Skip to content

Make chat timeline height estimates width-aware with attachment-aware tests - #167

Merged
juliusmarminge merged 11 commits into
mainfrom
t3code/fix-chat-attachment-height
Mar 5, 2026
Merged

juliusmarminge merged 11 commits into
mainfrom
t3code/fix-chat-attachment-height

Conversation

@juliusmarminge

@juliusmarminge juliusmarminge commented Mar 4, 2026

Copy link
Copy Markdown
Member

Summary

  • extract timeline row height estimation into timelineHeight.ts and reuse it from ChatView
  • make message height estimates responsive to measured timeline width via ResizeObserver
  • trigger virtualizer re-measure when timeline width changes to improve scroll/row sizing accuracy
  • add focused unit coverage for assistant/user sizing, attachment row behavior, newline handling, uncapped long messages, and width-dependent wrapping

Testing

  • Added/updated unit tests: apps/web/src/components/timelineHeight.test.ts
  • Concrete checks covered:
  • assistant sizing baseline behavior
  • user attachment row increments (1-2 vs 3-4 attachments)
  • long user message height growth (no cap)
  • explicit newline line counting
  • narrower vs wider width wrapping for both user and assistant messages
  • Not run: bun lint
  • Not run: bun typecheck
  • Not run: test suite execution in this PR description context

Note

Medium Risk
Changes ChatView virtualization sizing logic and introduces width-driven remeasurement, which can affect scrolling behavior and performance across viewports. Adds new Playwright/MSW browser-test infrastructure and CI steps that may introduce flakiness if timings or browser deps differ in CI.

Overview
Chat timeline virtualization now estimates row heights using measured width. The inline estimateTimelineMessageHeight logic is extracted to timelineHeight.ts, updated to account for wrapping based on timelineWidthPx (and attachment row sizing), and wired into MessagesTimeline via a ResizeObserver + rowVirtualizer.measure() on width changes.

Adds automated coverage and CI for real browser layout parity. Introduces unit tests for estimateTimelineMessageHeight, plus a Vitest+Playwright browser test (ChatView.browser.tsx) that boots the full app with MSW-mocked WS/HTTP and asserts measured DOM heights stay within tolerance across multiple viewport sizes and with attachments; CI now installs/caches Playwright Chromium and runs apps/web test:browser. Also factors router creation into getRouter, adds MSW worker config/asset, and updates .gitignore for Playwright artifacts/screenshots.

Written by Cursor Bugbot for commit 93412ac. This will update automatically on new commits. Configure here.

Note

Make apps/web/src/components/ChatView.MessagesTimeline height estimation width-aware using timelineHeight.estimateTimelineMessageHeight and add Playwright-driven browser tests to validate attachment and text wrapping behavior

Introduce a width-aware timelineHeight.estimateTimelineMessageHeight and wire it into ChatView.MessagesTimeline; add browser tests with Playwright/Vitest and MSW; update CI to install/cache Playwright and run browser tests; refactor router setup into a factory.

📍Where to Start

Start with the estimator in timelineHeight.ts, then review its integration in ChatView.tsx and validations in ChatView.browser.tsx.

Macroscope summarized 93412ac.

- extract `estimateTimelineMessageHeight` into `timelineHeight.ts`
- account for timeline width and explicit newlines when estimating wrapped lines
- re-measure virtualized rows on width changes to fix attachment/message height sizing
- add Vitest coverage for assistant/user wrapping and attachment row height rules
@coderabbitai

coderabbitai Bot commented Mar 4, 2026

Copy link
Copy Markdown

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 0d941110-f4e5-4a28-a68b-50fcf6e47356

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

  • 🔍 Trigger review
✨ Finishing Touches
🧪 Generate unit tests (beta)
  • Create PR with unit tests
  • Post copyable unit tests in a comment
  • Commit unit tests in branch t3code/fix-chat-attachment-height

Comment @coderabbitai help to get the list of available commands and usage tips.

