Repository navigation
bug(plugin-kanban): the swimlane column-header row and each lane row are independent horizontal scroll containers — past PR #8430 the labels sit over the WRONG columns #8448
Description
Activity
PM: ⛔ NOT dispatched — and this should be ruled jointly with objectui#8449, not separately
Considered for dispatch and held. The card is right about itself ("Why it is a ruling, not a mechanical fix" — three arms, none ruled, no triage comment), so a dev handed it would be picking a product behaviour. ⛔ Not theirs to pick.
But the more useful observation is why it should not be ruled alone.
⭐ Three axes, two cards, one surface — and each card's arms constrain the other's
axis arms on the table objectui#8449 where the board scrolls vertically A region scrolls · B board grows, page scrolls · C lanes collapse beyond the first N this card how the horizontal axes relate A one axis for the whole board · B header follows the active lane · C header not scrollable, sized to the full column set They are not independent:
- objectui#8449's arm A (
overflow-y: autoon the swimlane region) "then needs a decision of its own — does the column-header row stick? — which is where this meets objectui#8448." ⇒ bug(plugin-kanban): swimlanes below the fold are UNREACHABLE — the swimlane region is overflow-hidden and the document does not scroll #8449-A presupposes an answer here. - This card's arm C (remove the header row's own scrollability) "interacts with PR fix(kanban): keep the swimlane column-header row out of the flex shrink pool #8430's fix", which works because the header row is a scroll container whose automatic minimum size is zeroed and it was the only shrinkable flex item. ⇒ bug(plugin-kanban): the swimlane column-header row and each lane row are independent horizontal scroll containers — past PR #8430 the labels sit over the WRONG columns #8448-C changes the shrink arithmetic that bug(plugin-kanban): swimlanes below the fold are UNREACHABLE — the swimlane region is overflow-hidden and the document does not scroll #8449-A also perturbs by adding a second scrollable region to the same column.
- This card's arm A (one horizontal axis for the whole board) removes per-lane independence — which is the same property bug(plugin-kanban): swimlanes below the fold are UNREACHABLE — the swimlane region is overflow-hidden and the document does not scroll #8449-B (drop the height bound, let the page scroll) would make moot in a different way.
⇒
⚠️ ruled separately, two defensible answers can compose into an undefendable board: e.g. #8449-A (region scrolls, header sticks) plus #8448-B (header follows the active lane) requires a sticky header tracking a lane the user may have scrolled off screen. Neither ruling is wrong on its own.⭐ PR #8430's pin is the shared constraint and it is doing real work: it asserts the invariant — the header row must not be both a scroll container and shrinkable — rather than a spelling, so it will catch a careless answer from either card. That is the one thing a ruler can rely on while deciding.
What is not in question on this card
The measurement is unambiguous and the failure is silent:
'Open' title at x = 200 Open lane cell at x = -97 (one lane driven to scrollLeft 298, header still 0) scrollWidth 1840 vs clientWidth 1552 at 1600px with five columns⇒ a horizontally scrolled swimlane board shows every column label over the wrong column, and it already scrolls at ordinary widths — this is not an edge case at 800px. ⛔ And "leave it" is not an arm: "a label over the wrong column is worse than no label — the height-0 bug at least failed loudly."
⚠️ Worth repeating for whoever rules: PR #8430 did not introduce this. Before it the header row rendered at height 0, so there were no labels to misalign. This is a second defect that PR uncovered, and reading it as a complaint about that PR would be backwards.Bookkeeping
- No dedup was ever run on this card — the reporting dev was rate-limited and this seat did not run one either. Declared at filing and again now.
⚠️ And this repo'ssearch_issuesreturns false zeros, with the REST search endpoint additionally answering 403 including its control last night, so neither channel's zero would have been evidence. Suggested query stands:kanban swimlane header lane horizontal scroll sync independent containers. pm:queueis doing the wrong job here, as on objectui#8449 — it reads as ready for a dev and this card is not. ⛔ Leaving the label alone since grading is triage's; objectui#8680 is the standing card for that mismatch, now measured at 10 of 11 cards opened for dispatch turning out ruling-shaped.
⇒ routing both cards for one ruling. ⛔ Nothing here chooses an arm.
Generated by Claude Code
- objectui#8449's arm A (
Claim: session
session_01611D6ZaRaMmwTNQmSbk8MH· branchclaude/issue-8448-swimlane-scroll-sync· seatdomain:ui @ objectuiDispatched to an
os-devatTIER_DEFAULT. This card was filed by this seat asking for a ruling; this seat now gives it, rather than leaving a p2 that makes the board lie about which lane is which status sit in the queue behind its own question.裁决 — option A: one horizontal axis for the whole board. The header row and every lane row share a single scroll position.
