…390)

@tanstack/virtual-core 3.19.6 (TanStack/virtual#1326) measures the scroll
viewport with clientWidth/clientHeight instead of offsetWidth/offsetHeight. The
inbox attachments test gave happy-dom elements a 640 x 640 viewport through the
offset box only, so under 3.19.6 the transcript measured 0 x 0, rendered no rows,
and "an attachment in the conversation previews in Related" could not find the
attachment. The stack lane resolves dependencies fresh (pnpm 11's one-day
minimum release age made 3.19.6 eligible on 2026-10-17 ~11:20 UTC), so the lane
went red on main and on every open PR from then on.

The fixture now sizes both boxes, so it models a viewport whichever one the
virtualizer measures. Browsers report real client sizes; product code is
unaffected.

Claude-Session: https://claude.ai/code/session_01VvRPgo3mVuL9ULgDSeZGBV

@tanstack/virtual-core 3.19.6 (TanStack/virtual#1326) measures the scroll
viewport with clientWidth/clientHeight instead of offsetWidth/offsetHeight. The
inbox attachments test gave happy-dom elements a 640 x 640 viewport through the
offset box only, so under 3.19.6 the transcript measured 0 x 0, rendered no rows,
and "an attachment in the conversation previews in Related" could not find the
attachment. The stack lane resolves dependencies fresh (pnpm 11's one-day
minimum release age made 3.19.6 eligible on 2026-10-17 ~11:20 UTC), so the lane
went red on main and on every open PR from then on.

The fixture now sizes both boxes, so it models a viewport whichever one the
virtualizer measures. Browsers report real client sizes; product code is
unaffected.

Claude-Session: https://claude.ai/code/session_01VvRPgo3mVuL9ULgDSeZGBV

Bumps @tanstack/react-virtual ^3.14.19 -> ^3.14.21 (@tanstack/virtual-core
3.19.4 -> 3.19.6) and repairs the jsdom unit tests the bump broke.

virtual-core 3.19.6 (TanStack/virtual#1326) measures the scroll viewport
with clientWidth/clientHeight instead of offsetWidth/offsetHeight so that
borders and scrollbars no longer inflate it. jsdom has no layout engine, so
every dimension reads 0 and the virtualizer renders no rows unless the
tests stub the dimensions it reads. `setupVirtualListEnv` and the
CommandList test only stubbed the offset pair, so every virtualized list
under test rendered empty. The stub now covers the client pair too (rows
are still measured with offset dimensions), and the CommandList test uses
the shared helper instead of its own copy of the stub.

pnpm pruned the now-stale `@tanstack/*@<old version>` entries from
`minimumReleaseAgeExclude` (`minimumReleaseAgeExcludePrune`).

Supersedes #15438.

Bumps @tanstack/react-virtual ^3.14.19 -> ^3.14.21 (@tanstack/virtual-core
3.19.4 -> 3.19.6) and repairs the jsdom unit tests the bump broke.

virtual-core 3.19.6 (TanStack/virtual#1326) measures the scroll viewport
with clientWidth/clientHeight instead of offsetWidth/offsetHeight so that
borders and scrollbars no longer inflate it. jsdom has no layout engine, so
every dimension reads 0 and the virtualizer renders no rows unless the
tests stub the dimensions it reads. `setupVirtualListEnv` and the
CommandList test only stubbed the offset pair, so every virtualized list
under test rendered empty. The stub now covers the client pair too (rows
are still measured with offset dimensions), and the CommandList test uses
the shared helper instead of its own copy of the stub.

pnpm pruned the now-stale `@tanstack/*@<old version>` entries from
`minimumReleaseAgeExclude` (`minimumReleaseAgeExcludePrune`).

Supersedes #15438.

@tanstack/virtual-core 3.19.6, which @tanstack/react-virtual 3.14.21 brings,
measures the scroll element with clientWidth and clientHeight instead of
offsetWidth and offsetHeight (TanStack/virtual#1326), so borders and
scrollbars no longer count as visible room. The tests that scroll a long List
or DataTable on the page gave the document's scrolling element an 800px
offsetHeight, which the virtualizer no longer reads; happy-dom reports a
clientHeight of 0, so only the overscan rows were drawn and key-200 never
appeared. They now set clientHeight and clientWidth, which is also what a
browser reports as the viewport for the document, whose offsetHeight is the
whole page. The assertions on the rows a person sees are unchanged.

@tanstack/react-virtual 3.14.21 pulls in @tanstack/virtual-core 3.19.6,
whose `getRect()` now measures the scroll viewport with
`clientWidth`/`clientHeight` instead of `offsetWidth`/`offsetHeight`
(upstream TanStack/virtual#1326, "measure the visible scroll viewport" —
a deliberate correctness fix so borders and scrollbars no longer inflate
the viewport and obscure end-aligned items).

jsdom performs no layout, so those properties are 0 on every element. The
virtualizer therefore computed a 0-height viewport, `getVirtualItems()`
returned nothing, and 39 "renders a row" assertions failed across
NotificationBell.test.tsx (35) and Sidebar.test.tsx (4).

