R2 of the unused-code program (#4806): promote the unused-IMPORT subclass of no-unused-vars to error and clear the measured 114 sites

[os-trump]

Filed by triage (daily full round, 2026-08-26) to pick up a dropped baton: #4806's recorded tiered disposition (2026-08-16 PM comment) chartered "R2 as its own card after R1 lands" — R1 landed the same day (PR #4844) and the R2 card was never opened.

Charter (from the #4806 disposition and measurements — re-take all numbers on the current ref, they are ten days old)

1. Promote the unused-import subclass of @typescript-eslint/no-unused-vars to error (narrow gate: imports only, not the whole rule), and clear the measured ~114 genuinely-unjustified unused imports in the same PR (the sweep-batch pattern: itemised list in the PR body, zero changes outside it).
2. Scheduling discipline: a gate-tightening PR collides with in-flight PRs' merge windows — follow the [finding] 类型门收紧 PR 与旧式抑制指令的合并窗口相撞已两例(第二例弄红了 main)—— 值得一道结构性防线 #3535 semantic-collision discipline recorded in finding: 未使用的具名导入/局部变量在本仓结构上不可能让 CI 变红 —— no-unused-vars 是 warn(无 --max-warnings),且 38 个包 extends 的根 tsconfig 关着 noUnusedLocals(base 里开着的那条 0 个包在用) #4806's R2 note (land against fresh main, re-merge before queue).
3. Scope notes recorded on finding: 未使用的具名导入/局部变量在本仓结构上不可能让 CI 变红 —— no-unused-vars 是 warn(无 --max-warnings),且 38 个包 extends 的根 tsconfig 关着 noUnusedLocals(base 里开着的那条 0 个包在用) #4806 that bind this card: the jsxPragma question (421 dead React imports visible only to tsc) is a SEPARATE review, not a rider; and the "_-prefixed but actually dormant" class R1 hid (e.g. let _projectRoot) is deliberately outside this card's 114 — decide explicitly whether to list it as excluded or add a class, don't let it fall out silently.
4. R3 (the tsconfig.base.json positioning, a 0-radius maintainer call) and R4 (per-package noUnusedLocals flips) stay on finding: 未使用的具名导入/局部变量在本仓结构上不可能让 CI 变红 —— no-unused-vars 是 warn(无 --max-warnings),且 38 个包 extends 的根 tsconfig 关着 noUnusedLocals(base 里开着的那条 0 个包在用) #4806 — ⛔ not riders here.

Anchor: #4806 (tracking; carries the full measurements: 822 warnings, 74.6% _-convention noise, 114 real, two gates watching disjoint populations). Related: #4833 (stale eslint-disable, closed), #4853 (reportUnusedDisableDirectives precondition, separate gate-strength question), #5612/#5613 (the recurrence that re-proved the gap).

[yinlianghui]

Claim:

