Skip to content

feat(sidebar): show full session titles while the list is scrolled - #15224

Merged
iamwhatever merged 1 commit into
kirodotdev:mainfrom
Pearcekieser:feat/sidebar-widen-on-scroll
Oct 13, 2026
Merged

iamwhatever merged 1 commit into
kirodotdev:mainfrom
Pearcekieser:feat/sidebar-widen-on-scroll

Conversation

@Pearcekieser

@Pearcekieser Pearcekieser commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: While the sessions list is scrolled, long session titles can be read in full without resizing the sidebar.

Session titles are one line and truncate. At the default 260px width most titles are cut off, and the only way to read them is to hover each row for its tooltip or drag the sidebar wider and back again.

Why it matters

Scanning the list is how people find a session. A list of titles that all end in an ellipsis after four words is hard to scan, and a permanently wider sidebar costs chat width all the time.

Not a goal

  • No change to the stored sidebar width, the row width, the resize handle, or the board view.
  • No wrapping or two-line titles; rows keep one line and one height.
  • No change on mobile, where the list is a full-width drawer.

What changed (motivation → approach → change)

When you scroll the sessions list (wheel, touch, or arrow / Page / Home / End keys), every title in view that its row cuts off by 16px or more is drawn in full in a floating layer laid exactly over the row. The rows never change: the card keeps the user's stored width, nothing scrolls sideways, and the session-row-fixed-height rule in website/AUTOSDE.yaml holds as written. The layer is click-through, so a click on a shown title selects the row under it. No title is drawn for the row under the pointer (its hover actions stay visible), for a title behind a stuck folder header, or for a title whose start (inside the sidebar) another element covers. That check samples only the title's start, so an element over just the part drawn past the sidebar's edge is not detected; the 4s ceiling bounds it. While it is up, every lane scroll and every change to the sidebar's rows (a mount, a rename, a re-sort slide) re-measures, so it follows the virtualized rows. It ends at once on a press or a non-scrolling key anywhere on the page, or when the window loses focus. Otherwise it stays while the pointer is over the sidebar, up to 4s after the last scroll input, and hides 150ms after the pointer leaves or 0.7s after the last touch or key input. The layer is portaled above the dashboard shell at z-[65], so a popover or banner that opens on its own while a peek is up can sit under the titles for up to those 4s. It is off by default and desktop-only (the reason is stated at each place that switches it off on a phone): Settings → Chat → Sessions → Show Full Titles While Scrolling turns it on.

Piece What it does
website/src/pages/chat-sidebar/scrollPeek.ts useSidebarScrollPeek: listens for wheel, touchmove and scrolling keys in the lane (ignoring zoom wheels, horizontal wheels, caret keys in a text field and a lane that cannot scroll), measures the clipped titles fully in view below the search dock on the next frame, follows lane scrolls, and runs the hide timers. useFullTitlesOnScrollSetting reads the setting.
website/src/pages/chat-sidebar/ScrollPeekLayer.tsx The layer: portaled to document.body, position: fixed, pointer-events: none, aria-hidden; each title starts on the row's own title and is capped at the window edge.
website/src/pages/ChatSidebar.tsx, website/src/pages/ChatPage.tsx The sidebar renders the layer and gains an opt-in scrollPeek prop that the desktop drawer passes; the card's width is untouched.
website/src/pages/chat/ChatSettings.tsx, website/src/pages/settings/ChatPanel.tsx The fullTitlesOnScroll chat setting (default off) and its toggle; every catalog under website/src/i18n/locales/ carries its two strings, and the settings registry (settingsRegistry.gen.ts, src/kiro_crew/docs/settings-registry.generated.json) is regenerated so search and deep links find it.
docs/feature-map/README.md, docs/system-specs/modules/history.md List the two new files on the Sessions row and in the sidebar owner table.

Backwards compatibility

Compatible: scrollPeek is a new optional prop and the setting defaults off, so every sidebar renders exactly as before until a user turns it on.

