Repository navigation
Conversation
|
Thanks, Kirill. Stopping a helper once nobody wants its picture is worth having. But this is stacked on #67's helper machinery ( To make it mergeable, please rebase it onto main on its own, without #67. On main the change can be small: wait for the helper in short slices, kill it when the response's |
8f833ec to
84ad9fe
Compare
|
Thanks for the clear direction. Reworked this independently on main Kept one QML regression, |
[Medium impact] The PR appears safe to merge. SummaryDiscarded image requests now stop their external thumbnail helpers so workers can serve newer images sooner.
No actionable issues found.
|
84ad9fe to
0d0e0f0
Compare
Summary
Pass discarded image requests' cancellation flags to main's existing thumbnailer and
pdftoppmprocesses. Wait in 100 ms slices within the existing 15-second deadline, kill the helper when cancelled, and avoid recording a thumbnail or failure marker for cancelled helper work. Sharp-preview fallbacks pass the same flag.Reworked independently on main
7c058ce, without #67's process sessions,Ranor shutdown machinery. No new dependency or renderer.Evidence
thumbnailsStopWorkWhenItsImagesGoAwayregression fails on main: the follow-up image waits about 2.5 seconds, beyond its 1-second deadline.After: the same real QML/provider regression passes and checks both thumbnail and failure caches after the workers drain. It also passes with
slowinTMPDIR.bin/test: 233 passed, 0 failed, 1 skipped (the root-drive trash test needs a writable/). Local Qt 6.12, English locale, offscreen/private session bus, image plugins limited to the package dependencies; not a container-CI claim.git diff --checkpasses. Three files, one retained QML regression.Merge Danger
Door: two-way; reverting restores the original waits, with no cache migration.
Blast Radius: external thumbnail helpers. This stops the direct
QProcess; process-session/descendant cleanup remains outside this patch. In-process decoding is not interrupted.