Skip to content

feat: opt-in chat table breakout and resizable columns - #15547

Merged
buluoray merged 1 commit into
mainfrom
fix/crewmate-table-in-bubble
Oct 3, 2026
Merged

buluoray merged 1 commit into
mainfrom
fix/crewmate-table-in-bubble

Conversation

@buluoray

@buluoray buluoray commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: Chat tables stay at the reading column's width unless the reader asks for more, and the reader can resize any chat table's columns.

Today every top-level assistant table automatically spreads into the pane's gutters (#13066). A short table then sits wider than the prose around it, and the reader has no say in it. A wide table can be scrolled, but a cramped column cannot be widened.

Why it matters

The reader decides how much room a table takes. Prose and tables line up by default, a wide table is one press away from the pane's full width, and a cramped column can be dragged wider wherever a table renders, including inside a crewmate bubble, which keeps its tables in the card (#15509).

Not a goal

What changed (motivation → approach → change)

  • Opt-in breakout: the index.css breakout width rule now applies only to .markdown-table[data-expanded]. MarkdownTable owns an expanded state (off by default) and a new Widen toggle in its action row, beside the copy buttons. Its visible word flips Widen/Narrow (width words, not "Expand", which this app uses for disclosure), the accessible name and title flip with it ("Widen table"/"Narrow table") so a voice-control user can say what they see, aria-expanded carries the state, and its strings use i18n in all 12 locales. It shares the row's touch treatment (HOVER_NONE_ACTIONS_ROW_CLS): always visible with a 40px target on touch, and on desktop it shows on hover or keyboard focus. The toggle is display: none except on a top-level table in an unbordered transcript bubble (the breakout selector) on a viewport at least 880px wide; below that, the pane minus the rail band is no wider than the reading column. The gate is the viewport, not the pane, so in a narrow split pane on a wide screen the toggle shows but cannot widen the table past the pane. The bubble's overflow: visible / flow-root rules stay unconditional, so row heights and the raw-view height contract are the same in both states.
  • Column resize: new useMarkdownTableColumns, used by MarkdownTable. useTableColumnWidths does not fit, because it assumes declared px defaults and a markdown table has none. So the table stays on auto layout until the first drag. That drag snapshots the laid-out widths (rounded up, so no header label that fitted gets clipped) and switches to table-layout: fixed with a <colgroup>. Widening a column widens the table, and the wrapper's existing scroll absorbs it. The grips reuse ColumnResizer (ARIA splitter, Enter or double-click resets) as an overlay inside the scroll wrapper at each header cell's right edge. With no real layout (SSR/jsdom) no grips render, and when every column is reset the table returns to auto layout.
  • Keyboard: a table's grips are one roving tab stop. ArrowUp/ArrowDown move between them and ArrowLeft/ArrowRight resize, so a transcript with many tables doesn't add a Tab press per column. ColumnResizer gains optional tabIndex/onFocus/onNavigate props; existing callers are unchanged.
  • Touch: a touch device (useIsTouchDevice) gets no grips. A 6px strip can't be hit with a finger, and its touch-action: none would swallow a horizontal swipe. On touch, the Widen toggle and the wrapper's horizontal scroll are the way to read a wide table.
  • Measurement: only the thead and its header cells are observed, not the table, which grows every frame while a reply streams. Observer bursts collapse into one measurement per frame. The span check visits only cells that declare a span. A width change is detected even when no right edge moved.
  • A raw-HTML table with a colspan/rowspan cell or a multi-row header gets no grips and keeps auto layout. One grip per first-row th would otherwise map widths onto the wrong logical columns.
  • ColumnResizer's faint "all boundaries" hint now also lights when a .markdown-table's header is hovered, because those grips sit outside the thead.
  • scripts/capture-chat-table-breakout.mjs now asserts the default prose width and where the toggle is visible, then expands before running the existing breakout and turn-rail geometry checks.
  • The feature map row is updated.

Backwards compatibility

This is a deliberate default change: a top-level transcript table no longer breaks out until the reader presses Widen. Expanded geometry is identical to before. The renderer goldens gain the toggle button in each table's action row, and nothing else in their output changed.