Tests

  • website/src/pages/chat-sidebar/scrollPeek.test.tsx (21 tests):
    • measureClippedTitles picks only titles clipped past the threshold and fully in view below the dock, capped at the window edge.
    • A lane wheel shows the clipped titles on the next frame and never changes the root width.
    • Input outside the lane, horizontal and zoom wheels, an unscrollable lane, and a disabled hook show nothing; turning it off ends a live peek.
    • A lane scroll re-measures row positions. The pointer keeps it up, a leave or any press ends it, touch is not hover, scrolling keys work, and caret keys in a text field do not.
    • It skips the row under the pointer (re-resolved from the pointer position on each measure), a title behind a stuck folder header, and a title whose start another element covers.
    • It follows rows that move without a scroll: a DOM change, and a re-sort slide that moves rows by style alone.
    • A press or non-scrolling key anywhere on the page, window blur, and the 4s idle ceiling each end it.
    • The layer renders nothing when empty, and is click-through and hidden from assistive technology when not.
  • website/src/test/ChatSidebar.scrollPeek.test.tsx renders the real sidebar, scrolls a lane of clipped titles, and asserts the layer is up while the card stays at 260px and no row gets a width.
  • website/src/test/ChatSidebar.ownerComposition.test.ts pins the two new owner hooks in the facade's call order.

The 512 related frontend test files pass locally.

Manual verification

Built the dashboard from this branch and served it from an isolated dev gateway with 32 long-titled sessions, driven by a Playwright script at 1440×900 with the setting turned on. While scrolling with the pointer over the list, 13 titles were shown in full and the sidebar stayed at 260px; 1.5s after the pointer moved to the chat pane the layer was gone and the sidebar was still 260px.

Screenshots / video

At rest, titles truncated:

Sessions list at rest with truncated titles

While scrolling, clipped titles are shown in full over their rows; the sidebar keeps its width:

Full titles shown over the rows while scrolling

The setting, under Settings → Chat → Sessions (shown on; it defaults off):

Settings Sessions tab showing Show Full Titles While Scrolling

After the pointer leaves, the layer is gone

Sessions list after the pointer leaves

Scroll, read, move away:

Related Issues

no linked issue: requested directly by the author; no tracked issue exists.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

@Pearcekieser
Pearcekieser requested a review from a team September 30, 2026 01:02
@Pearcekieser
Pearcekieser requested a review from a team as a code owner September 30, 2026 01:02
@Pearcekieser

Copy link
Copy Markdown
Contributor Author

Intent: Let a user read long session titles by scrolling the sessions list: the card widens over the chat pane while it is scrolled and settles back when the pointer leaves, without changing the stored sidebar width or reflowing the chat.
Not a goal: Title wrapping, a new setting, any change to the resize handle, board view or mobile drawer.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author

Note on pre-push review: the model-pinned local reviewers (GPT and Opus lanes) and the cold-reader check could not start on the author's host (every subagent failed with Request initialize timed out after 30s). The pre-push pass was a self-review against the extracted CI contracts, not an independent model review. It found and fixed three items before this push: width transition leaking onto non-peek width changes, an unthrottled per-wheel-event layout read, and ctrl+wheel zoom triggering the peek.

@github-actions github-actions Bot added fork Pull request from a fork (external contributor) readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Sep 30, 2026
@Pearcekieser
Pearcekieser force-pushed the feat/sidebar-widen-on-scroll branch from b996429 to 37995b0 Compare September 30, 2026 01:30
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention labels Sep 30, 2026
@github-actions

github-actions Bot commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Opus 5.5, fork) — ✅ PASS

Design-level review of f49fd4955c3c665630bf5407fe2ffe35c977551a via the fork AI-review pipeline — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Check the z-[65] layer against the shell's own overlays (toasts, approval banners, dialogs): one that opens on its own during a peek can sit under the titles for up to 4s.

[DESIGN-REVIEWED] f49fd49

Description read

[DESCRIPTION-READ] 1b5b2ab1db869af18d295a84d23979eb146f327cfc79d2cf1a3bad651b21d2fe

That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Opus 5.5, fork) — 🟡 CONCERNS

UX-level review of f49fd4955c3c665630bf5407fe2ffe35c977551a via the fork AI-review pipeline — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

UX-Verdict: CONCERNS

In scrolling.png, a peek title covers the hovered row's action buttons, and chips run over the composer. No first-time reader has seen any of these screens.

