-7258 (linked worktree off origin/main; primary checkout untouched)
Domain: core chart-series label resolution + i18n locale keys
File surface: packages/core/src/utils/chart-series.ts (labelOf / buildChartSeries), the i18n locale bundles that already carry report.aggregate.* keys (en + zh), tests beside chart-series; spec dependency bump ONLY if the repo consumes @objectstack/spec by pin and the field is needed at type level
Container & model: general-purpose subagent reading objectstack/.claude/agents/os-dev.md, inherit tier
Clause-②: no (consumer side; reads an optional field already ruled in #14492)
Serial constraints: PR stays draft until objectstack#14492 is merged so the contract exists upstream; code against the optional field structurally so the branch compiles before the spec bump
Generated by Claude Code
hotlong commentedon Sep 3, 2026
os-dev-report
{
"issue": 7258,
"status": "done",
"branch": "fix/chart-builtin-aggregate-label-7258",
"pr": "#7535",
"premise_still_valid": true,
"summary": "Consumer half of ruling B, DRAFT PR #7535 at 799c84a (kept draft: Depends on objectstack#14492 landing upstream for the field to be populated; inert on spec 17.2.0 by construction). buildChartSeries()/labelOf() now resolves a result field that carries the optional wire discriminator builtinAggregate (closed enum count|sum|avg|min|max|count_distinct, validated at the lookup by isBuiltinAggregate) through a caller-supplied locale map, and otherwise keeps fields[].label verbatim (objectui#4106) then the name; no name/label-text inference. Core stays React/i18n-free, so the map arrives as ChartSeriesOptions.builtinAggregateLabels (the nullCategoryLabel division); a new i18n seam builtinAggregateLabels(tt) reads the existing report.aggregate.* keys (all ten packs already carry them, bundles untouched), and the three chart callers (ObjectChart, DatasetWidget, DatasetReportRenderer) pass it. Type level: @objectstack/spec is consumed by npm RANGE (core ^17.2.0, lockfile 17.2.0; no workspace link, no SHA pin); adding the key ON ChartResultField turned the compile-time pin keyof ChartResultField subset-of spec fields[] (dataset-result-field-spec-parity, TS2344 line 121) red, so the key lives BESIDE it as BuiltinAggregateCarrier and buildChartSeries takes ChartMeasureField[] (ChartResultField intersected with the carrier) so a spec-typed DatasetResultField[] flows in unchanged today and carries the key once the spec does; NO pin bump needed for this PR, follow-up TODO(objectstack#14492) in chart-series.ts to fold the key into ChartResultField after the spec release ships. File-surface breach declared in the PR body: 3 caller edits (one option each), the i18n seam module + index export, and a DatasetWidget surface pin under a real I18nProvider, each necessary so the seam is wired rather than declared. Two standard-text deviations, both from the dispatch word: branch name is the Claim's (fix/chart-builtin-aggregate-label-7258, not the claude/issue-N form) and the PR body ends with the dispatch-mandated footer with the session id carried in prose. CI: green — all 32 check runs on 799c84a completed with 0 failures (29 success incl. Build & E2E, Lint, Type Check, Test shards 1-4, README Export Check, Changeset Declaration, Bundle Analysis; 3 skipped: coverage shards and dependabot), read via REST check-runs (17 polls) after one gh pr checks --watch burst tripped the GraphQL abuse limiter. PR left as DRAFT, not marked ready, not merged.",
"tests": "Union at 799c84a (final commit; ablation restored to that exact blob first). type-check (tsc --noEmit + tsc -p tsconfig.test.json) for core, i18n, plugin-charts, plugin-dashboard, plugin-report: all five Done, exit 0, after building the plugins' dependency closure; --listFiles shows each new test file in its package's test project (1/1/1). vitest run --maxWorkers=2 over 16 files (3 new + chart-series/dataset-format/spec-parity/label-net siblings + ObjectChart.datasetMeasureLabel + two DatasetWidget label suites): Test Files 16 passed (16), Tests 256 passed (256), exit 0. Ablation after commit: labelOf mutated back to field?.label ?? name (on-disk: marker count 1, removed call 0), core+dashboard pins ran 4 failed | 14 passed (18) — red exactly at zh built-in count (core), every-member, multi-measure, zh built-in count (surface); observed direction note: the en pins stayed green under the mutation because the en locale label IS the wire Count, so they are controls; restore via git checkout HEAD -- path, git hash-object equals HEAD blob bcae7ca08, git diff HEAD empty, marker 0; no dist involved (vitest aliases @object-ui/* to src). ESLint --format json over the 9 ts/tsx files of the diff: exit 0, 0 errors, 74 warnings, 0 on any added line (74 pre-existing: ObjectChart 58, DatasetWidget 11, DatasetReportRenderer 5); narrowing proof: universe = eslint files glob, count = 9 JSON entries, grep -cE projectService|parserOptions|project: eslint.config.js = 0 (control files:|ignores: = 9). check-changeset-presence (9 source files / 5 released packages / 1 changeset), -fixed, -no-major, -overwrite: exit 0 each. check-control-bytes OK (6206 files) + targeted perl scan of diff files 0. check-i18n-call-site-keys, check-i18n-en-drift (no en value changed), check-phantom-dependencies: exit 0. NOT MEASURED locally: check:readme-exports (257 type-entry-not-on-disk, 22 unbuilt packages) and check:spec-floors (12 findings all no-artifact on unbuilt packages; the 19 built ones incl. all five touched judged 185 pairs clean) — CI measures both on a full build. os-verify-lock.sh ran in DECLARED UNLOCKED MODE (no flock on this macOS host), nothing serialized.",
"mcp_calls": "0 — every GitHub read/write went through the gh CLI; one gh pr checks --watch burst tripped the GraphQL abuse limiter, after which CI was read via REST check-runs only",
"open_questions": [],
"out_of_scope_findings": [
"filed as #7534: buildDatasetFieldHelpers().headerLabel (KPI caption, table/pivot headers, report metric caption, app-shell DatasetPreview) still prints the wire Count — same class one seam over, mechanical follow-up with resolveMeasureLabel across five call sites, Blocked-by objectstack#14492"
]
}
hotlong commentedon Sep 3, 2026
ACCEPT (pending producer) — PR #7535 verified in-seat: 10 files, all inside the declared surface plus the three one-option caller edits and the i18n seam the PR body declares; 29 checks success / 3 skipped / 0 failed at 799c84a; consumer reads the optional
builtinAggregatestructurally (no spec pin bump needed — @objectstack/spec is an npm range here and the cloud image links the pinned framework at build time, so the field arrives via the cloud pin bump). Order: objectstack#15017 merges first → this PR goes ready + queue → cloud.objectui-shabump. #7534 (KPI / table / pivot captions, same class) is tracked for the follow-up batch.Generated by Claude Code
hotlong commentedon Sep 3, 2026
Producer objectstack#15017 is ready and in the objectstack merge queue (clause-② PASS, all checks green), so #7535 is now marked ready and enqueued as well. The behaviour switches on once cloud's
.objectstack-shaand.objectui-shabumps carry both sides (epic cloud#1841).Generated by Claude Code