Tests

  • chatTableBreakout.test.tsx: no ungated breakout width rule exists. The default table has no data-expanded. One press sets it along with aria-expanded="true", the Narrow word, and the "Narrow table" name and title; a second press restores "Widen table". Reverting the name/title flip turns it red. The toggle is hidden by default and only revealed by the top-level selector at ≥880px. Each test was mutation-checked: dropping the [data-expanded] gate, the click toggle, aria-expanded, or the hidden-by-default rule turns it red.

  • MarkdownTable.columnResize.test.tsx (stubbed layout) covers:

    • one grip per header cell, and auto layout until a resize
    • ArrowRight widening only that column, Enter resetting to auto layout
    • the minimum-width clamp, and no grips without layout
    • no grips on spanned raw-HTML tables
    • one roving tab stop per table, with ArrowUp/ArrowDown navigation
    • no grips on a touch device, and a pointer drag
    • the round-up freeze (sub-pixel getBoundingClientRect stub)
    • a width change with unchanged right edges
    • header-only observation and one measurement per frame

    Each of these was mutation-checked too.

  • Locally: the 51 test files that touch markdown tables, ColumnResizer or i18n pass (837 tests). tsc -b, eslint and i18n:check (diff-scoped against main) are clean.

Manual verification

scripts/capture-chat-table-breakout.mjs was run against a local Vite server (Chromium, light and dark themes, SDK and main hosts, at 1500, 1100, 768, 390, 320 and back to 1500px). The default table matched the prose width at every width. The toggle showed at 1100 and 1500 only. After Widen, every existing breakout, turn-rail, raw-view and MCP-overlay assertion passed. Column-drag behaviour was verified earlier in the same harness: Signal widened by the drag distance, and the other columns stayed within a pixel.

Not run: the full frontend suite locally (left to CI) and a live gateway pod.

Screenshots / video

Default: the table keeps the reading column's width (hovered, so the Widen toggle shows)

default

Expanded: after pressing Widen, the table uses the pane's spare width and the toggle reads Narrow

expanded

Column grips on header hover: each column edge shows a grip while the header row is hovered

grips on hover

After a drag: the first column widened by dragging its grip; the table switches to fixed layout and scrolls horizontally inside its own wrapper instead of widening the message

resized after drag

Keyboard: a grip focused from the keyboard shows the accent focus state; ArrowRight widened the column further

keyboard focus on a grip

Recording: widen, narrow, then a pointer drag on a grip (captured before the label rename, so the button still reads Expand/Collapse; behaviour is unchanged)

expand-collapse-resize.webm

Related Issues

Builds on #13066 and #15509.

Pattern harvest

Not generalizable: a feature change with no recurring defect shape.

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

Contribution License Agreement

@github-actions

github-actions Bot commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Fable 5) — ✅ human override accepted

@buluoray overrode this lane for f8e1b6a6ea33ddefa4c2fcdac178a406756c30dd via /ai-review override (this commit only).

@github-actions

github-actions Bot commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Fable 5) — 🟡 CONCERNS

Design-level review of f8e1b6a6ea33ddefa4c2fcdac178a406756c30dd — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: CONCERNS

A reader's Widen press and column drags silently revert on ordinary scrolling: VirtualTranscript unmounts off-window rows, and both choices live in the row's useState.

Watch

  • expanded (MarkdownTable.tsx) and the width snapshot (useMarkdownTableColumns) are component state inside a virtualized row; VirtualTranscript.tsx renders "only the viewport window (plus overscan)", so scrolling a few screens away and back remounts the table at defaults. The description's non-goal draws the loss boundary at reload ("both last as long as the rendered message"), but in the main transcript the real boundary is a scroll — the reader's decision, the PR's own stated point, evaporates mid-session, most often on exactly the long replies that contain tables. Hoisting the state into a per-message map keyed by message id (still ephemeral) fixes it without touching the persistence non-goal.
    Clears when: the state survives a scroll-away-and-back in the virtualized transcript (test or capture-script assertion), or a human accepts scroll-lifetime state explicitly in the PR.