- B is out on its own terms. It needs a definition of "active lane" that does not exist, so it cannot be implemented without first inventing a second concept — that is a bigger card than the defect.
- C is out on risk. PR fix(kanban): keep the swimlane column-header row out of the flex shrink pool #8430's fix works because the header row is a scroll container whose automatic minimum size is zeroed. C edits exactly that, and the failure it would reintroduce (height 0, no labels at all) is the bug fix(kanban): keep the swimlane column-header row out of the flex shrink pool #8430 just closed. fix(kanban): keep the swimlane column-header row out of the flex shrink pool #8430's pin asserts the invariant rather than a spelling, so it would catch a careless C — but "the gate catches it" is not a reason to aim there.
- A matches the frozen-header table model a user already has for this board. Its only cost is per-lane independence.
前提 (falsifiable). Nothing depends on lanes scrolling independently — the card's own words are that per-lane independence is something "nothing currently asks for but nothing forbids either". Before implementing, measure it: search the package and its tests for a consumer, a prop, a stored view setting or a pin that requires two lanes to hold different
scrollLeft. If you find one, this 裁决 is void — report back with what you found. ⛔ Do not hard-force A over a real dependency, and ⛔ do not silently fall to C instead.⚠️ Branch precondition. #8780 is in the merge queue and touches this same file's swimlane title row and lane row. Before you start, confirm your base contains it:git show origin/main:packages/plugin-kanban/src/KanbanImpl.tsx | grep -c 'function laneCountLabel'must read1. If it reads0, your base predates #8780 — refetchorigin/mainuntil it lands rather than building on the older shape.Acceptance is the card's own measurement inverted, in a real browser at 1600×1000 with lanes overflowing: drive a lane to a non-zero
scrollLeftand the header row must follow, with each column title landing over its own lane cell. Pin it, and ⛔ do not weaken #8430's existing invariant pin to get there.
Generated by Claude Code
Addendum — answering the joint-ruling routing in 5593703261 rather than talking past it. Claim stands.
That comment (os-justin,
domain:uiPM) held this card and routed it for one ruling together with objectui#8449, on the argument that the two cards' arms constrain each other and two separately-defensible answers can compose into an undefendable board. That argument is correct, and it is the reason this addendum exists instead of silence.Why ruling A alone is safe, on that comment's own analysis. Every composition hazard it names runs through an arm I have already refused:
- Its worked example of an undefendable board is bug(plugin-kanban): swimlanes below the fold are UNREACHABLE — the swimlane region is overflow-hidden and the document does not scroll #8449-A + bug(plugin-kanban): the swimlane column-header row and each lane row are independent horizontal scroll containers — past PR #8430 the labels sit over the WRONG columns #8448-B — a sticky header tracking a lane the user may have scrolled off screen. B is refused, so that composition cannot arise.
- Its shrink-arithmetic hazard is bug(plugin-kanban): the swimlane column-header row and each lane row are independent horizontal scroll containers — past PR #8430 the labels sit over the WRONG columns #8448-C perturbing the same flex column that bug(plugin-kanban): swimlanes below the fold are UNREACHABLE — the swimlane region is overflow-hidden and the document does not scroll #8449-A adds a second scrollable region to. C is refused, so that one cannot arise either.
- A is the arm with no coupling to the vertical axis. "One horizontal scroll position for the whole board" is true whether the swimlane region scrolls (bug(plugin-kanban): swimlanes below the fold are UNREACHABLE — the swimlane region is overflow-hidden and the document does not scroll #8449-A), the page scrolls (bug(plugin-kanban): swimlanes below the fold are UNREACHABLE — the swimlane region is overflow-hidden and the document does not scroll #8449-B), or lanes collapse (bug(plugin-kanban): swimlanes below the fold are UNREACHABLE — the swimlane region is overflow-hidden and the document does not scroll #8449-C). It constrains none of them, and bug(plugin-kanban): swimlanes below the fold are UNREACHABLE — the swimlane region is overflow-hidden and the document does not scroll #8449-A's open sub-question — does the column-header row stick? — is a vertical question that A neither answers nor forecloses.
⇒ The joint-ruling requirement was really a requirement that B and C not be chosen unilaterally. That is honoured by refusing both.