Watch

  • Peek title hides hover actions. scrolling.png, row at y≈478: the "Fix composer paste highlighting…" chip sits on top of the row's ⋮ / copy / × toolbar, which was showing for "Rebase…" in before.png. The browser's hover state doesn't update during a wheel scroll, but rowAtPoint only skips a row that is under the pointer when it measures. So the toolbar stays visible and clickable while a chip half-covers it. Users then click a control they can barely see. Severity: medium, because it happens every scroll that ends with the pointer resting on the list. Smallest fix: also skip the row that holds the hover toolbar (the hover-hold seat), not only the row under the pointer.
  • Chips cover the composer. In scrolling.png, the "Design review…" and "Clean up stale worktrees…" chips are drawn over the composer's border, and one of them covers its "Normal" chip, because the layer sits at z-[65]. Clicks pass through, but for up to 4s the user reads two overlapping texts where they type. Severity: low-medium, because it only happens when the lowest rows are clipped. Fix: don't draw a chip whose bounds overlap the composer, or cap each chip's width at the chat pane's content start for bottom rows.
  • Comprehension is unverified. The setting toggle, the peek layer and the before/after states each appear in a committed screenshot, and demo.webm exists. But no blind read has checked that a first-time user understands floating chips over rows, or the label "Show Full Titles While Scrolling".

Suggestions

  • Change full_titles_on_scroll_description to "Long session titles show in full while you scroll the list". The timing details ("hide when the pointer leaves…") are mechanism, not decision-critical, so they can go.

[UX-REVIEWED] f49fd49

Description read

[DESCRIPTION-READ] 4bf9141217099f5c7b40abf349e5ebac10e23f468af90a8d8cd8c8b7d32f050e

That sha256 names everything this verdict read: the PR description, plus the 7 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

GPT 6.1 Review (fork) — ✅ no blocking findings

Reviewed f49fd4955c3c665630bf5407fe2ffe35c977551a via the fork AI-review pipeline; updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] f49fd49

@github-actions

github-actions Bot commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5, fork) — 🟡 CONCERNS

Premise-level review of f49fd4955c3c665630bf5407fe2ffe35c977551a via the fork AI-review pipeline — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

First-Principles-Verdict: CONCERNS

The new scrollPeek prop does the same job as the existing fillsHost prop: both are set on exactly the two hosts that have no chat pane beside them.

Not justified as shipped

  • Item 4 — duplicate of fillsHost (website/src/pages/ChatSidebar.tsx:2297): the new prop says "the embedded view has no chat beside it either". fillsHost's existing docstring gives the same reason: "the mobile drawer, the sessions embed… reserves no chat pane beside it". I counted the 2 <ChatSidebar call sites. The embed passes fillsHost and no scrollPeek. The drawer passes fillsHost={isMobile} and scrollPeek={!isMobile}. The hook then checks !isMobile a third time.
    Subtraction: delete the scrollPeek prop and the scrollPeek={!isMobile} line in ChatPage, and gate the hook on fullTitlesOnScroll && !fillsHost && !boardLaneActive.
    Clears when: the scrollPeek prop is removed and the peek is gated on !fillsHost.

What this change ships

Inventory (6 items) — 5 justified

Intent: let a desktop user read long session titles in full while scrolling the list, without widening the sidebar. This is an ADDITION.

  1. New toggle Settings → Chat → Sessions → Show Full Titles While Scrolling, off by default — justified
  2. While the list scrolls, cut-off titles are drawn in full in a click-through layer over the chat — justified
  3. The layer ends on any press, any non-scrolling key, window blur, pointer leave, or the 4s ceiling — justified
  4. Peek is off on the phone drawer and the embedded Sessions view, via a new scrollPeek prop — duplicate of fillsHost
  5. The setting is findable through command-palette search and the find_ui index (regenerated) — justified
  6. Review screenshots and a recording committed under temp-screenshots/ — justified

[FIRST-PRINCIPLES-REVIEWED] f49fd49

Description read

[DESCRIPTION-READ] a5551dd410c496e550950f580577fbb27c987f5fcdc838175cc37ec7d6d5fdc1

That sha256 names the PR description this verdict read. Recompute it with .github/scripts/pr-description-capture.sh; if it differs, the description changed after this verdict was formed and any finding drawn from the description is unproven.

