[arimendelow]

Cause, two faults:

• App: leaving with a jump still on its way logged "its target kept" and then wrote it only after reading the page. The jump's chapter, still loading, held that read until iOS had ended Gatefold, so the reopen came back at the book's start. Run 43277705735: "leaving a jump still on its way; its target kept at … 0.62883", no place write after it, then "place read: opening on none".
• Test: testAScrubberJumpKeepsThePlace read the page it settled on 20 s plus sleep(2) after the drag. In that run the jump's chapter was still loading 25 s after the drag (load average 528), so the test read the page from before the jump and failed at :72.

Fix:

• The jump's target is written at once as he leaves (writeJumpTargetAtOnce, in owedPlaceOnLeaving), as a turn is (Reopen after Home and an evicted app lands pages before where he left (build 25) #301). The ordinary leave write still follows. The jump's own report, if the app lives on, still writes his page and words over it. This covers every jump that sets jumpOnItsWay: the scrubber, bookmarks, highlights and search results.
• -GatefoldHoldLeaveWrites (UI tests only) holds the ordinary leave write for good. testAScrubberJumpLeftWhileItsChapterLoadsKeepsItsTarget now runs with it, so only the at-once write can reach the reopen.
• The scrubber and contents jump tests wait for the jump to land (veil down, new words) in up to 1200 reads. They read the page left and the reopen with ReaderReads.settle, never a sleep. ReaderReads.swift is ReviewerChecks' Page Numbers test waits in reads on the reader's ready and count reports, never 120 s (#122) #331's file, copied unchanged; main is merged in once ReviewerChecks' Page Numbers test waits in reads on the reader's ready and count reports, never 120 s (#122) #331 lands. The "moved the page" check is unchanged.

Proof:

• Probes, each run once, only these two tests on iPad:
  • Red, the at-once write removed: run 43313011769. Leaving logged the target kept, the held write never landed, and the reopen read "opening on none". The test failed on the reopen showing the book's start, which has no words, for 200 reads after ready. testAScrubberJumpKeepsThePlace passed, the jump landing after 78 reads.
  • Green, this fix: run 43313016177. "the jump's target written on leaving, at once, at 0.628827", reopen "opening on … total 0.62883"; both tests passed, the jump landing after 24 reads.
  • Both runs are red only on the offline CI scripts check, which objects to the probe's trimmed workflow.
• Unit tests: two new LeavingAScrubberJumpOnItsWayTests (inactive and close, the write's wait held) fail without the fix and pass with it; the 19 leave tests pass.

An independent review found no blocker. Its should-fixes are in: the landing is judged with the veil down, and the helper is placed away from #325's so the two merge cleanly. When #325 lands, its writeContentsTargetAtOnce and this writeJumpTargetAtOnce could become one helper.

Fixes #309