Suggestions

  • The 880px media query re-encodes "reading column + 80px band" as a second constant; a comment naming the source values it was derived from (content-width setting, MIN_GUTTER_PX) would keep it correctable when either moves.

[DESIGN-REVIEWED] f8e1b6a

Description read

[DESCRIPTION-READ] f0f8a763368393f3c121f9fa6ebe9471195b17c16100fa5678a95100bc68d491

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

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of f8e1b6a6ea33ddefa4c2fcdac178a406756c30dd and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] f8e1b6a

False positive or not applicable? A repository writer can comment:
/ai-review override gpt f8e1b6a6ea33ddefa4c2fcdac178a406756c30dd: <one-sentence reason>

@github-actions

github-actions Bot commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Fable 5) — 🟡 CONCERNS

UX-level review of f8e1b6a6ea33ddefa4c2fcdac178a406756c30dd — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

UX-Verdict: CONCERNS

Over-dragged columns clip cell text with no visible scroll cue — the blind reader called the table "broken", seeing no way to read the hidden text.

Watch

  • Post-drag clipping (shot-04): "Purpose" is cut mid-word and the reader said "no visible way to scroll or see the rest… other than dragging the column back" — the hidden scrollbar plus edge fade doesn't read as scrollable once fixed layout clips cells. Every over-wide drag, friction, per message. Smallest fix: show the wrapper's scrollbar (or a stronger overflow cue) while columns.resized.
  • Silent Widen in a narrow split pane: the toggle is viewport-gated (min-width: 880px in index.css) where, per the description, "expanding is a no-op" — the word flips to Narrow with nothing visibly changing, breaking the label's promise. Rare, momentary. Decide: accept, or gate on pane width.
  • Stale evidence: shots 03–05 and the recording still read "Expand" (pre-rename), so the blind reader concluded Widen/Expand might be two actions — "the two names made me doubt myself" — and found shot-04 and shot-05 identical despite the claimed focus state. Recapture at HEAD with a visibly distinct focused grip.
  • Both primary controls were read correctly but only as "a guess" (Widen: "I think the table would stretch out sideways"; grips: "dragging would resize"). Low risk — both guesses were right and reversible — but a human should confirm the hover-revealed affordances are discoverable enough.

[UX-REVIEWED] f8e1b6a

Description read

[DESCRIPTION-READ] 66a7b4fe7140e5fed96993fbab075388c864825afe5c439edb01197569e41ad7

That sha256 names everything this verdict read: the PR description, plus the 8 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.

@buluoray
buluoray force-pushed the fix/crewmate-table-in-bubble branch from 95b647e to b11e18b Compare September 30, 2026 20:18
@buluoray

Copy link
Copy Markdown
Contributor Author

Dispositions for the review of 95b647ea7a, fixed in b11e18bf19:

GPT 5.6 BLOCKING (ArrowRight can shrink naturally wide columns): fixed. Confirmed: a column auto layout set wider than MAX_COL (720px) was clamped to 720 on its first ArrowRight. Each column's drag bounds are now min(MIN_COL, laidOut) to max(MAX_COL, laidOut), and the grip's aria-valuemin/aria-valuemax report the same bounds. New test never narrows a column auto layout made wider than the drag cap on a widen gesture lays out an 800px column and asserts ArrowRight keeps it at 800 and ArrowLeft moves it to 784. Reverting the bound to a flat MAX_COL turns exactly that test red.

Design Review CONCERNS (opt-out :not(.crewmate-bubble) fails open): acknowledged, not changed here. The concern is valid in principle: a future filled bubbleClassName host would inherit the breakout. This PR keeps the opt-out because it leaves the existing #13066 behaviour byte-identical for every host that exists today. Flipping to an opt-in marker on the plain transcript bubble is a polarity decision for the #13066 owner and is left to a follow-up.