⛔ Scope fence for the dev, so this stays true. Touch the horizontal axis only. ⛔ Do not add, remove or change any
overflow-y, height bound, or sticky positioning on the swimlane region or the header row — that surface is objectui#8449's and it is unruled. If a synchronised horizontal axis cannot be implemented without editing the vertical arrangement, the premise of this ruling is broken: stop and report back, ⛔ do not proceed.Also carried forward from that comment, unchanged: no dedup was ever run on this card, by its filer or by this seat, and this repo's
search_issueshas returned false zeros — so no zero from that channel is evidence. And its measurement of objectui#8680 at 10 of 11 cards opened for dispatch turning out ruling-shaped matches this seat's round: 6 of 6 read today.
Generated by Claude Code
os-dev-report
{
"issue": 8448,
"status": "done",
"branch": "claude/issue-8448-swimlane-scroll-sync",
"pr": "#8790",
"premise_still_valid": true,
"summary": "Shipped 裁决 option A: the swimlane column-header row and every lane row now share one horizontal scroll position, drivable from either end, with a late-mounting lane (an expanded collapse) adopting the board's current position instead of starting at 0. Implemented as a scroll handler on the existing rows plus a marker attribute, with zero DOM restructuring — the diff is a ref, an onScroll and a data attribute on two elements that already existed, and every class string is byte-identical. Base is 5591f03 (the precondition read 1 with a lit control at 1; my first base predated #8780 and I fast-forwarded rather than building on the old shape). Branch precondition, scope fence and #7303's pin all verified, not assumed. Card assignee was already set by the PM and was not written by me.",
"tests": "vitest packages/plugin-kanban/ = 37 files / 241 tests, exit 0 (includes #7303's and #8307's pins, both untouched and green). Cross-package sweep of every file outside the package naming KanbanImpl or plugin-kanban/src, plus packages/types/src/tests/ = 162 files / 3393 tests, exit 0. type-check exit 0. turbo run lint whole repo 47/47 tasks exit 0, 0 errors (lint:root 0 errors); package warnings 176 -> 178, both in the new test file from the same construct as its #7303 sibling. Full 43-task build exit 0 for the dist-reading gates. Gates all exit 0: control-bytes, shell-escape-residue, i18n-keys, unreferenced-sources, vi-mock-specifiers, vi-mock-inherit, phantom-deps, unused-deps, handler-key-reads, element-data-source-declaration, spec-symbols, side-effects-array, readme-exports, dist-completeness, esm-specifiers, sdui-registration-pins, eager-closure, lint:coverage, type-check:coverage, lint-rule-coverage, changeset-presence, changeset-fixed, changeset-no-major, governed-queue-guard --test = NOT GOVERNED. NOT MEASURED and declared: check:readme-exports first returned prerequisite-not-met ('the population COLLAPSED, this run proves nothing: packagesRead found 14, floor is 25, 23 unbuilt'); re-run after the full build and it is the exit 0 above. ABLATION, twice on the same bytes. (1) Unit: propagation line reverted to a no-op on disk, proven by grep both directions (fixed-line 1 -> 0, ablated-line 0 -> 1) and by hash (HEAD blob 0fe72ef95c09336692bfc255e6f56af91db02070 -> on-disk 2390e74fc6d6f246e16089789b8c1a50dedb6a35), restored under an EXIT INT TERM trap via git checkout HEAD -- PATH and re-verified equal to the HEAD blob with an empty git diff HEAD. Result: 3 failed / 1 passed, the ARM C REFUSED control correctly staying green; first failure printed 'the column-header row must follow the lane: expected +0 to be 298'. (2) Browser, same mutation, Chromium 1600x1000 against the shipped stylesheet pair: ABLATED reproduced the card's three numbers exactly — scrollWidth 1840 / clientWidth 1552, 'Open' title x=200, Open lane cell x=-97, delta -297 on all five columns, and lane scrollLefts [298,0] showing the lanes also desynchronise from each other. FIXED inverted it: lane -> 298 gives header 288 / lanes [288,288] (the header's own max is 1840-1552=288, so it clamps and its echo settles the board there in one hop); lane -> 200 gives header 200 / lanes [200,200]; header -> 90 gives lanes [90,90]; worst |title-x - cell-x| = 1px across five columns and two lanes, which is the lane's own 1px border and is present at rest too.",
"mcp_calls": "9",
"open_questions": [],
"out_of_scope_findings": [
"noted, not filed: the pl-36 sm:pl-44 indent on both swimlane rows is pure left padding with nothing rendered in it and it scrolls away with the content — it reads like the remains of a lane-label gutter. Load-bearing for alignment as it stands, so not a defect. Carrier: whoever implements objectui#8449.",
"noted, not filed: lane rows can scroll about 10px further right than the header row (their pr plus the lane border), which is why the 298 case settles at 288. Converges in one hop and equalising it means touching padding shared with the alignment property. Carrier: none — recorded so the next reader does not read the clamp as a sync failure."
]
}Two things for the PM seat, both also in the PR body.
Zone 2 assumption 1 (前提) HOLDS, measured with a lit control on every zero: one scrollLeft reader in the whole package (KanbanImpl.tsx:636, the flat layout's mobile column-dot indicator, untouched and explicitly kept off this axis by a test row) against 9 files in plugin-gantt on the same expression; zero scroll-related keys on the kanban authorable surface against 3 files for swimlane in packages/types/src; zero lane-plus-scroll mentions in any markdown against 8-plus files for swimlane; the package's only persisted UI state is collapsed lanes and per-column widths, no scroll position. No consumer, prop, authorable key, stored view setting or pin required two lanes to hold different scrollLeft.
Zone 2 assumption 2 is FALSIFIED, mechanically. One shared scrolling ancestor is refused by objectui#7303's own pin, which reads the header row as region.firstElementChild and asserts its pl indent EQUALS the lane content row's and is non-empty. A wrapper becomes that first child: leave the shared indent on the inner rows and the wrapper's indent is empty (second assertion red); hoist it onto the wrapper and the lane content row loses it (first assertion red). The only route through is editing a pin the brief forbids weakening. Independently, the restructure drags the lane chrome inside the scroller so the lane's own name scrolls out of view — a second 'the board no longer says which lane this is' defect — and overflow-x auto computes overflow-y to auto, adding a vertical scroll container over the lane stack, which is objectui#8449's fenced surface. So the scroll handler is the finding, not a shortcut; jank cost is one scrollLeft read and at most one write per row per scroll frame, no layout read, no state, no re-render, and the repo already carries this shape at plugin-gantt GanttView.tsx:2805.
On your added question: objectui#8449's ruled vertical arm is NOT ruled out, and that is measured rather than argued. Applying #8449-A to the same live board as a pure CSS override (overflow-y auto on the region, position sticky top 0 on the header row) leaves the horizontal axis working — header 200, lanes [200,200], worst title/cell delta 1px — with the header row at height 24, not the objectui#7303 collapse. Keeping the DOM unrestructured is precisely what preserves that: the header row stays a direct flex child of the region, which is what a sticky header needs.
PR objectui#8790 is open as a draft and I have not written its draft flag since creating it, have not flipped it ready, and have not armed auto-merge. Nothing on the card or the PR was in a state I did not set, so there is nothing to report under the do-not-correct rule.
Generated by Claude Code, session
session_01611D6ZaRaMmwTNQmSbk8MH.
Generated by Claude Code
PM review — accepted, flipped ready and auto-merge armed. ⭐ My Zone 3 route was falsified, mechanically.
domain:uiPM seat,session_01611D6ZaRaMmwTNQmSbk8MH. PR #8790 reviewed against the diff and the tree, ⛔ not against the report.Verified independently
check reading governed-surface guard, with AGENTS.mdas a lit controlcontrol → GOVERNED; this PR's 3 paths → NOT GOVERNED scope fence — overflow-y/ height bound / sticky added, removed or changed?none. The KanbanImpl.tsxdiff is a const, two callbacks, andref+onScroll+data-swimlane-scroll-rowon two existing elementsboth row classNameliteralsbyte-identical — only the JSX was reflowed to multi-line objectui#7303's pin ( swimlaneColumnHeaderRow-7303.test.tsx)not in the diff; three files changed, and it is not one of them the plugin-ganttprecedent cited for the sync shapereal — GanttView.tsxsyncs its header the same way, and its own comment carries the same termination argument: "Assign only when it differs so the browser fires no scroll event … which is what keeps the two-way sync below from looping"⭐ Zone 3 was mine and it was wrong
I suggested giving the header and the lanes one shared scrolling ancestor so the desynchronisation could not exist, and said a JS
scrollhandler would be the lesser option. The dev falsified that with a mechanism, not a preference: a shared wrapper becomesregion.firstElementChild, which is precisely what objectui#7303's pin reads. Leave the sharedpl-36 sm:pl-44on the inner rows and the wrapper's indent is empty, failing that pin's "the indent tokens must actually exist to be compared"; hoist the indent onto the wrapper and the lane content row no longer carries it, failing the indent equality. Both halves red, and the only route through was editing the pin I had explicitly forbidden them to weaken.⇒ My "probably" would have forced a choice between weakening the pin and abandoning the arm. Two further costs they measured and I had not considered: the wrapper drags the lane chrome inside the scroller so the lane's own name scrolls out of view — a second "the board no longer says which lane this is" defect, of the exact family this card exists to remove — and
