fix(web): sort pull request lists and timelines by time#23588

Problem

The pull request timeline and several list rankings compare timestamp text. ISO-8601 text order can disagree with time order when hosts return UTC offsets or mixed fractional precision.

For example, 2026-07-05T01:00:00+02:00 is 23:00 UTC on the previous day. It currently sorts ahead of 2026-07-05T00:30:00Z, which is 90 minutes newer. An update at 00:30:00.5Z also sorts behind one at 00:30:00Z.

Users see older activity first in the Timeline and older requests above newer requests, including with Recently updated selected.

Change

Sort timeline events with the existing instant() helper. Extract the list's existing numeric recency comparison and use it for merged listings, the Updated sort and its groups, authored/review-requested groups, search ties, merge-readiness ties, and pages appended by Load more. Both "Blocked on me" rankings already compared this way and now call the shared function.

Priority, search score, readiness, and diff-size precedence stay intact. Unparseable list timestamps sort after parseable ones; equal timestamps keep their input order. No timestamp format or server contract changes.

Two focused regressions cover offsets and mixed fractional seconds: one through the timeline and one through the merged list order. The list recency rules now share one comparator, so the regression tests it once at that layer instead of again through each ranking and view.

Scope and approval

This is a small, focused correction to an obvious ordering bug, submitted under the focused-fix exception. There is no linked issue or maintainer approval claim. The same files already parse timestamps numerically for other recency decisions.

The change affects the web pull request page and timeline, including Electron's shared web UI. Mobile uses separate code and is outside this change.

Verification

With the fix reverted, both regressions fail. The timeline returns opened / commit / comment instead of comment / commit / opened. The merged list returns [1, 2, 3] instead of [3, 2, 1]. With the fix, both affected suites pass: 101 tests on the branch, and 108 on a merge with current main.

From apps/web:

To see the failures without the fix, add -t "orders host timestamps by instant".

From the repository root, vp run check:changed passes on the branch and on a merge with current main: format, lint, knip, typecheck, the affected tests and, on the merge, the React compiler check.

Built both web revisions with pnpm exec vp build from apps/web, plus the unchanged server with pnpm exec vp pack from apps/server. Drove the production app in Chromium on Linux. A test proxy supplies controlled PR read responses at the WebSocket boundary, with the same timestamps for each build. The app's production bundles render and sort them. The request titles describe their fixture timestamps; they are not generated UI labels. This verifies rendering and ordering without requiring a live host account. The latest commit changes only tests, so these captures still match the app code.

Desktop: ran the Electron app on Linux with vp run dev:desktop, under Xvfb, on the PR base and on this head, with the same isolated state. A stand-in glab on the server's PATH answers the GitLab merge request reads with the same three timestamps. With Recently updated selected, the base lists the older offset update first, and this head lists the fractional update first. On the Timeline of the offset request, with Newest first selected, the base shows opened, then commit, then comment, and this head shows the reverse.

Not checked:

• Live GitHub, GitLab, or other host responses. A maintainer's live check saw whole-second UTC timestamps from GitHub and consistent UTC from git.cafe, so this PR does not show a host that sends offsets or fractional seconds today.
• Packaged Electron builds, macOS, Windows, or mobile apps. Desktop ran in Linux dev mode only.
• The repository-wide test suite.

Prior PRs: #23395 by Baker-Harrison touched the four list comparisons for a different performance change, replacing locale collation with plain string comparisons. It is closed. That approach keeps text ordering and does not address offsets or fractions. This change reuses no code from it.

Related open PR: #21530 by sanjaibalajee changes the same mergePullRequestLists sort for readable-server fallback. Its current head conflicts with this change at that line; whichever merges second will need to preserve both changes.

Base production list with Recently updated selected: the older offset update appears first in Others.

Base pull request list

Fixed production list with Recently updated selected: Others shows the fractional update first, then the UTC update, then the older offset update.

Fixed pull request list

Base production Timeline: opened, commit, then the newer comment.

Base pull request timeline

Fixed production Timeline: comment group, commit, then opened, with Newest first selected.

Fixed pull request timeline

Base desktop app (Linux Electron under Xvfb) with Recently updated selected: the older offset update is listed first.

Base desktop pull request list

Fixed desktop app (Linux Electron under Xvfb) with Recently updated selected: the fractional update is first, then the UTC update, then the older offset update.

Fixed desktop pull request list

Base desktop Timeline with Newest first selected: opened, commit, then the newer comment.

Base desktop pull request timeline

Fixed desktop Timeline with Newest first selected: comment, commit, then opened.

Fixed desktop pull request timeline

Generated with GPT-6.1 Sol and Claude Opus 5.5 in T3 Code Nightly