@buluoray

Copy link
Copy Markdown
Contributor Author

UX Review (CONCERNS, 95b647ea7a) dispositions. Everything below is advisory, and no code changed for it:

  • Grip discoverability: accepted as is. The grip, its hover highlight and its title hint are the same ColumnResizer the dashboard data tables already ship. Relabelling that shared widget is outside this PR.
  • Tall rows after resizing a cramped bubble table: accepted-and-deferred. The rows are tall because the columns that haven't been scrolled into view are narrow and their text wraps; that is ordinary auto-layout behaviour in a narrow bubble and is not caused by the resize. Clipping those cells would hide their content instead, which is a separate design call.
  • No visible scroll affordance on the contained table: accepted-and-deferred. The in-bubble horizontal scroll is the behaviour that existed before fix: let chat tables use the available pane width #13066. Adding a scroll-edge fade to markdown tables is a new surface and belongs in a follow-up, not in this fix.
  • Widths lost when the virtualizer unmounts a row: rebutted. The PR states widths last only "as long as the message stays rendered". An unmounted row is no longer rendered, so the promise holds. Keeping widths after an unmount would mean storing them outside the component, the same kind of persistence that "Not a goal" rules out for reloads.

@github-actions

github-actions Bot commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5 Review — ✅ no blocking findings

Reviewed f8e1b6a6ea33ddefa4c2fcdac178a406756c30dd — this comment is updated in place on each push.

Review details

The Widen toggle's state is row-local in a virtualized transcript, so the reader's choice is dropped on scroll — advisory, not blocking.

FINDING — website/src/components/markdown/MarkdownTable.tsx:73 — const [expanded, setExpanded] = useState(false) keeps the Widen choice in row-local state while useVirtualChat emits only the window (out.push({ … mounted: true … }) for start..end only, off-window rows omitted), so widening a table in a long thread and scrolling it past the overscan band unmounts the row and the remount loses data-expanded — the table silently snaps back to the reading column and the button reads "Widen" again → Fix: back it with useRowDisclosure from pages/chat/rowDisclosure.tsx (written for exactly this: "Any expand/collapse state held in useState inside a row is therefore destroyed"), which needs a stable key plumbed from AssistantMessage/MarkdownRenderer — untouched files, so the remedy cannot land inside this diff.

[OPUS-REVIEWED] f8e1b6a

Verdict parsed from the review's SHA-scoped output markers for commit f8e1b6a6ea33ddefa4c2fcdac178a406756c30dd.

False positive or not applicable? A repository writer can comment:
/ai-review override fable f8e1b6a6ea33ddefa4c2fcdac178a406756c30dd: <one-sentence reason>

@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
@buluoray
buluoray force-pushed the fix/crewmate-table-in-bubble branch from b11e18b to 1a89c53 Compare September 30, 2026 21:54
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision merge conflict Branch has merge conflicts with its base — author must resolve before merge and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Sep 30, 2026
@buluoray
buluoray force-pushed the fix/crewmate-table-in-bubble branch from 1a89c53 to a1fc031 Compare October 1, 2026 05:21
@buluoray buluoray changed the title feat: resizable chat table columns, crewmate tables stay in bubble feat: resizable chat table columns Oct 1, 2026
@buluoray

buluoray commented Oct 1, 2026

Copy link
Copy Markdown
Contributor Author

Rebased onto main. #15509 already keeps a crewmate (bordered) bubble's table inside its card via :not([data-bordered]), which covers what this PR's crewmate-bubble marker did. So I took main's selectors in index.css / chatTableBreakout.test.tsx and dropped the now-unused marker from crewmateBubbles.ts. This PR is now column resizing only, and the title and body are updated to match.

@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: passed Eligible automated validation passed for the current revision merge conflict Branch has merge conflicts with its base — author must resolve before merge labels Oct 1, 2026
@buluoray
buluoray force-pushed the fix/crewmate-table-in-bubble branch from a1fc031 to e072c41 Compare October 1, 2026 05:33
@github-actions

github-actions Bot commented Oct 2, 2026

Copy link
Copy Markdown
Contributor

Human judgment recorded

@buluoray marked the gpt AI finding as false positive, not applicable, or explicitly accepted for 992b7a17250c1d0dd7a495e548147265f3e96f1e.

maintainer decision — column resizing is a pointer-precision affordance and does not need a touch-sized handle; touch readers keep the content-width default and the Expand control, and the per-table row keeps three controls (Expand, Copy Markdown, Copy CSV)

This decision applies only to this commit. A new push requires a new judgment.