@cursor cursor Bot 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.

Cursor Bugbot has reviewed your changes and found 1 potential issue.

Autofix Details

Bugbot Autofix prepared a fix for the issue found in the latest run.

  • ✅ Fixed: ResizeObserver never activates for initially empty threads
    • Replaced useRef with a state-backed callback ref (useState + setTimelineRootNode as the ref callback) so the useLayoutEffect re-runs when the timeline div mounts after transitioning from the empty-thread state.

Create PR

Or push these changes by commenting:

@cursor push fd13dceb67
Preview (fd13dceb67)
diff --git a/apps/web/src/components/ChatView.tsx b/apps/web/src/components/ChatView.tsx
--- a/apps/web/src/components/ChatView.tsx
+++ b/apps/web/src/components/ChatView.tsx
@@ -3187,12 +3187,11 @@
   onImageExpand,
   markdownCwd,
 }: MessagesTimelineProps) {
-  const timelineRootRef = useRef<HTMLDivElement | null>(null);
+  const [timelineRootNode, setTimelineRootNode] = useState<HTMLDivElement | null>(null);
   const [timelineWidthPx, setTimelineWidthPx] = useState<number | null>(null);
 
   useLayoutEffect(() => {
-    const timelineRoot = timelineRootRef.current;
-    if (!timelineRoot) return;
+    if (!timelineRootNode) return;
 
     const updateWidth = (nextWidth: number) => {
       setTimelineWidthPx((previousWidth) => {
@@ -3203,7 +3202,7 @@
       });
     };
 
-    updateWidth(timelineRoot.getBoundingClientRect().width);
+    updateWidth(timelineRootNode.getBoundingClientRect().width);
 
     if (typeof ResizeObserver === "undefined") return;
     const observer = new ResizeObserver((entries) => {
@@ -3211,11 +3210,11 @@
       if (!entry) return;
       updateWidth(entry.contentRect.width);
     });
-    observer.observe(timelineRoot);
+    observer.observe(timelineRootNode);
     return () => {
       observer.disconnect();
     };
-  }, []);
+  }, [timelineRootNode]);
 
   const rows = useMemo<TimelineRow[]>(() => {
     const nextRows: TimelineRow[] = [];
@@ -3645,7 +3644,7 @@
   }
 
   return (
-    <div ref={timelineRootRef} className="mx-auto w-full min-w-0 max-w-3xl overflow-x-hidden">
+    <div ref={setTimelineRootNode} className="mx-auto w-full min-w-0 max-w-3xl overflow-x-hidden">
       {virtualizedRowCount > 0 && (
         <div className="relative" style={{ height: `${rowVirtualizer.getTotalSize()}px` }}>
           {virtualRows.map((virtualRow: VirtualItem) => {

Comment thread apps/web/src/components/ChatView.tsx Outdated
Comment thread apps/web/src/components/timelineHeight.ts Outdated
- add browser E2E tests validating timeline height estimator against rendered DOM
- configure Playwright test runner and scripts in `apps/web`
- run browser tests in CI with Playwright browser caching and install step
- Replace Playwright e2e timeline-height test setup with Vitest browser config
- Add `ChatView.browser.tsx` parity tests for text wrapping and attachment height estimation
- Add row data attributes in `ChatView` to support in-browser measurement targeting

@cursor cursor Bot 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.

Cursor Bugbot has reviewed your changes and found 1 potential issue.

Autofix Details

Bugbot Autofix prepared a fix for the issue found in the latest run.

  • ✅ Fixed: Inconsistent width measurement APIs cause initial value mismatch
    • Replaced getBoundingClientRect().width (border-box) with clientWidth - paddingLeft - paddingRight (content-box) for the initial measurement, making it consistent with the ResizeObserver's contentRect.width.