@github-actions

github-actions Bot commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review (fork) — ✅ no blocking findings

Reviewed f49fd4955c3c665630bf5407fe2ffe35c977551a via the fork AI-review pipeline; updated in place on each push.

Review details

No findings that block the merge.

FINDING — website/src/pages/chat-sidebar/scrollPeek.ts:989 — when a running session's title is clipped, the chip still lands 13px left of the row's own text, so it does not sit exactly over the title as ScrollPeekLayer claims. The [data-session-title] div also holds the inline .session-live-dot (7px wide plus a 6px right margin) or the title-regenerating Loader before the text. The chip takes left: r.left and width: el.scrollWidth from that whole div, but its text comes from text: el.textContent ?? '', which carries no glyph. The opaque chip therefore covers the running marker, and it is wider than its text by the same 13px → Fix: measure the title's text node with a Range (range.selectNodeContents(textNode).getBoundingClientRect()) and use that for left and width, or draw the chip starting after the leading glyphs.

[OPUS-REVIEWED] f49fd49

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Sep 30, 2026
@Pearcekieser
Pearcekieser force-pushed the feat/sidebar-widen-on-scroll branch from 37995b0 to a871102 Compare September 30, 2026 02:15
@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Repeated wheel events compound overflow during the width transition

outcome: fixed
Fixed in a871102: the hook now takes the card's resting width and counts only the part of the peek that has rendered (root width minus base), so the target is absolute: next = min(cap, rendered + more).
A wheel event mid-transition re-derives the same target instead of adding the residual clip again.
Regression test: does not compound the clip while the width transition is still running (five wheels at 60ms spacing stay at 52px).

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Lowered cap after a viewport resize keeps the old overhang

outcome: fixed
Fixed in a871102: when the current extra exceeds the freshly read ceiling, the next scroll input calls apply(cap).
Regression test: shrinks a live peek to a ceiling a window resize lowered.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Resize grip on a peeking card snaps and trails the cursor

outcome: fixed
Fixed in a871102: a drag now starts from sidebarWidth + peekExtra (clamped to SIDEBAR_MAX), the edge the user grabbed, so the first move lands under the cursor instead of snapping back by the peek width.
A click on the grip without moving still persists the stored width unchanged.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Mid-transition re-measure ratchets the peek to the cap

outcome: fixed
Fixed in a871102, same change as the GPT lane's blocker: the measure subtracts the rendered peek, so the target no longer grows with each throttled measure.
Test: does not compound the clip while the width transition is still running.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Touch at desktop width leaves the strip live for 700ms after lift

outcome: rebutted
The strip that stays for 700ms is the visible, widened card, not an invisible hit area: a tap there lands on the rows the user can see, like any overlay that is on screen.
Collapsing on finger lift would end the peek before the user can read the titles it opened, which is the point of the feature on touch.
A tap on a row now collapses at once (a871102), so choosing a session never leaves the card over the chat.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Always-on default when most titles truncate

outcome: fixed
Addressed in a871102: an opt-out ships. Settings → Chat → Sessions → Widen List While Scrolling (widenSessionsOnScroll, default on) turns the peek off; the sidebar reads it live via mc-config-changed.
The default stays on because the requester asked for this behaviour; a maintainer can flip the default in one line in ChatSettings.tsx DEFAULTS.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Collapse the peek when a session row is clicked

outcome: fixed
Fixed in a871102: a click inside the lane collapses the card at once, so opening a session gives the chat its width back immediately.
Test: collapses at once when a row is clicked.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Keyboard scrolling never widens the card