@github-actions

github-actions Bot commented Oct 2, 2026

Copy link
Copy Markdown
Contributor

Human judgment recorded

@buluoray marked the first-principles AI finding as false positive, not applicable, or explicitly accepted for 992b7a17250c1d0dd7a495e548147265f3e96f1e.

maintainer decision — transcript tables default to the reading-column width and break out only when the reader presses Expand; this intentionally supersedes the #13066 breakout-by-default

This decision applies only to this commit. A new push requires a new judgment.

@buluoray
buluoray force-pushed the fix/crewmate-table-in-bubble branch from 992b7a1 to de3a6d1 Compare October 2, 2026 17:33
@buluoray

buluoray commented Oct 2, 2026

Copy link
Copy Markdown
Contributor Author

Rebased onto main to clear a conflict with #15850 (table overflow fade). Merge: the table node feeds both the overflow fade (attachTable) and the column sizer (tableRef) through one callback ref; the CSS keeps both blocks; the observer test now scopes to the column sizer's own observer (it still reddens if the sizer watches the table). No behaviour change in this PR's scope. New head: de3a6d1 — overrides from 992b7a1 need re-posting.

@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running merge conflict Branch has merge conflicts with its base — author must resolve before merge labels Oct 2, 2026
@buluoray

buluoray commented Oct 2, 2026

Copy link
Copy Markdown
Contributor Author

Dispositions for de3a6d1 (rebase-only head; same findings as 992b7a1):

  • GPT 5.6 — no findings.
  • Opus 5 BLOCK (max-two-buttons-per-row) — the three-button row is kept by maintainer decision; awaiting the override on this head.
  • First Principles BLOCK (opt-in breakout default) — product decision; awaiting the override on this head.
  • Design CONCERNS (Widen/width state reset when the virtualized transcript unmounts a row) — known and listed under Non-goals in the description; persisting it is out of scope here.
  • UX CONCERNS (Widen is a visible no-op in a narrow split pane on a wide viewport) — known; the gate is the viewport, documented in the feature map and the CSS comment. A pane-width gate is a follow-up.

Inherited reds: Frontend Tests (1)/(2), Frontend Coverage Merge and Coverage Gate fail on main too (#14807 cross-merge); the fix is #16210. Not patched here.

@buluoray

buluoray commented Oct 2, 2026

Copy link
Copy Markdown
Contributor Author

/ai-review override fable de3a6d1: maintainer decision — the per-table action row keeps three controls (Expand, Copy Markdown, Copy CSV) as one compact table toolbar; opt-in expand needs its own visible control and merging the copy buttons into a menu was considered and rejected

@buluoray

buluoray commented Oct 2, 2026

Copy link
Copy Markdown
Contributor Author

/ai-review override first-principles de3a6d1: maintainer decision — transcript tables default to the reading-column width and break out only when the reader presses Expand; this intentionally supersedes the #13066 breakout-by-default

@github-actions

github-actions Bot commented Oct 2, 2026

Copy link
Copy Markdown
Contributor

Human judgment recorded

@buluoray marked the fable AI finding as false positive, not applicable, or explicitly accepted for de3a6d174e3b8c573a15a8955cf5bb4e78a723c6.

maintainer decision — the per-table action row keeps three controls (Expand, Copy Markdown, Copy CSV) as one compact table toolbar; opt-in expand needs its own visible control and merging the copy buttons into a menu was considered and rejected

This decision applies only to this commit. A new push requires a new judgment.

@github-actions

github-actions Bot commented Oct 2, 2026

Copy link
Copy Markdown
Contributor

Human judgment recorded

@buluoray marked the first-principles AI finding as false positive, not applicable, or explicitly accepted for de3a6d174e3b8c573a15a8955cf5bb4e78a723c6.

maintainer decision — transcript tables default to the reading-column width and break out only when the reader presses Expand; this intentionally supersedes the #13066 breakout-by-default

This decision applies only to this commit. A new push requires a new judgment.