Create PR

Or push these changes by commenting:

@cursor push e1d5bbad14
Preview (e1d5bbad14)
diff --git a/apps/web/src/components/ChatView.tsx b/apps/web/src/components/ChatView.tsx
--- a/apps/web/src/components/ChatView.tsx
+++ b/apps/web/src/components/ChatView.tsx
@@ -3203,7 +3203,12 @@
       });
     };
 
-    updateWidth(timelineRoot.getBoundingClientRect().width);
+    const cs = getComputedStyle(timelineRoot);
+    updateWidth(
+      timelineRoot.clientWidth -
+        parseFloat(cs.paddingLeft) -
+        parseFloat(cs.paddingRight),
+    );
 
     if (typeof ResizeObserver === "undefined") return;
     const observer = new ResizeObserver((entries) => {

Comment thread apps/web/src/components/ChatView.tsx Outdated
- Replace mocked router hooks with a memory router + RouterProvider
- Render ChatView through a test route to exercise attachment height layout in realistic routing context
- Run ChatView browser tests through the real app router/store stack
- Add MSW worker setup and mocked WS/attachment responses for full-app rendering
- Extract shared `getRouter` setup and add a timeline root data hook for reliable measurement

@cursor cursor Bot 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.

Cursor Bugbot has reviewed your changes and found 1 potential issue.

Autofix Details

Bugbot Autofix prepared a fix for the issue found in the latest run.

  • ✅ Fixed: Document title no longer set after refactor
    • Fixed the meta format from { name: "title", content: ... } to { title: APP_DISPLAY_NAME } and added the missing <HeadContent /> component to both render paths in the root route so TanStack Router's head management actually takes effect.

Create PR

Or push these changes by commenting:

@cursor push 7f77794bdd
Preview (7f77794bdd)
diff --git a/apps/web/src/routes/__root.tsx b/apps/web/src/routes/__root.tsx
--- a/apps/web/src/routes/__root.tsx
+++ b/apps/web/src/routes/__root.tsx
@@ -1,5 +1,6 @@
 import { ThreadId } from "@t3tools/contracts";
 import {
+  HeadContent,
   Outlet,
   createRootRouteWithContext,
   type ErrorComponentProps,
@@ -27,26 +28,30 @@
   component: RootRouteView,
   errorComponent: RootRouteErrorView,
   head: () => ({
-    meta: [{ name: "title", content: APP_DISPLAY_NAME }],
+    meta: [{ title: APP_DISPLAY_NAME }],
   }),
 });
 
 function RootRouteView() {
   if (!readNativeApi()) {
     return (
-      <div className="flex h-screen flex-col bg-background text-foreground">
-        <div className="flex flex-1 items-center justify-center">
-          <p className="text-sm text-muted-foreground">
-            Connecting to {APP_DISPLAY_NAME} server...
-          </p>
+      <>
+        <HeadContent />
+        <div className="flex h-screen flex-col bg-background text-foreground">
+          <div className="flex flex-1 items-center justify-center">
+            <p className="text-sm text-muted-foreground">
+              Connecting to {APP_DISPLAY_NAME} server...
+            </p>
+          </div>
         </div>
-      </div>
+      </>
     );
   }
 
   return (
     <ToastProvider>
       <AnchoredToastProvider>
+        <HeadContent />
         <EventRouter />
         <DesktopProjectBootstrap />
         <Outlet />

Comment thread apps/web/src/routes/__root.tsx
Comment thread apps/web/src/components/ChatView.tsx Outdated
- replace manual `createRoot` mounting with `render`/`unmount` from vitest-browser-react
- add `vitest-browser-react` to web devDependencies and lockfile
- Replace manual timeout polling loops with `vi.waitFor` in test helpers
- Improve reliability of locating and measuring virtualized user rows in attachment-height tests
- recompute timeline width from the root element on ResizeObserver updates
- rerun width effect when message/working state changes
- reduce minimum user chars per line for narrow layouts and add regression test
- Add viewport matrix coverage for long text and attachment row-height parity
- Reuse a mount/measure test harness to validate resize behavior in one session
- Ensure production CSS and viewport sizing are applied before measurements

@cursor cursor Bot 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.