• Session: session_01BGMDbrVa8JjZcCQ7DWYH1b (execution seat, domain:devx @ objectui, seat post ⚪ RELEASED · last seat session_01BGMDbrVa8JjZcCQ7DWYH1b · R40 closed 2026-09-02T07:15Z · tail: #6467 in flight (dev PR pending) · #7325 #7326 queued (Part of #5174) · governed drafts awaiting human merge #7317 (#6970+#6452+#7185) #7264 (#6451) · decision box #6600 · awaiting-maintainer #4912 #5748, round R40)
• Branch: claude/issue-6467-unused-import-gate
• Worktree: /home/user/objectui-6467 (dedicated per-task worktree, off origin/main)
• Domain: domain:devx (root eslint.config.js gate strength + a repo-wide mechanical sweep; type Task)
• File surface: eslint.config.js (the @typescript-eslint/no-unused-vars rule block, currently warn at line 75 on origin/main @ dd5b01b59) plus the unused-import sites the re-measurement finds (charter figure ~114, ten days old — re-take on the current ref). Sweep-batch pattern: itemised list in the PR body, zero edits outside it. ⛔ No jsxPragma rider; ⛔ R3/R4 stay on finding: 未使用的具名导入/局部变量在本仓结构上不可能让 CI 变红 —— no-unused-vars 是 warn(无 --max-warnings),且 38 个包 extends 的根 tsconfig 关着 noUnusedLocals(base 里开着的那条 0 个包在用) #4806; the _-prefixed-but-dormant class is listed as excluded or given its own class, explicitly, never dropped silently.
• Container & model: subagent os-dev, opus, background. Sizing note: the charter is broad but mechanical (one rule change + a removal-only sweep with no design decision), so this seat sizes it M and dispatches as a subagent rather than reserving it for the cloud lane; if the re-measured population is materially larger than the charter's 114 or needs judgment per site, the dev stops and reports the count instead of landing a partial sweep.
• Clause ②: not applicable (no packages/spec surface; removing unused imports changes no published API).
• Serial constraints cleared: in-flight cards touch scripts/** (finding(test-config): the unit project's isolate: false is justified by a comment that is false — registry-writing AND registry-absence tests both live there #7134 landed, tier 2 of #5043: compile the README code blocks (+ bidirectional pins for documented interface blocks) — needs a baseline decision before it can be a gate #6214 → PR feat(scripts): pin documented interface blocks against the shipped type, both ways #7306), root vitest.setup.* (finding(test-infra): the ECONNREFUSED 127.0.0.1:3000 class is closed four times over yet still prints in 12 of 16 GREEN full-suite shards — and it reads as a red gate whenever a run is cut short #6640), console test files' mock factories (finding(test-infra): importOriginal() on the source-aliased @object-ui/app-shell barrel costs ~10s per console test file — measured once, and up to 12 files are in the same shape #6580), content/docs/** (check-doc-snippet-types.mjs collects only .mdx under content/docs — 40 .md guides are uncovered with no UNGATED_DOCS entry, contradicting its own "covered by default" rule #5174) and vitest.config.mts (finding(ci): a claim about a package's BUILT artifact cannot be pinned — turbo's test task builds dependencies only, so a dist-importing test is NOT MEASURED in CI #7183/finding(test-config): the unit project's isolate: false is justified by a comment that is false — registry-writing AND registry-absence tests both live there #7134). An unused-import sweep may touch any source file, so the dev excludes from the sweep every file changed by an open PR on this lane (test(ci): a dist vitest project so built-artifact claims can be pinned #7291, test(infra): attribute and gate the ECONNREFUSED network-escape class repo-wide #7304, test(e2e): tell an open modal from a blank viewport in the boot-continuity coverage rule #7305, feat(scripts): pin documented interface blocks against the shipped type, both ways #7306, test(config): state the unit project's real isolate: false invariant, and enforce it #7309 and finding(test-infra): importOriginal() on the source-aliased @object-ui/app-shell barrel costs ~10s per console test file — measured once, and up to 12 files are in the same shape #6580's eventual PR) and lists those exclusions in the PR body; per the [finding] 类型门收紧 PR 与旧式抑制指令的合并窗口相撞已两例(第二例弄红了 main)—— 值得一道结构性防线 #3535 discipline recorded on finding: 未使用的具名导入/局部变量在本仓结构上不可能让 CI 变红 —— no-unused-vars 是 warn(无 --max-warnings),且 38 个包 extends 的根 tsconfig 关着 noUnusedLocals(base 里开着的那条 0 个包在用) #4806, the PR is re-merged against fresh main immediately before it is flipped into the queue.

---

Generated by Claude Code

[yinlianghui]