outcome: fixed
Fixed in a871102: ArrowUp/ArrowDown/PageUp/PageDown/Home/End inside the lane are scroll input too, and focus inside the sidebar keeps it wide the same way pointer presence does; blur starts the collapse timer.
Test: widens on keyboard scrolling and stays wide while focus is inside.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Feature-map text misstates the key-input stop

outcome: deferred
Correct: onScrollIntent schedules the 0.7s stop only when no pointer is inside the sidebar. The follow-up PR rewords the Sessions row to "0.7s after the last touch or key input when no pointer is over the sidebar". This head is a rebase-only retrigger, so the docs change ships there rather than resetting these verdicts.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author

chore (non-blocking): rebased on main as e4e9b84, no behaviour change. The three reds on be687ac came from outside this diff. Frontend Tests shard 1 failed in focusRingContrast.test.ts on AskAnswersCard.tsx:41 and in ChatPage.dirStaging.test.tsx; main fixed both in #18580 and #17929, and Frontend Coverage Merge failed only because shard 1 did. SAST (Semgrep) never scanned: Docker Hub refused the semgrep/semgrep:1.78 pull with toomanyrequests. The rebase had no conflicts, the PR diff is still 31 files, +5858/-5, and tsc, the feature-map gate and the two failing test files (73 tests) pass locally.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Every row becomes a floating card

outcome: deferred
The follow-up PR drops the border, shadow and semibold weight, keeps only the elevated background, and stops each label at the chat pane edge so it no longer spills over the chat area. This head is a rebase-only retrigger, so the change ships in the follow-up PR rather than resetting these verdicts.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • A peek label hides a hover toolbar

outcome: deferred
Agreed. The follow-up PR skips the row useHoverHold is holding as well as the row under the pointer, and retakes attachment-03 against that build. This head is a rebase-only retrigger, so the change ships in the follow-up PR rather than resetting these verdicts.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Wrong stacking context for the layer

outcome: deferred
Agreed. The follow-up PR mounts the layer in the app shell's overlay slot below z-[70] and adds a case to AppRootDialogZIndex.test.tsx pinning that a shell dialog paints above it. This head is a rebase-only retrigger, so the change ships in the follow-up PR rather than resetting these verdicts.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Setting description wording

outcome: deferred
Agreed. The follow-up PR rewrites full_titles_on_scroll_description to name the session list and say titles stay while the pointer is over it. This head is a rebase-only retrigger, so the change ships in the follow-up PR rather than resetting these verdicts.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • scrollPeek duplicates fillsHost

outcome: deferred
Agreed: at both call sites scrollPeek equals !fillsHost. The follow-up PR deletes the prop and gates the hook on fullTitlesOnScroll && !fillsHost && !boardLaneActive. This head is a rebase-only retrigger, so the change ships in the follow-up PR rather than resetting these verdicts.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author

chore (non-blocking): rebased on main as 7a7bd5b, no behaviour change. The only conflict was the hook list in website/src/test/ChatSidebar.ownerComposition.test.ts, where main added useCrewEnable; the resolution keeps main's lines and re-adds this PR's useSidebarScrollPeek. The PR diff is still 31 files, +5858/-5, and tsc plus the two affected test files (8 tests) pass locally. This push also reruns Frontend Tests (4), Frontend Coverage Merge and Coverage Gate, which failed on e4e9b84 in ChatInput.lexicalKeys.test.tsx the same way main's own CI failed on aee520e.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Overlay over the hovered row's actions

outcome: deferred
The code already skips the hovered row; scrolling.png predates that fix. The follow-up PR retakes scrolling.png and attachment-03 against the current build and also skips the row useHoverHold keeps highlighted. This head is a rebase-only retrigger, so the change ships there rather than resetting these verdicts.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Chip chrome crowds the screen

outcome: deferred
Agreed. The follow-up PR drops the border, shadow and semibold weight from each chip, stops each chip at the chat pane edge before the composer, and fades chips in and out with no fade under reduced motion. This head is a rebase-only retrigger, so the change ships there rather than resetting these verdicts.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Shorter setting description

outcome: deferred
Agreed. The follow-up PR shortens full_titles_on_scroll_description to one sentence that names the session list and leaves the hide timing out. This head is a rebase-only retrigger, so the string change ships there rather than resetting these verdicts.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • scrollPeek duplicates fillsHost