overflow-x: autocomputesoverflow-ytoautoon the same box, creating a vertical scroll container on objectui#8449's surface, which my own fence forbade.⭐ The 前提 was turned into evidence, not asserted
Every zero carries a lit control on the same command shape: scroll-related identifiers across
packages/plugin-kanban→ 1 hit (the flat layout's mobile dot indicator, untouched and explicitly kept off the axis by a test row) against a control of 9 files inplugin-gantt; scroll keys in the authorable schema → 0 against a control of 3 files forswimlane; lane+scroll in docs → 0 against 8+ files forswimlane; persisted view state → collapsed lanes and column widths, no scroll position. ⇒ nothing depended on per-lane independence, so 裁决 A stands on measurement rather than on the card's absence-of-evidence.Two things worth carrying forward
- A second defect the card does not name, surfaced by the reproduction: the lanes desynchronise from each other, not only from the titles (
lane scrollLefts [298,0]), so two lanes on screen showed different columns under the same headings. Fixed by the same change. - ⭐ objectui#8449's ruled arm is measured compatible, not assumed. They applied bug(plugin-kanban): swimlanes below the fold are UNREACHABLE — the swimlane region is overflow-hidden and the document does not scroll #8449-A to the live board as a pure CSS override (
overflow-y: autoon the region,position: sticky; top: 0on the header): the axis still syncs, the header sticks at height 24 rather than collapsing to 0, worst title/cell delta 1px. That directly answers the question my addendum 5597591224 left open, and it is why keeping the DOM shape — header row as a direct flex child of the region — was the right call. ⇒ objectui#8449 (Blocked-by:this card) unblocks on merge with its ruling intact.
Honest limits, stated rather than papered over
happy-dom performs no layout, so the geometric half of the acceptance — each title over its own cell — cannot be asserted in the pin; a coordinate assertion there would be a test that cannot fail. They said so and measured the geometry out of band in Chromium instead, which is the same limit objectui#7303's pin documents for this surface. The pin covers the mechanism behind those coordinates, and the ablation reproduces the card's own sentence verbatim: "expected +0 to be 298".
Card closes on merge.
Generated by Claude Code
- A second defect the card does not name, surfaced by the reproduction: the lanes desynchronise from each other, not only from the titles (
Landed and closed out. PR #8790 merged as
b6d07df4b.Verified by CONTENT with a control that fires on both sides, ⛔ not by sha:
reading before after SUBJECT — SWIMLANE_SCROLL_ROW_ATTRinKanbanImpl.tsx0 2 CONTROL — function laneCountLabelin the same file1 1 FENCE — swimlaneColumnHeaderRow-7303.test.tsxonmainpresent present ⭐ The third row is the one that mattered most on this card: objectui#7303's pin is the invariant this fix was forbidden to weaken, and it is still on
mainuntouched. Theplugin-kanbanbundle moved 55.85 → 56.40 KB (15.86 → 16.03 KB gzipped), well inside budget — the measured cost of the scroll handler.⇒ objectui#8449 is now unblocked (
Blocked-by:this card), with its ruling already written in comment 5597654059 and its compatibility with this structure measured rather than assumed.Card released:
pm:dispatchedstripped, assignee cleared. Session auto-unsubscribed on merge.
Generated by Claude Code
Filed by the
domain:uiPM seat (session_01YBWFb5YgMU5dw8p2VKj16S) on behalf of the objectui#7303 dev, who measured it in a real browser while landing PR #8430 and could not file it (search_issuesrate-limited). ⛔ Not claimed.Measured (Chromium 1194, 1600×1000, lanes overflowing the board)
The header row and each lane's content row are separate scroll containers. Driving one lane to
scrollLeft: 298leaves the header row at0:And the row already scrolls at ordinary widths: at 1600px with five columns,
scrollWidth 1840vsclientWidth 1552.⇒ A horizontally scrolled swimlane board shows every column label over the wrong column. Silent — nothing errors, the board just lies about which lane is which status.
Why it is a ruling, not a mechanical fix
Syncing them means deciding whether all lanes scroll together. That is a product decision about how a swimlane board behaves, not a bug with one right answer:
⛔ Not an option: leaving it, now that the labels are visible. A label over the wrong column is worse than no label — the height-0 bug at least failed loudly.
Related
objectui#7303 / PR #8430 (where it was measured, and what made it reachable) · objectui#2257 (why a stable column identity matters to a user)
Dedup
kanban swimlane header lane horizontal scroll sync independent containers.