SUMMARY I haven't verified it yet but it appears KRunner launch count stuff isn't persisted properly anymore. Especially now that the recent docs runner results are all over the place, this gets more noticeable, that it doesn't reliably "learn" your preferences anymore. STEPS TO REPRODUCE 1. Search "spotify" 2. Get pointless recent documents results and Spotify app at the end of the list 3. Launch Spotify app 4. Search "spotify" again 5. Get Spotify at the top of the list because of launch count as expected 6. Now kquitapp5 krunner 7. Search "spotify" again OBSERVED RESULT Gets same pointless results as at the beginning EXPECTED RESULT Remembered Spotify app launch count thingie and prefer it over the recent docs results, to "learn what the user launches" Same also happens with Telegram. SOFTWARE/OS VERSIONS Linux/KDE Plasma: (available in About System) KDE Plasma Version: 5.19.80 KDE Frameworks Version: 5.72 git build of 2020-07-08 Qt Version: 5.14.2
I can confirm this issue but it is not a regression. I will do some testing, but already have pretty much identified the issue. Will write a patch soon :-).
There are actually two issues: 1: The entries don't get properly persisted, this can be fixed by syncing them after each session. Just like the history entries. 2: In milou lib/resultsmodel.cpp line 87+ we prefer the type over the relevance. If we launch an entry very often the position changes only inside of its group, pretty much the same issue as in BUG 340283.
Git commit 45d42bb9520ff3130753535012ec3939df232a37 by Alexander Lohnau. Committed on 04/08/2020 at 05:02. Pushed by alex into branch 'master'. Save RunnerContext after each match session If we only save the context in the deconstructor the entries don't get saved when we kill krunner/shutdown/reboot. Currently they only get saved when we end krunner using `kquitapp5 krunner`. The history items are also synced after each session. M +4 -6 src/runnermanager.cpp https://invent.kde.org/frameworks/krunner/commit/45d42bb9520ff3130753535012ec3939df232a37
Marking this as fixed: A better rating system is mentioned in other bugs and a different system of storing launch counts is already in progress.