outcome: deferred
Agreed: at both render sites scrollPeek equals !fillsHost. The follow-up PR deletes the prop, its JSDoc and the ChatPage hunk, and gates the hook on fullTitlesOnScroll && !fillsHost. This head is a rebase-only retrigger, so the change ships there rather than resetting these verdicts.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Evidence files not declared in the description

outcome: deferred
Correct, the description does not list the five evidence files under temp-screenshots/sidebar-scroll-peek/. The follow-up PR retakes those stills and its description lists every committed evidence file. This head is a rebase-only retrigger, so a body edit now would only restart the lanes for an item the review says needs no change.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author
  • Pointer already inside the sidebar when the listener binds

outcome: deferred
Agreed, and the fix is built and tested on the follow-up branch: pointermove and wheel events set pointer presence, covered by 'counts a wheel as pointer presence when no pointerenter fired'. It ships in the follow-up PR right after this merges. Until then the case is bounded: the peek collapses 0.7s after the last scroll, the same timing as touch and keyboard.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author

chore (non-blocking): rebased on main as 298c635, no behaviour change. The only conflict was the sidebar root div in website/src/pages/ChatSidebar.tsx, where main (#18443) added role="region" and the sessions.list guide location. The resolution keeps main's div line and re-adds this PR's ScrollPeekLayer under it. The PR diff is still 31 files, +5858/-5, and tsc plus the 3 affected test files (29 tests) pass locally.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author

issue (blocking): fixed in 1b459fa. Build Wheel and both Build Desktop jobs failed on 298c635 in gen-ui-index --check: "src/kiro_crew/docs/ui-index.generated.json is stale". main added that index in #18443, and it lists every setting, so this PR's new chat.show-full-titles-while-scrolling setting and its 12 locale labels were missing from it. The amend adds the output of npm run gen:ui: one setting row, one label per locale, and the updated digests and counts (setting 238 to 239). No other file changed, and gen-ui-index --check now passes locally. The PR diff is 32 files, +5873/-7.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author

issue (blocking): fixed in 684476d. The index check still failed on 1b459fa in Build Wheel, both Build Desktop jobs, Frontend Lint & Type Check and E2E. CI checks the merge with main, and main had gained 9 commits since the base, including b16b26b in the sidebar. The index's input_digest covers those source files, so the merge result was stale even though the branch alone passed. 684476d is the same change rebased onto main 6166905 with the index regenerated there, and only input_digest moved. gen-ui-index --check and tsc pass locally. The PR diff is still 32 files, +5873/-7.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author

note (non-blocking): two reds on 684476d come from main, not from this diff.

Frontend Tests (3) fails GuideLayer.test.tsx "pressing the highlighted control on the last step finishes the guide, as Done does". The same test fails in main's own CI run on 6166905 (run 38033018042) and in at least three other open PRs. Frontend Coverage Merge is red only because shard 3 is.

E2E (stub ACP backend, offline) fails in check-i18n-render.mjs --build at the vs-base step: "[plugin kirocrew-ui-auto-stamp] src/pages/settings/ChatPanel.tsx changed after the auto stamp manifest was written". The base tree is built against a symlink to this checkout's website/node_modules (check-i18n-render.mjs:729), and vite.config.ts reads the stamp manifest from ./node_modules/.cache/kc-ui-auto/sites.json. So the base build reads the manifest the HEAD build wrote, and any PR that edits a stamped file (ChatPanel.tsx here, where this PR adds its setting) fails the base build. Both came in with #18443. The E2E suites themselves passed.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author

chore (non-blocking): rebased on main d32c084 as 81c3477, no behaviour change. The only conflict was the generated src/kiro_crew/docs/ui-index.generated.json. The resolution takes main's copy and reruns npm run gen:ui, and gen-ui-index --check and tsc pass locally. The PR diff is still 32 files, +5873/-7. The GuideLayer red in Frontend Tests (3) also fails on main. The E2E i18n-render base-build red is fixed upstream by #18793, which is still open, so E2E may stay red on this head until that merges.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author