os-dev-report

{
 "issue": 6467,
 "status": "done",
 "verdict": "done",
 "branch": "claude/issue-6467-unused-import-gate",
 "pr": "https://github.com/objectstack-ai/objectui/pull/7332",
 "head_sha": "58ad7ff1fdf8b3695876d85c5547318edae8c609",
 "files_changed": 87,
 "premise_still_valid": true,
 "summary": "R2 landed as chartered. The unused-IMPORT subclass of @typescript-eslint/no-unused-vars is now `error` via a new local rule `object-ui/no-unused-imports` that RUNS the upstream rule unmodified and forwards only reports on import bindings, configured from an identical options object so the two halves cannot drift; no new dependency, and every other subclass keeps the severity it had. Re-measured on origin/main @ dd5b01b59 the population is 108 sites in 82 files (charter said ~114, ten days old), all mechanical, 0 needing per-site judgement and 0 excluded, so the sizing ruling permitted landing rather than stopping at measurement. The population was derived twice by independent instruments that agree on count, distribution and per-site list; the sweep itself is the upstream rule's own fixer under `--fix-type problem`, and the resulting diff touched exactly the 82 expected files with zero unexpected files and zero misses. The `_`-prefixed-but-dormant class is EXCLUDED and said so explicitly in the PR body: it is not an import class at all, 0 of the 108 carry a `_` prefix, and every member sits in the 102 findings left at warn. No jsxPragma rider, no R3/R4 rider.",
 "tests": "All exit codes captured by redirect before any pipe; heavy runs through os-verify-lock.sh (slot dev-6467). REPO-WIDE LINT, merged head: `pnpm exec eslint . --format json` exit 0 — 4142 files, 0 errors, 0 object-ui/no-unused-imports findings, @typescript-eslint/no-unused-vars 102. BASE dd5b01b59 for comparison: exit 0, 0 errors, 11801 warnings, no-unused-vars 210, unused-import subclass 108. Both deltas are 108 and they are the same 108 (11801->11693 warnings; 210->102). NON-VACUITY, all in ONE shell invocation: three swept files restored to pre-sweep content from the base commit -> `PLANT_LINT_EXIT=1`, `202 problems (10 errors, 192 warnings)`, the 10 errors being exactly the 10 sites those files carried — layout/src/index.ts:9 SidebarNav, plugin-list/src/ListView.tsx 10:133/10:141/10:156/10:168/10:183/12:88/12:112/25:10, types/src/views.ts:22 FormField, every one `object-ui/no-unused-imports`. Mutation proven on disk BEFORE the lint (each reintroduced import text grepped, count 1 each); restore proven AFTER by empty `git diff HEAD` AND `git hash-object` equal to each path's HEAD blob (9743fc77.., dde38cdc.., 510c730a..). CONTROL: scratch file gives `3 problems (1 error, 2 warnings)` — unused import errors AND warns, unused local warns only; `eslint --print-config` shows the pair as [2,{...}] and [1,{...}] with identical options. BUILD `turbo run build` exit 0 `Tasks: 44 successful, 44 total`. TYPE-CHECK `turbo run type-check` exit 0 `Tasks: 81 successful, 81 total` (each package task is `tsc --noEmit && tsc -p tsconfig.test.json`, so test files are type-checked too). RULE SUITE `vitest run eslint-rules/no-unused-imports.test.js` exit 0, 21/21. TESTS from repo root, exit 0 in every batch, 1220 files / 16192 assertions total: touched-tests+tooling 39/780; core,types,react,cli,auth,layout,fields,schema-catalog 467/8781; components renderers+plugin-view 128/1145; rest of components 126/1171; plugin-list 62/787; plugin-gantt,designer,ai,charts,calendar,timeline,map 165/1391; plugin-grid,plugin-detail 233/2137. GATES: check-changeset-presence.mjs exit 0 '72 source file(s) of 21 released package(s) changed, and this change declares 1 changeset(s) ... Every one of them has an EMPTY frontmatter'; check-changeset-no-major.mjs exit 0 'No changeset declares a major bump.'; check-control-bytes.mjs exit 0 'OK (scanned 6012 tracked text file(s); skipped 85 binary)'; check-lint-coverage.mjs exit 0 '46/46 packages linted, 0 with outstanding errors'; check-entry-guard.mjs exit 0 '58 scripts/ file(s) — no entry guard outside the baseline'; tsc -p tsconfig.scripts.json exit 0; check-phantom-dependencies.mjs exit 0 'Every in-scope import is declared by the package that publishes it.' Gate list derived from .github/workflows/lint.yml plus the touched paths; scripts/__tests__/lint-workflow.test.ts is the one that judges a new error-level object-ui rule and it passes unedited (it deliberately keeps no enumeration). MERGE: origin/main merged as the LAST step before this report, dd5b01b59 -> c93b4d5f3, nine commits, no conflicts, pnpm-lock.yaml unchanged, and the repo-wide lint above IS the re-run on that merged head — the commits that landed meanwhile brought no new unused import. The changeset follow-up commit 58ad7ff1f changes one .changeset markdown file and no file any eslint glob matches, so that lint reading carries to the current head unchanged (`git diff --name-only 68fd2a0cc 58ad7ff1f` = one .md).",
 "mcp_calls": "9",
 "deviations": [
 "PROCESS DEFECT, self-inflicted, found by CI and fixed: the empty-frontmatter changeset was written AFTER the main commit and never staged, so it sat untracked. check-changeset-presence.mjs compares the WORKING TREE, so it went green locally on a file that was not in any commit, while CI — which sees only committed content — went red on head 68fd2a0cc. Fixed in 58ad7ff1f (added, committed, no amend, no force-push); both changeset gates re-run afterwards from a CLEAN working tree, so the green now reads about committed state: presence exit 0 '... declares 1 changeset(s): .changeset/6467-unused-import-gate.md', no-major exit 0. Worth recording as a class: a working-tree-reading gate cannot tell 'added' from 'about to be lost', and `git status` was the only thing that would have caught it.",
 "DECLARED NARROWING: packages/app-shell's full vitest suite is the one batch not run locally. It exceeded the container's ~10-minute foreground cap three times (whole suite twice, then src/views/ alone), each time with no failure printed — NOT MEASURED, read as neither green nor red. Its seven swept source files are covered on the compile side by turbo type-check 81/81, and its five swept TEST files ran green in the touched-tests batch. CI runs it in full. Stated as a narrowing in the PR body too, not passed off as a pass.",
 "The sweep instrument (a temporary minimal flat config enabling only no-unused-vars with enableAutofixRemoval) is NOT part of the PR; the shipped gate declares no autofix option, so `eslint --fix` does not delete imports. A first fixer run without `--fix-type problem` touched 115 files instead of 82 because ESLint also stripped 33 files' worth of eslint-disable directives that are unused only under that probe config and entirely live under the real one; that run was reverted wholesale via `git checkout HEAD -- .` before the narrowed re-run, and the 82-file result was then diffed against the measured list (zero unexpected files, zero misses).",
 "The changeset's prose differs from the wording suggested in the patch instruction — kept the version already quoted in the PR body, which states the same thing (empty frontmatter, no published behaviour change) with the actual counts. The frontmatter, which is the part the gate reads, is exactly the empty form."
 ],
 "open_questions": [],
 "out_of_scope_findings": [
 "filed as #7336: packages/fields/src/standard-widgets.test.tsx covers 7 of the 10 standard widgets it imports — DateField, PhoneField and TextAreaField had no case in the file named for that set. Filed specifically because this sweep DELETED the only visible trace of the gap: the three unused imports were the symptom, removing them is correct, and the file now reads as a tidy complete suite with no signal that three members are missing. Unassigned, labelled finding/domain:devx.",
 "NOT filed, recorded here as a judgement call for the PM to overturn if wanted: packages/components/src/custom/input-group.tsx imported Input and Button from ../ui/* and rendered neither. Unlike the fields case there is no residual defect once the imports are gone — InputGroup takes startContent/endContent as ReactNode and is a container by design, so the imports were simply leftovers, not evidence of missing behaviour. Filing it would have been a card with nothing left to do."
 ]
}