Cursor Bugbot has reviewed your changes and found 1 potential issue.

Bugbot Autofix prepared a fix for the issue found in the latest run.

  • ✅ Fixed: Condition inversion changes system message height estimation
    • Changed message.role !== "user" back to message.role === "assistant" so system messages use user sizing rules (matching the original behavior) instead of being incorrectly routed to assistant sizing.

Create PR

Or push these changes by commenting:

@cursor push 7fbf52b7af
Preview (7fbf52b7af)
diff --git a/apps/web/src/components/timelineHeight.ts b/apps/web/src/components/timelineHeight.ts
--- a/apps/web/src/components/timelineHeight.ts
+++ b/apps/web/src/components/timelineHeight.ts
@@ -67,7 +67,7 @@
   message: TimelineMessageHeightInput,
   layout: TimelineHeightEstimateLayout = { timelineWidthPx: null },
 ): number {
-  if (message.role !== "user") {
+  if (message.role === "assistant") {
     const charsPerLine = estimateCharsPerLineForAssistant(layout.timelineWidthPx);
     const estimatedLines = estimateWrappedLineCount(message.text, charsPerLine);
     return ASSISTANT_BASE_HEIGHT_PX + estimatedLines * LINE_HEIGHT_PX;

Comment thread apps/web/src/components/timelineHeight.ts
- Route `system` messages through assistant height estimation
- Keep user attachment height logic scoped to user messages
- Add regression test for system-message sizing behavior
@juliusmarminge
juliusmarminge merged commit 7c7fcf4 into main Mar 5, 2026
4 checks passed
@juliusmarminge
juliusmarminge deleted the t3code/fix-chat-attachment-height branch March 5, 2026 18:20
aorwall added a commit to aorwall/t3code that referenced this pull request Sep 14, 2026
Merges `pingdotgg/t3code` into the fork: `0c5771d60` → `1bbca0e78`, 32
commits, landed as a merge commit.

`201` files changed in the upstream range and `204` landed. The gap is
exactly three files, all fork docs written by this merge —
`docs/fork/gaps.md`, `inventory.json` and `upstream-merge-log.md`.
Nothing in the range failed to land.

## Conflicts

Nine files, each resolved to the verdict `preflight.mjs` printed. No
conflict was left undecided, so this is not a draft.

Most of them are one upstream change hitting fork gates from several
sides: the compact-sidebar work (pingdotgg#11525, pingdotgg#9417, pingdotgg#11644) rewrote
`Sidebar.tsx`, `LegacySidebar.tsx`, `ui/sidebar.tsx`,
`SidebarThreadHeader.tsx` and `ThreadStatusIndicators.tsx`. Took
upstream's structure whole in each and re-applied the fork delta at its
anchor; four of the five were import-block unions or a single re-placed
element.

`AGENTS.md` — upstream's only change in range was the reusable dev
credential, restated into the fork's own prose rather than taken as a
block. `auth.ts` — took upstream's new `urlCredential` parameter on top
of the fork's Moatless cookie-session probe; a desktop credential is now
exchanged only when the server advertises `desktop-bootstrap`, which the
Moatless auth descriptor never does, so a stale one is never sent.
`pnpm-lock.yaml` was reset to `upstream/main` and the fork edges
re-derived by `vp i`.

### Two things that did not announce themselves

**A gated control was replaced, so the gate had to move rather than
survive.** pingdotgg#11678 deleted the `enableLegacyTokenStreaming` switch the
fork gates and put a three-way `responseStreamingMode` select in its
place. Taking upstream drops the gate along with the row it sat on and
un-hides the replacement — no conflict marker, and no test covers a
surface that should not render. The gate moved onto the new row. The
same check turned up its second half: upstream also added a
`response-streaming` entry to `settingsSearch.ts`, and a gate on a row
does not reach the command palette, so the result would have landed on
General at a hash for a control that is not rendered. Added
`assistantStreamingOnly` beside the existing `providerConfigurationOnly`
filter.

**`authBootstrap.test.ts` auto-merged into something neither side
wrote.** Upstream's two new tests assert `status: "requires-auth"`; the
fork renamed that gate status to `requires-login`. No marker and no
script catches this — `resolution-check.mjs` documents it as the case it
cannot see. The fork's own suite was the only thing that caught it.

## What the merge brings in

### Usable as-is

**Sidebar and chrome (pure client, no RPC).** `ca2cc1339` compact
sidebar rail, `77bca8b2d` compact thread list mode, `df7ccc8fd` compact
thread row badges, `0118b5229` sparse sidebar shelves, `46140c96a`
unified panel resizing, `3689c98d2` hover-highlight fix. The two new
settings keys are `ClientSettings`, which the fork persists to
`localStorage` through `splitPatch`, so nothing reaches
`server.updateSettings`.

**`7b6109988` — linked pull request in the compact rail.** Resolves
through `pullRequests.summary`, the one method in that group Moatless
serves, and opens a surface gated on `capabilities.threadPullRequests`,
which Moatless reports.

**`2ec59ca1f` — hide the back button for a single linked pull request.**
Same capability, so this surface is live here and the fix applies.

**`42b6bcc6f` — opt-in in-app thread notifications.** Derived entirely
from the `environmentShell` snapshot; the toggle is a `ClientSettings`
key. No new RPC.

**`6e5e986f1` — badge background thread notifications.** The web half
rides `navigator.setAppBadge` and window focus; the desktop IPC half is
inert in a browser tab.

**`4a39cade9` / `e62868393` — routing recovery.** A `notFoundComponent`,
and `router.invalidate()` in place of `reset()` so startup retries once
the server is back. Both improve the fork's failure path directly.

**`8b3ddf51c`** — `resolveProjectSettings` accepts a null project; filed
as a mobile fix, but the helper is shared and the looser signature is
safer for every caller.

**Mobile-only, in tree but unverified against Moatless** (`564719165`,
`9086a1f71`, `17f8e2a8a`, `0a91b9a11`, `20a8f1de3`): merge cleanly and
change nothing this fork tests — see _Mobile testing against Moatless is
undocumented_ in gaps.md.

### Unsupported in Moatless / needs implementation

**`1bbca0e78` — response streaming mode.**
`apps/web/src/components/settings/SettingsPanels.tsx`,
`settingsSearch.ts`, `packages/contracts/src/settings.ts`. Methods:
`server.updateSettings` (refuses), `server.getSettings` (answers `{}`).
Flag: `FEATURES.assistantStreaming`. Gated in this PR, palette entry
filtered. Closing it properly means Moatless dispatching
`server.updateSettings` *and* honouring the mode server-side — see the
backend bucket below.

**`5e961d3d7` + `f26198d79` — the Connections page, flattened and then
reorganized by environment.** `ConnectionsSettings.tsx`,
`EnvironmentRow.tsx`, `LoadBalancingSettings.tsx`,
`CloudEnvironmentConnectList.tsx` and friends. Flag:
`FEATURES.connections`, via
`FEATURE_BY_SETTINGS_PATH["/settings/connections"]` — the nav entry is
dropped and a typed URL redirects. Both rewrites land entirely on a page
this build does not show: device pairing, SSH environments, WSL, server
exposure. `CloudEnvironmentConnectList` is T3 Connect, which the
inventory's `cloud-relay-connect` concern says not to adopt. Nothing to
implement unless the fork grows a multi-environment story.

**`db6e0531e` — route pull request operations across matching GitHub
accounts.** New:
`packages/client-runtime/src/state/pullRequestRouting.ts`,
`connection/githubRoutingPermissions.ts`,
`apps/web/src/components/settings/GitHubRoutingSettings.tsx`, plus
`apps/server/src/pullRequest/*`. Methods: `pullRequests.routing` and
`pullRequests.routingIdentity`, both declaring `PullRequestRpcError` —
so they arrived already refusing and the derivation reports nothing to
add. `routingIdentity` answers which GitHub account an environment is
authenticated as; `routing` picks the environment whose account can act
on a given PR. That is a git-host integration Moatless does not have,
and it means nothing before the `pullRequests.*` group underneath it is
served, so it closes with the group rather than separately.

**`66e39ca2a` — apply device settings to selected environments.**
`DeviceHostsSettings.tsx`, `IntegrationsSettings.tsx`,
`deviceHostsSettings.logic.ts`. Methods: `server.updateSettings` plus
the whole `device.*` group (`configure`, `list`, `testHost`, `open`,
`close`, `shutdown`, `detail`, `action`) and the `subscribeDeviceState`
stream — all nine refuse. Flag: `FEATURES.deviceHub`. Rewires device
toggles onto `useScopedSettings` so they fan out to the selected scope.
Closing this means Moatless running an iOS Simulator or Android emulator
host for a task — real work, since the hub needs Xcode or the Android
SDK on the host plus a second stream beside the RPC connection.

**`d7c71f91d` — float the pull request comment composer.** New
`PullRequestCommentComposer.tsx`. Methods: `pullRequests.comment`, on a
panel fed by `pullRequests.detail` and `.activity`. Flag:
`FEATURES.pullRequestSurface`, plus `capabilities.pullRequests`, which
Moatless does not report — `ChatView` renders the unavailable state
before the composer can mount.

**`20363c32c` — provider selector in the pull request toolbar.**
`routes/_chat.pull-requests.tsx`. No fork gate needed and none added:
the route computes support from `capabilities.pullRequests`, so with
none it short-circuits and the new menu never renders.

### Backend behavior to consider reproducing in Moatless

**`c07575f57` — deliver finished paragraphs and closed code blocks
mid-turn.**
`apps/server/src/orchestration/Layers/ProviderRuntimeIngestion.ts`. The
most interesting item in this range. Upstream splits the buffered
assistant message at the last blank line or closing fence that is *not*
inside an open code block, flushing at most once every 400 ms. That
needs no token stream — it is a buffer-and-split on the message the
backend already builds, which makes it the reachable middle ground
between what Moatless does now (whole messages, which is why
`FEATURES.assistantStreaming` is off) and real streaming. The client
half merged inert and is ready: the fade-in is gated on `data-streaming`
and the smooth end-scroll on `isWorking`. Reproduce the split boundary
faithfully — a naive split on `\n\n` breaks fenced blocks and tables.

**`21d53ca2e` — browse ignored files and load folders on demand.**
`packages/contracts/src/project.ts`,
`apps/server/src/workspace/WorkspaceEntries.ts`. This one is a served
method that now under-serves: `projects.listEntries` gained an optional
`directoryPath` (present → one directory's immediate children including
gitignored ones; omitted → the old recursive indexed listing), an
`ignored` flag on `ProjectEntry`, and a `directory_list_failed` literal.
Nothing breaks — `useDirectoryEntries` filters by parent path, so a full
listing still yields the right rows — but it degrades quietly: every
folder expansion re-reads the entire workspace index, the new four-way
request pool has nothing to pool, and gitignored files stay invisible.
Small, well-scoped backend change with a clear payoff on large repos.

**`3b75e607e` — the handshake now fails on an environment-id mismatch.**
The server half is upstream's own dev-auth feature, which the fork does
not adopt. The client half applies regardless: after `initialSync`,
`packages/client-runtime/src/rpc/session.ts` compares
`config.environment.environmentId` against the id the connection was
registered under and fails with a `ConnectionBlockedError` instead of
proceeding. **This is a live risk, not a missing feature** — it passes
every check in this repository and would fail on first contact with a
deployment. Moatless must echo, in its `subscribeServerConfig`
descriptor, exactly the `environmentId` the client registered.

**`dd6ba84dc` — fall back when a new worktree is unavailable.**
Preflights whether the project cwd is a git repository and whether the
base ref names a real commit, degrading to the project checkout rather
than failing thread creation. Largely moot today
(`FEATURES.worktreeSelection` is off, Moatless never sets
`worktreePath`); recorded because the shape is the one to copy if
worktrees ever arrive.

**`3138f5716` — task lifecycle for monitors and background shells.**
Grok-adapter-specific and not reusable, but the contract-level behaviour
is: the client already knows how to render a task's lifecycle, so a
Moatless task that spawns a background shell or watcher lights that
surface up for free by emitting the same events.

**`9bf349cf6` — preserve internal agent errors.** A `(?!\[internal\])`
so an over-broad transport-error pattern stops erasing the agent's own
diagnostics. Cursor-specific; the failure mode generalizes anywhere
Moatless normalizes sandbox or agent errors into transport-versus-agent
buckets.

## Verification

`tripwires`, `resolution-check`, `unsupported-methods`, `fmt:check`,
`lint` and `typecheck` all pass. Tests pass per package: web `419`,
mobile `319`, relay `30`.

- `resolution-check` — 15 fork-delta paths still differ from upstream,
18 carry upstream's change, 17 `theirs-verbatim` paths byte-identical.
- `tripwires` — Clerk/T3 Connect 4 files, device pairing 90 files, T3
session bootstrap 8 matches, exactly the 5 known deletions, 3 active
workflows all allowed.
- `unsupported-methods` — **ADD 0, DROP 0**, so
`packages/contracts/src/rpc.ts` needed no edit. 151 methods, 93
declaring `UnsupportedMethodError`, 67 backend-dispatched.
- Sweep — 8 owned-concern hits, all false positives: auth and pairing
strings inside upstream's own `apps/server` auth work, which was not
taken.
- No route file was added, deleted or renamed upstream, so
`regen-route-tree.mjs` correctly skipped.

Three caveats, none of them a merge regression:

1. **`@t3tools/desktop` fails on a missing `libsecret-1`.**
`scripts/browser-secret-native.test.mjs` shells out to `pkg-config
--cflags --libs libsecret-1`, which is not installed in this sandbox.
The file is untouched by this merge (`git log HEAD^1..HEAD -- <path>` is
empty). Already recorded in gaps.md as _The desktop suite needs
libsecret_.
2. **`duplicate-adds.mjs` reports 2 false positives.** `auth.ts` → `if
(` is a line written by the `decide` resolution: it appears once in the
merge and in neither parent, so it cannot be a kept-twice duplicate.
`authBootstrap.test.ts` →
`expect(testApi.calls.browserSession).toEqual([]);` traces to two
genuinely distinct tests, one from each parent. The script's own tell is
that a real duplicate breaks lint, typecheck and test at once; all three
are green.
3. `@t3tools/mobile` failed once under the parallel run and passed alone
— CPU contention.

## Fork documents

- `docs/fork/inventory.json` — four new `pathPolicy` rows for files
resolved this merge that had no policy: `settings-surface-gates`,
`settings-search-filters`, `sidebar-fork-chrome`, `root-route`.
`inventory-check.mjs` passes.
- `docs/fork/gaps.md` — extended the pull-requests entry with the
routing pair, and added four entries: the `listEntries` degradation, the
streaming gate and the paragraph-split behaviour behind it, the
environment-id handshake guard, and three more server fixes under
_Runtime fixes upstream made to its own server_.
- `docs/fork/upstream-merge-log.md` — dated entry.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---
Moatless task:
https://moatless.soaplabstest.com/tasks/13aa8647-9eb6-4052-8edf-e417893d7abf
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.

1 participant