Fix two Windows CI timing failures: outline landing wait, and unlaid chunks the freeze sweep skipped

[ben-dev-au]

Problem

• test_a_heading_inside_a_chunk_the_scroll_cannot_reach_keeps_the_cursor failed on Windows CI (29 Sep on two dependabot PRs, 11 Oct on Give every example and test one made-up person, in everyday words #196): the fixture's tail brings Findings to the reading line, assert 1 == 2.
• The tests' _landed helper skipped the arrival check the outline's own _landed makes (fnd/tui/outline_view.py): a glide can apply its scroll after the navigation commits, so a slow runner read the reading line at the old offset.

Change

• _landed in tests/test_outline_pane.py also waits until the scroller reaches landing_destination(), as the outline does.

Evidence

• Delaying scroll_relative by 0.6 s (the mechanism test_a_jump_whose_scroll_starts_late_still_places_only_its_target injects) reproduces the CI failure every time: offset 0 against a destination of 36, assert 1 == 2.
• With the wait, the same delayed probe passes 3 of 3.
• All 43 outline tests pass; full gate with both fixes 5,956 passed, all 12 tmux scenarios passed.
• Other outline tests that have failed on Windows CI go through the same helper; not checked whether this was their cause too.

Second Windows failure: a freeze sweep that met unlaid chunks never came back

• test_a_requery_serves_the_unmatched_chunks_and_builds_the_matched_one timed out on Windows (10 Oct on a dependabot branch, 11 Oct here): the first query's sweep never captured the far chunk. The diag held mount served=0 built=60 but no backfill froze=: the post-fill freeze sweep froze nothing.
• Cause: the sweep's await_settled can time out or run out of rounds on a loaded runner. The just-filled chunks then have height 0, freeze rightly refuses each one, and nothing retried. The chunks stayed live (the DOM the freeze exists to drop), and a requery rebuilt them instead of reusing captures.
• Reproduced by making await_settled a no-op: 47 of 47 refusals at height 0, and the CI timeout.
• Fix (fnd/tui/preview/presenter.py): a sweep that met chunks not yet laid out comes back for them in the background (FREEZE_UNLAID_RETRIES = 20, FREEZE_UNLAID_RETRY_SECONDS = 0.1), so no navigation waits on it. Each sweep cancels a pending retry and bumps a generation, so the newest window always wins; app exit cancels it.
• tests/test_a_freeze_sweep_returns_for_unlaid_chunks.py injects the no-op settle: fails without the fix (and with retries set to 0), passes with it. The 22 files touching freeze and captures: 412 passed.
• Follow-up, not in this PR: a same-file navigation to an already-mounted chunk does not cancel the earlier mount task, so its fill sweep can still freeze the new focus chunk (reviewer probe: 1 to 2 in 10, with or without this change).

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: ASSERTIVE
• Plan: Advanced
• Run ID: 8c9a463c-d75e-4ccf-9cc4-76f29cb31809

---

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

---

_{Comment @coderabbitai help to get the list of available commands.}