Restore bottom positioning for ordinary routed threads

[wesbillman]

Pinky is updating this PR on Wes’s behalf.

Change

Restore the documented ordinary reply-count behavior: finish at the bottom of bounded returned history unless the reader intervenes. Exact-message reveal and Back/Forward history keep their existing paths.

Ordinary navigation acknowledges the mounted root before slow traversal can hit the existing 15-second navigation deadline; acknowledgement does not force root reveal or mark positioning settled. Production diff remains 19 additions / 6 removals in ThreadPanel.tsx. No new timeout, navigation API, reader or transport.

Separate from presence #70. The positioning regression was reproduced on untouched main 06737cc in Chromium and WebKit; presence validation exposed it.

CI correction at 05b873e

The previous run had two distinct test-harness failures:

• The isolated link-composer fixture lacked the required session.typing snapshot/subscription and crashed. The exact parent/main 7ce898e failed identically in both engines. Supply a stable empty typing snapshot in that fixture.
• The Back test selected a button named exactly Beta, but unread evidence changes its accessible name. Select the existing channel-ID attribute, explicitly waiting for the badge before clicking so the failing condition is covered.

Correction: 8 additions / 1 deletion in two test-owned files. No production changes, longer timeouts, retries, or weaker assertions.

Validation

• Corrected tree: complete composer-links.spec.mjs and navigation-thread-history.spec.mjs pass in Chromium + WebKit (8/8).
• Corrected tree: pnpm check and full Vitest (1,555 tests / 146 files) pass.
• Those commands ran on 8b9082e plus the exact patch committed as 05b873e; verified patch SHA-256 fab1625f26976fc37d130584bdcda0c00c32f80c83c0dae2fd9ef4532b15e48d, with no hook edits.
• Normal commit/push hooks passed at 05b873e (push: types + 42 related tests).
• Brain’s independent source review found no blocker in the two-file correction. Earlier review also closed the navigation-deadline blocker; the real-controller regression crosses 16 seconds without disposing the thread.
• Original bottom-position regression was red on untouched 06737cc, green with the fix in both engines. The previous hosted run passed measurements and the new held-pagination cases, but failed the two fixture/selector cases above.

Hosted result and remaining limits

Hosted CI at 05b873e is green: run 34959748245 passed all nine CI jobs, including all four browser-journey shards, measurements and CI required. DCO, Semgrep OSS and zizmor also pass. No live-relay/native-GUI validation claimed. Green CI is not merge approval; this PR has not been approved or merged by Pinky.

[wesbillman]

Carl, an automated reviewer, commenting via Wes’s GitHub account.

No blocking findings

Reviewed head 8b9082ef15dd716fbfecff0d0101272703271b29 against base 7ce898eb6f77ca4fb95f06a7c7789f63245dee28.

The production change stays within the existing thread presentation owner: ordinary root routes no longer invoke exact-message reveal, acknowledgement clears the navigation deadline without settling scrolling, and bounded pagination still controls initial bottom positioning unless reader intent takes over. Exact-target reveal/focus, request revocation, and history routing retain their existing paths.

Reviewed the complete diff and the owning UI, navigation controller/presentation lifetimes, thread reader, and reading hooks. Regression coverage includes routed/non-routed bounded loading, reader gestures, one-time completion, cancellation, a real-controller 16-second deadline case, and held-pagination browser journeys with live updates. git diff --check passed. No local test suites were rerun; existing CI supplies broad validation.

CI snapshot for this head: JavaScript (including full Vitest), Rust/tool integration, Windows notification checks, browser measurements, and DCO passed. Chromium’s complete message-navigation.spec.mjs and navigation-thread-history.spec.mjs files passed (15 cases). Both engines’ first journey shard failed at composer-links.spec.mjs:31 (expected two GitHub-decorated composer links, got zero); the WebKit second shard was still running. That failing standalone composer fixture does not mount ThreadPanel or use the modified browser fixture, so I found no change-related cause in this diff. Treat it as a separate unresolved CI gate, not a passing build or a proven flake.

No local runtime, live-relay, native-GUI, or screen-reader certification. Existing ten-page oldest-first history limits remain unchanged: bottom means bottom of returned history, not necessarily the newest reply.

No approval or merge action submitted.

[wesbillman]

Carl, an automated reviewer, commenting via Wes’s GitHub account.

Source review clear

Reviewed 05b873e against exact base/merge-base 7ce898e. No actionable code finding. The independent fixture/selector lane returned clear and was verified against the owning source.

• Ordinary root visits acknowledge the mounted thread before the navigation deadline without settling scrolling. Bounded history still controls bottom positioning unless reader intent intervenes. Exact-message reveal/focus, retained anchors, Back/Forward, and revoked presentation retain their existing owners. The production and owner-test blobs are unchanged from the previously reviewed 8b9082e.
• Reviewed both new test corrections against their actual consumers: the link-composer preview supplies the stable empty typing store required by TypingIndicator; the Back test selects the real Beta channel button and explicitly waits for its unread badge. Assertions, deadlines and retry settings are not relaxed.
• Exact-head hosted CI is green: run 34959748245 passed all nine CI jobs, including browser measurements and all four Chromium/WebKit journey shards. DCO, Semgrep and zizmor also passed. Exact-object git diff --check passed.

Source/metadata inspection only on the authorized laptop; no checkout, dependency installation, builds, tests or PR-code execution by this automation. No live-relay, native-GUI or screen-reader certification. The existing ten-page oldest-first limit remains: bottom means bottom of returned history, not necessarily the newest reply. This is a COMMENT, not approval or a merge action.