@buluoray
buluoray force-pushed the fix/crewmate-table-in-bubble branch from de3a6d1 to 203b303 Compare October 2, 2026 23:40
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Oct 3, 2026
A top-level assistant table now keeps the reading column's width by
default. A Widen toggle beside the copy actions (keyboard-operable,
aria-expanded, i18n, a 40px target on touch) widens it into the pane's
spare width, and a second press (Narrow) restores it. The accessible
name and title flip with the visible word. The toggle is gated on a
top-level table in an unbordered bubble and a viewport at least 880px
wide; the gate is the viewport, not the pane, so in a narrow split pane
on a wide screen it shows but cannot widen past the pane.

Columns are resizable from a grip on each header cell's right edge. The
first drag freezes the auto-laid-out widths (rounded up) and switches
the table to fixed layout, so widening a column widens the table and
never squeezes a neighbour. Widths live for the rendered message only.
A table's grips are one roving tab stop: ArrowUp/ArrowDown move between
them, ArrowLeft/ArrowRight resize, Enter or double-click resets. Touch
devices get no grips, so a horizontal swipe still scrolls the table.
Only the header boxes are observed, measured at most once per frame.
@buluoray

buluoray commented Oct 3, 2026

Copy link
Copy Markdown
Contributor Author

GPT 5.6 on 203b303: both findings were real, both fixed in the amended commit.

  • F1 (stale baseline after a full reset): when the last resized column resets, the hook now clears the baseline along with the widths, so the next resize re-measures the table instead of re-adopting the first resize's widths. Test: measures fresh widths on a resize after a full reset, not the first resize's (reddens when the baseline clear is removed).
  • F2 (motionless click froze the table into fixed layout): resize now returns early when the table is on auto layout and the clamped width equals the laid-out one, so a click without movement — or a widen at the cap — changes nothing. Test: leaves the table on auto layout after a motionless click on a grip (reddens when the guard is removed). The cap test now asserts the at-cap widen keeps auto layout.

Related test files pass (75), tsc -b and eslint clean.

@buluoray

buluoray commented Oct 3, 2026

Copy link
Copy Markdown
Contributor Author

Dispositions for f8e1b6a:

  • GPT 5.6 — no blocking findings (F1/F2 from 203b303 are fixed in this head).
  • First Principles BLOCK (opt-in breakout default) — product decision; awaiting the override on this head.
  • Design CONCERNS (Widen / column widths reset when the virtualized transcript unmounts a row) — known, listed under Non-goals in the description; persisting per-message table state is out of scope.
  • UX CONCERNS (a column dragged narrower clips its cell text with no cue) — by design: a narrowed fixed-layout column clips rather than painting over its neighbour, the header ellipsizes, and Enter / double-click on the grip restores the auto width. A visible clipped-cell cue is a follow-up, not a fix for this PR.

@buluoray

buluoray commented Oct 3, 2026

Copy link
Copy Markdown
Contributor Author

/ai-review override first-principles f8e1b6a: maintainer decision — transcript tables default to the reading-column width and break out only when the reader presses Expand; this intentionally supersedes the #13066 breakout-by-default

@github-actions

github-actions Bot commented Oct 3, 2026

Copy link
Copy Markdown
Contributor

Human judgment recorded

@buluoray marked the first-principles AI finding as false positive, not applicable, or explicitly accepted for f8e1b6a6ea33ddefa4c2fcdac178a406756c30dd.

maintainer decision — transcript tables default to the reading-column width and break out only when the reader presses Expand; this intentionally supersedes the #13066 breakout-by-default

This decision applies only to this commit. A new push requires a new judgment.

@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.

Re-reviewed at f8e1b6a; all five points fixed. Each table is one roving tab stop (useMarkdownTableColumns.tsx:96-99, 177-193), touch devices get no grips while a mouse on a touchscreen laptop still does, the change check compares width and label, and only the header and its cells are observed, measured at most once per frame. The feature-map row is updated, and new tests cover a pointer drag, a motionless click, the round-up width and the observed elements. setTable feeds both tableRef and #15850's attachTable (MarkdownTable.tsx:63-69), so the overflow fade still tracks the table. Clearing my request.

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

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants