fix(channels): prevent clipped activity rows and remove separators

[tellaho]

Overview

Category: fix

User Impact: Channel activity previews have smooth, unclipped hover backgrounds without dividers between threads.

Problem: The channel hover popover clips its activity rows' rounded corners, and separators between multiple rows make the hovered background look disconnected.

Solution: Match the rows to the popover's inner curve and remove the local dividers. The outer frame, spacing, shared design tokens, scrolling, and thread-opening behavior stay unchanged.

Changes

File changes

src/bundled/channels/Channels.module.css
Derive the activity-list row radius from the existing panel radius minus its inset and border, and remove the adjacent-row border rule. This stays local to channel activity previews.

tests/browser/channel-activity-corners.spec.mjs
Add single- and multiple-row journeys through the actual app and production broker, using the existing isolated test relay. Check painted browser geometry, divider absence, light/dark modes, enlarged text, constrained viewports, and keyboard thread opening.

Current validation at 190807d7:

• Fixed the browser regression's layout setup: close the hover popover before resizing, then reopen through the real channel trigger. The previous test resized an open popup away from its stationary pointer and raced normal dismissal. Production code, timeouts, retries, and animation behavior are unchanged.
• bin/pnpm test:browser tests/browser/channel-activity-corners.spec.mjs --no-deps --project chromium --project webkit --repeat-each=3: 12/12 passed, two workers, 35.8s total on Apple Silicon macOS. Individual cases took 4.0–5.1s. This is repeated local evidence, not Linux CI confirmation.
• The same single/multiple rows, all four corner radii, divider absence, light/dark modes, enlarged text, constrained viewport, scrolling, and keyboard thread opening remain covered. The multi-row case now explicitly asserts overflow; keyboard navigation stays in the short viewport.
• Repository staged-file and pre-push checks passed: formatting/lint, TypeScript, selected app-composition test, design typecheck, and design guards.
• Hosted run at the preceding head b5d3163e passed JavaScript (including the merged test(relay): stabilize per-channel replay boundary coverage #378 replay-test fix), Rust/tool integration, measurements, and four browser shards. Only this new hover regression failed on the other two shards. Fresh hosted CI for 190807d7 is pending.
• Based on 23929d76; the CSS feature patch and the original screenshots below are unchanged. Screenshots retain their original capture attribution.

Validation at d84986c:

• Rebased onto 5ce7836b without conflicts.
• bin/pnpm test:browser channel-activity-corners.spec.mjs --project chromium --project webkit --no-deps --workers=2: 4 passed (9.7s locally).
• Repository staged-file and pre-push checks passed: formatting/lint, TypeScript, selected app-composition test, design typecheck, and design guards.
• Browser cases: 2 added per engine; 0 removed. Rounded overflow clipping and computed corner geometry require a browser, not jsdom. Minimal fixture histories are reused.
• Fail-then-pass evidence: both Chromium cases failed the radius assertion with the original CSS during implementation; all four cases pass with the fix. The original multi-row capture also confirms the removed 1px divider.
• Remaining gaps: current-head hosted CI, native/live-relay acceptance, independent review, and explicit human acceptance of the final state. The earlier live-account check encountered relay throttling; isolated-relay checks are not live-data acceptance.

Reproduction Steps

1. Hover a sidebar channel with unread activity in a single thread.
2. Hover its activity row. The highlight should follow the popover's rounded inner edge without cropped corners.
3. Repeat with unread activity in multiple threads and move between the rows. There should be no separator lines.
4. Use a short viewport to scroll the list, then focus a row and press Enter. Its thread should open and the popover should close.

Screenshots/Demos

Real production-built app in dark mode with neutral, isolated test data; cropped to the activity popover. Before uses the base CSS at 5ce7836b in the otherwise identical rebased app; after is d84986c. No UI replica or component showcase.

State	Before	After
Single hovered row
Multiple rows, last row hovered

Implementation and validation performed by Carl, an AI agent, under the author's direction.

[wesbillman]

Star Lord — automated source review via Wes’s account

Reviewed head 190807d71ac44a4d40427aec80c6e9d34ac89e6a against base 23929d764697d1747d0ba711ad47598d172cd07e. This is a non-blocking COMMENT, not approval or merge authorization.

Finding

P2 — Remove internal deployment/agent identifiers from the public commit metadata. Both PR commits (b5d3163e34b8e2bd3a244dd0c4c6efe73b174bdd and 190807d71ac44a4d40427aec80c6e9d34ac89e6a) include a Co-authored-by address containing an internal coordination hostname and a stable agent identifier. In this public repository, that publishes an unnecessary association between the deployment and the agent identity. This is a public-material privacy finding, not a secret-key leak or a CSS defect. Please have the author sanitize those addresses using an owner-verified, public-safe attribution identity while preserving actual contributor credit and valid DCO certification. Do not invent a replacement identity or substitute someone else’s authorship. The sensitive strings are intentionally not repeated here.

Code assessment

No actionable production-code defect found in the two-file diff. Channels.module.css:628–635 scopes the radius override to the activity list and removes only its adjacent-row divider. The formula matches the current shared frame’s panel radius, list inset and 1px border (styles/materials.css:67–71, styles/components.css:916–950); NavigationItem consumes the inherited radius. No shared token, spacing, unread acknowledgement, or navigation owner is changed.

I traced the hover/close lifecycle, stale/profile-failure presentation, thread-opening handoff, thread error/retry rendering and close-focus restoration. The CSS introduces no new focus transition. Source inspection does not establish runtime focus correctness; the added test checks programmatic row focus plus Enter, not a complete keyboard-entry/error-recovery journey.

The two new cases per engine are appropriately browser-level geometry coverage, reuse minimal upstream fixture data, and resize only while the hover popup is closed. No cases are removed. Screenshots are captured artifacts, not automated pixel comparisons. Engineering assessment: 9/10 minimalness and elegance; 9/10 source-level code correctness; 8/10 overall until the concrete publication finding above is addressed.

Evidence and limits

• Inspected the full PR description, all four attached before/after images and both commit messages. No additional disclosure found in those images, the description or changed source. The images are explicitly attributed to older snapshots, not runtime evidence for this head.
• Hosted CI run 36496893087 identifies this head/base and reports success, including all six Chromium/WebKit shards, JavaScript, Rust/tool integration and measurements. DCO/security checks also report success; Windows native validation is skipped. This snapshot supersedes the description’s pending-CI wording, but is not my execution evidence.
• The description reports a current-head 12/12 repeated local browser run in 35.8s, with 4.0–5.1s cases; I did not reproduce or independently measure those timings. Native/live-relay acceptance and explicit final human acceptance remain unverified; the repository readiness checklist is not established by this review.
• Source-only: no PR code, tests, builds or apps executed; no installs or source changes. Reviewed an isolated archive whose 1,646 Git blobs match the pinned head and original content manifest, not a modified live checkout.