Stub `clientHeight`/`clientWidth` alongside the existing `offsetHeight`
stub in both files' `stubVirtualizerLayout()`, mirroring upstream's own
remediation in TanStack/virtual#1326. The `offsetHeight` stub stays: row
measurement still reads `offsetHeight` when the ResizeObserver supplies
no `borderBoxSize`. The stubs stay local to these two suites rather than
going into the shared testSetup.ts, since unrelated suites deliberately
rely on jsdom reporting `clientHeight === 0`.

No product code is touched.

…ates (#1872)

Squashes the grouped frontend `npm-deps` bump plus the test-harness fix it
requires.

`@tanstack/react-virtual` 3.14.19 → 3.14.21 pulls `@tanstack/virtual-core`
3.19.4 → 3.19.6, where upstream TanStack/virtual#1326 ("fix(virtual-core):
measure the visible scroll viewport") changed the scroll-viewport measurement
from `offsetWidth`/`offsetHeight` to `clientWidth`/`clientHeight` (and the
ResizeObserver from `border-box` to `content-box`). Mullion's jsdom harness
only stubbed `offsetHeight`, so the virtualizer measured a 0-height viewport,
`getVirtualItems()` returned an empty list, and 39 assertions failed across
`NotificationBell.test.tsx` (35) and `Sidebar.test.tsx` (4).

The fix adds `clientWidth`/`clientHeight` stubs next to the existing
`offsetHeight` stub in those two `stubVirtualizerLayout()` helpers. The
stubs stay local to those two files: `frontend/src/testSetup.ts` is shared,
and `terminalTouchScroll.test.ts` deliberately relies on jsdom reporting
`clientHeight === 0`. `offsetHeight` is kept because row measurement still
reads it when the ResizeObserver reports no `borderBoxSize`. No product code
is touched.

Follow-up: the dockview 8.6.1 `closeEmptyGroups` default is unchanged from
8.5.1, but the workspace-restore path is worth one manual check — #1903.

Unblocks dependabot PR #2595 (npm-minor-patch group). That PR fails
`Test (web)` / `Coverage (web)` with 20 failures from two of its bumps.
This change is backward compatible: it passes on `main`'s current
dependency versions and on #2595's lockfile.

## `@pierre/diffs` 1.2.12 → 1.6.1: product fix

When a hunk has fewer lines than its header declares (for example, `@@
-1,3 +1,3 @@` with only 2 lines per side), 1.6.1's lenient parser logs
`hunk line count mismatch` and rewrites
`hunk.additionCount`/`deletionCount` to the parsed counts. That makes
`previewMarkdownDocument`'s completeness check always pass, so a
**truncated patch would render as a complete Markdown document**.

`previewMarkdownDocument` now calls `parsePatchFiles(patch, undefined,
true)`, which has the same signature in 1.2.12 and 1.6.1. Strict mode
throws on that mismatch, and any parse error maps to `unavailable`. That
is the result the existing `refuses patches that do not hold one
complete document` test expects.

## `@tanstack/react-virtual` 3.14.17 → 3.14.21 (`virtual-core` 3.19.6):
test fix

virtual-core 3.19.6
([TanStack/virtual#1326](TanStack/virtual#1326))
measures the scroll viewport with `clientWidth`/`clientHeight` instead
of `offsetWidth`/`offsetHeight`. The timeline tests only mocked
`offset*`. jsdom's `clientHeight` is 0, so the virtualizer saw a 0px
viewport and rendered no rows. The three `session-timeline*` test files
now also mock `clientHeight`/`clientWidth`. Row items are still measured
with `offsetHeight`, so no other mock changes are needed.

## Validation

`vitest run --maxWorkers=1` on `session-timeline.test.tsx`,
`session-timeline-scroll.test.tsx`, `session-timeline-autofix.test.tsx`,
`markdown-preview.test.ts` and `session-changes-panel.test.tsx`:
- On `main` dependencies: 81/81 pass.
- On #2595's lockfile: 81/81 pass. Without this change, 20 of those
tests fail.
- Web `tsc --noEmit`, `eslint` and `prettier` are clean.

After this merges, `@dependabot rebase` on #2595 should turn it green.

---
*Created with
[Open-Inspect](https://open-inspect-prod.vercel.app/session/dd389a219f00b01bbb36eedceb2635c1)*

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

<!-- coderabbit-ai-generated:v1 -->

## Summary by CodeRabbit

* **Bug Fixes**
* Markdown previews are marked unavailable when patch data fails strict
parsing; truncated patch sections are no longer accepted as valid.
* **Tests**
  * Updated timeline tests to account for browser element dimensions.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

Co-authored-by: Cole Murray <2492022+ColeMurray@users.noreply.github.com>