chore (non-blocking): rebased on main 63c155a as bb07e16, no behaviour change. The base now includes #18793, which fixes the E2E i18n render base-build stamp manifest failure. Over the two rebases since 81c3477 the conflicts were the feature-map Sessions row and the ChatSidebar.ownerComposition.test.ts hook list, where main's lines were kept and this PR's scrollPeek entry and useSidebarScrollPeek re-added, and the generated ui-index, which was taken from main and regenerated with npm run gen:ui. The PR diff is still 32 files, +5873/-7. gen-ui-index --check, tsc and the 2 affected test files (8 tests) pass locally.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author

chore (non-blocking): rebased on main 65df088 as 6e14cf9, no conflicts and no behaviour change. Backend Tests shard 2 failed on bb07e16 on Linux and Windows in test_wait_discipline_ratchet only: test_surface_deny_notice_sites.py and test_webhook_permission_requests.py gained wait_for(10s) sites in #19259, past the baseline. This PR touches no file under test/, and main fixed the baseline in #19495, which this base includes. E2E (stub ACP backend, offline) on bb07e16 ran 314 Playwright tests with 0 unexpected, and failed only because the suite took 1016s against the 1020s bound in test_playwright_e2e.py:447; this push re-runs it. The PR diff is still 32 files, +5873/-7, and gen-ui-index --check and tsc pass locally.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author

chore (non-blocking): rebased on main as 296b986, no behaviour change. main de2d143 conflicted in three places, each next to a line this PR adds: the feature-map Sessions row, the sidebar module table in docs/system-specs/modules/history.md (main added models.ts) and the hook list in ChatSidebar.ownerComposition.test.ts (main added useSidebarModels). The resolution keeps main's lines and re-adds this PR's entries. The PR diff is still 32 files, +5873/-7. gen-ui-index --check, tsc and the 2 affected test files (8 tests) pass locally. The push also re-runs E2E (stub ACP backend, offline), which stopped at the 1020s Playwright bound on a1c8e49 with 326 passed and 0 flaky. A local run of 71 tests on both sides gave a median time ratio of 1.006 against main, and fork PRs run that job on ubuntu-latest in 975 to 1098s against 655 to 829s for same-repo PRs.

@Pearcekieser

Copy link
Copy Markdown
Contributor Author

chore (non-blocking): rebased on main as f49fd49, no behaviour change. After readiness passed on 296b986, main 1f6f5af conflicted only in the generated src/kiro_crew/docs/ui-index.generated.json. The resolution takes main's copy and regenerates it with npm run gen:ui. The PR diff is still 32 files, +5873/-7, and gen-ui-index --check and tsc pass locally. The commit still merges cleanly onto main 9b1fc3c.

@iamwhatever iamwhatever left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This adds an opt-in setting, off by default and desktop only, that draws clipped session titles in full in a click-through layer while the sessions list is scrolled. Row width, the stored sidebar width and the fixed row height are not touched. Our earlier note asked for the body to describe the real lifecycle, and it now does (the 4s ceiling, a press or non-scroll key anywhere ends it, window blur, the hovered, sticky-header and occluded rows are skipped, and the z-[65] tradeoff is named). I read scrollPeek.ts, ScrollPeekLayer.tsx and the ChatSidebar, ChatPage and settings hunks, and tried to make it get in the way. The layer is pointer-events: none and aria-hidden. Every listener is passive or only calls stop(), never preventDefault, so no wheel, key or click is swallowed. Escape or any non-scroll key, any press, and blur all end it. The portal sits outside the lane, so its own render cannot feed the MutationObserver into a measure loop (samePeek also guards that). Titles render as text, not HTML. Every effect removes its listeners on cleanup. What is left is cosmetic and was answered as deferred: the chip offset by the live dot, the duplicate scrollPeek/fillsHost prop, and a pointer already inside the sidebar when the listener binds, which only shortens the peek to 0.7s. The committed evidence is about 1.2 MB, as the fork rules allow. Checks are green and the base is main. — crew-pr-reviewer (deep re-check)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

fork Pull request from a fork (external contributor)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants