Dashboard table: numeric measure columns sort descending on first click #5845

Activity

1. os-zhuang

   self-assigned this

   on Aug 23, 2026
2. os-zhuang commented on Aug 23, 2026

   

   os-zhuang

   on Aug 23, 2026

   ContributorAuthor

   More actions

   Claim: direct-dispatch (hotcrm seat acting cross-repo under maintainer instruction)
   Session: session_61f28fb9-8e31-4a6f-b4d8-10f749a092d2
   Branch: claude/issue-5845-measure-desc-first
   Worktree: objectui-issue-5845
   File surface: packages/plugin-dashboard/src/DatasetWidget.tsx + __tests__/DatasetWidget.tablePolish.test.tsx (stop on breach; explain in the report)
   Container & model: S(判断面在既有语义上,按 M 待遇), mode:subagent, model: opus — Clause-②: no.
   Serial constraints cleared: #5834 merged (62c0fa1); the totals-row follow-up is queued BEHIND this card (same file) and is not in flight. No other open PR touches DatasetWidget.tsx (REST pulls listing checked at claim time).

   Authorization — maintainer ruling, verbatim (未译): 「我觉得合理的需求都应该开发」
3. os-zhuang

   added a commit that references this issue on Aug 23, 2026

   fix(dashboard): sort a numeric measure column descending on the first…

   ...

   2d4c360
4. os-zhuang commented on Aug 23, 2026

   

   os-zhuang

   on Aug 23, 2026

   ContributorAuthor

   More actions

   {"issue": 5845, "status": "done", "branch": "claude/issue-5845-measure-desc-first", "pr": "https://github.com/objectstack-ai/objectui/pull/5848", "premise_still_valid": true, "summary": "Numeric MEASURE columns in the flat dataset `table` widget now cycle descending -> ascending -> the dataset's own order; dimension columns and non-numeric measures keep ascending -> descending -> dataset order. The predicate is #5834's `numericColumns` set (a measure, or a measure's comparison column, whose every non-null value is a number) - the same set that decides right-alignment - so first-click direction follows the column's ROLE, not 'it looks like digits': a digit-keyed dimension (a year) still starts ascending and so does a min()/max() over a text field. PM MECHANISM ASSUMPTION, TESTED: 'two lines in cycleSort' is CONFINED correctly - nothing else in the file assumes asc-first, because aria-sort, the ChevronUp/ChevronDown indicator and the comparator all read activeSort.dir (the direction APPLIED, never the position in the cycle), so a desc-first initial state renders and announces itself truthfully with zero further change, and #5834's blanks-last arm is settled BEFORE the direction is applied and is therefore direction-agnostic by construction. But a literal two-line edit would have SHIPPED A BUG: the old cycle spelled both directions as literals (asc -> desc -> null), so swapping only the first yields desc -> desc -> null and a measure column that can never be sorted ascending. The middle step is now written as 'whatever firstDir is not', giving both column kinds all three states, and a column switch starts the new column at ITS OWN first direction rather than inheriting. One consequence named in the code: the descending arm is now what a measure's FIRST click runs, so #5834's negate-trap moved from the second interaction to the very first - a test pins blanks-last on that first click. Scope held: #5846 (totals row) untouched and the file not restructured for it.", "tests": "All from the repo root in a dedicated worktree; union re-run AFTER the final commit, at 2d4c3609b. Every exit code captured by redirecting to a file first, never read after a pipe. `pnpm --filter '@object-ui/plugin-dashboard^...' build` exit 0 (dependency closure). `pnpm --filter @object-ui/plugin-dashboard type-check` exit 0, script echoed ('tsc --noEmit && tsc -p tsconfig.test.json'), no diagnostics. `pnpm --filter @object-ui/plugin-dashboard lint` exit 0, verdict line '369 problems (0 errors, 369 warnings)' - the IDENTICAL count #5834 reported, so this diff adds no warning. `pnpm exec vitest run packages/plugin-dashboard/` exit 0, 'Test Files 77 passed (77)' / 'Tests 726 passed (726)'; the touched file itself went 11 -> 14 tests ('Tests 14 passed (14)'). NOT DONE, stated plainly: no separate baseline run on BASE d40d2953e - origin/main moved since #5834's 76/693 reading, so the 'nothing pre-existing moved' claim rests on the file-level 11 -> 14 delta and a green package suite, not on a re-measured baseline. 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 ('No changeset declares a `major` bump.'), check-control-bytes.mjs exit 0 ('OK (scanned 4868 tracked text file(s); skipped 85 binary)'), check:i18n-keys exit 0 ('Every in-scope call-site key resolves against the en pack (2928 keys)'). REVERSE VERIFICATION - direction stated BEFORE the run: reverting ONLY DatasetWidget.tsx to origin/main turns 6 of 14 red (the three #5845 cases plus #5827's 'new column at its own first direction', 'sorts numerically' and 'CSV export') and leaves the 8 alignment/dimension-cycle/blank-bucket/drill/hover cases green, since those are exactly the half the ruling does not touch. OBSERVED: 'Tests 6 failed | 8 passed (14)' and the six named failures are the six predicted. Mutation confirmed ON DISK by anchored greps, not by an editor exit code: `firstDir` 3 -> 0 and the deleted origin/main spelling `prev.dir === 'asc'` 0 -> 1. Restore leg confirmed the same way (3 / 0) plus an empty `git status --porcelain`, re-ran green at 14/14. No rebuild stands between edit and run - the suite imports the component by relative source path (../DatasetWidget), not through a package exports->dist hop - and the ablation script carried `trap ... EXIT INT TERM`, which did fire. NOT VERIFIED IN A BROWSER (same limit #5834 declared): component tests over the produced DOM, not a console rendering HotCRM's executive_dashboard. DECLARED NARROWINGS: the shared lock is inoperable on this host (os-verify-lock.sh --status prints 'CANNOT OPERATE THIS LOCK ... a stock macOS does not ship flock'), so commands ran directly rather than through the entry point; repo-wide check:* scanning left to CI; eslint was also run targeted at the two changed files (0 errors) in addition to the package-level run. objectui has no scripts/pm/dispatch-gates.mjs - the gate list above was derived by hand from this repo's package.json check:* scripts and .github/workflows, and objectstack's copy of that script was deliberately NOT pointed at objectui paths.", "open_questions": [], "out_of_scope_findings": []}
5. os-zhuang commented on Aug 23, 2026

   

   os-zhuang

   on Aug 23, 2026

   ContributorAuthor

   More actions

   PM review — ACCEPT. PR #5848. The "two-line change" premise was half false, and the falsification is the value of this run.

   The literal two-line edit would have shipped a bug

   The sibling dev's estimate (which my dispatch carried) assumed swapping the first direction sufficed. Measured: the old cycle spelled both directions as literals (asc → desc → null), so swapping only the first yields desc → desc → null — a measure column that can never be sorted ascending, green on a naive "first click is descending" test. The middle step is now "whatever firstDir is not", verified in the diff; both column kinds get all three states, and a column switch starts the new column at its own first direction instead of inheriting the previous column's.

   Two structural facts made the rest clean, both verified claims: aria-sort, the indicator and the comparator all read the applied direction (never the cycle position), so a desc-first initial state announces itself truthfully with zero further change; and #5834's blanks-last arm was settled before direction is applied, so it is direction-agnostic by construction — with a new test pinning blanks-last on the first click, where the negate-trap now lives.

   Verification

   Predicted 6 red / 8 green under ablation; observed exactly those six, and the eight greens are precisely the half the ruling does not touch (alignment, dimension cycle, blank bucket, drill pairing, hover). On-disk mutation confirmation both legs; 14/14 restored. Lint count identical to #5834's (369), so zero new warnings. Honestly declared: no re-measured baseline on moved main (file-level delta + green package suite carry the claim), and no browser verification — the PM runs the visual half in the overlay, as with the previous two cards.

   Disposition

   Ready + merge on CI green. #5846 (totals row, same file) dispatches after this lands — scope was held, the file was not restructured for it.
6. os-zhuang

   closed this as completedin #5848on Aug 23, 2026
7. os-zhuang

   removed their assignment

   on Aug 23, 2026
8. akarma-synetal

   added a commit that references this issue on Sep 1, 2026

   fix(dashboard): sort a numeric measure column descending on the first…

   ...

   1d9100e