https://bz.apache.org/ooo/show_bug.cgi?id=126990
Peter <[email protected]> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|CONFIRMED |ACCEPTED --- Comment #17 from Peter <[email protected]> --- After all the years i am on a fix. yay. I try to get it in the next release. Test plan: 1 Functional: These exercise the paths the change touches or deliberately routes around: a) Save existing doc (the fixed path) — Writer, Calc, Impress b) Save As to a new name — non-overwrite branch, must be unchanged c) Save to a network share and a USB stick — the sibling must land on that volume, not %TEMP%. This is the case that would silently lose atomicity if the sibling logic is wrong; check in ProcMon that the temp is created on the target volume d) Save with the doc open in another process holding a handle → should still save via fallback, and the trace should fire. This is the intended graceful degradation e) Password-protected document f) Large document (perf sanity — you accepted the cost, but worth knowing the number) g) Linux, if you test there: symlinked document → must take the fallback and leave the symlink intact 2. The actual crash test VM with a snapshot, so it's repeatable. Run it against an unfixed build first — if you can't reproduce the bug, the harness proves nothing. a) Save a document. b) Within ~1-2 s, hard-kill: notmyfault64.exe /crash, or the hypervisor's Reset (not a clean shutdown — that flushes). Boot, hex-dump the file. c) Unfixed: expect all-zeros on some fraction of trials — it's a race, so run it several times. Fixed: never zeros; you get either the previous document or the new one. I currently cant do the VM, my laptop does not have the free space for that. Linux build is currently difficult too. I do not own a large document, but maybe i can produce one. (lots of pictures) I will note when the patch lands. -- You are receiving this mail because: You are the assignee for the issue.