---

Generated by Claude Code

[yinlianghui]

ACCEPT — PR #7332 (head 58ad7ff1f, after the changeset fix commit), reviewed against GitHub by the domain:devx seat (session_01BGMDbrVa8JjZcCQ7DWYH1b, R40 tail, handled after the shift-end brief per the maintainer's "finish what is unfinished" instruction).

Checked:

• Charter met exactly: only the unused-IMPORT subclass is promoted, via a local rule object-ui/no-unused-imports that runs @typescript-eslint/no-unused-vars unmodified and forwards import-binding reports at error; the warn half is untouched (control shows the same import warned + errored, an unused local still only warned). No new dependency; the two alternatives (eslint-plugin-unused-imports, base-rule options) are measured out in the PR body and the rule header.
• Population re-measured at 108 sites / 82 files by two independent instruments; the sweep is the upstream fixer under --fix-type problem (the first attempt without it touched 33 files' eslint-disable directives and was reverted wholesale), reconciled 82/82 against the measured list; the 23 sole-specifier removals read individually, none a side-effect import.
• Deferred sites: none, measured against every open PR on the lane and the finding(test-infra): importOriginal() on the source-aliased @object-ui/app-shell barrel costs ~10s per console test file — measured once, and up to 12 files are in the same shape #6580 exclusion set. The _-prefixed dormant class is excluded by construction (0 of 108 carry it) and left on finding: 未使用的具名导入/局部变量在本仓结构上不可能让 CI 变红 —— no-unused-vars 是 warn(无 --max-warnings),且 38 个包 extends 的根 tsconfig 关着 noUnusedLocals(base 里开着的那条 0 个包在用) #4806 for R4; jsxPragma, R3, R4 not touched.
• Non-vacuity: three swept files restored to pre-sweep content lint red with exactly their 10 sites named; restore proven by blob hash. Gates on the merged head: repo-wide eslint . 0 errors / 0 unused-import findings, turbo run build 44/44, turbo run type-check 81/81, changeset presence (empty-frontmatter changeset, now committed — the first push had left it out and CI's Changeset Declaration went red; fixed on 58ad7ff1f, green), no-major, control-bytes, lint-coverage, entry-guard, phantom-deps. Tests from the repo root in seven batches (1221 files / 16192 tests); packages/app-shell's full suite declared narrowed (container cap), covered by type-check and by CI.
• origin/main merged last per the [finding] 类型门收紧 PR 与旧式抑制指令的合并窗口相撞已两例(第二例弄红了 main)—— 值得一道结构性防线 #3535 discipline; the gate re-run on the merged head reports 0.

The PR flips out of draft and enters the queue once CI on 58ad7ff1f is green (Lint, Type Check and the four test shards are running). Card closes on merge via Fixes #6467.

---

Generated by Claude Code