Since Fedora-Rawhide-20260128.n.0 , the openQA Japanese input test is failing. In the test, we add Japanese (Anthy) input source and enable it. Then we go to a console and type 'yama'. This correctly shows やま ('yama' in hiragana). Next we press space, and expect it to produce 山 (the most likely kanji character). That's what it did until Fedora-Rawhide-20260127.n.0 . However, now it produces や真 , which is pretty useless. I can reproduce this locally on my F44 Silverblue system. If I hit space again to get the menu, it seems to offer me completions for *each character alone*, not for the whole string - like it's offering me potential kanji for 'や' alone, not the whole word 'やま', and if I hit the right arrow key it scrolls over to 'ま' and starts offering me completions for that character alone. This is a general issue, it doesn't just affect this one test word...it's the same with e.g. ぼく (completion gives 墓句 not 僕) and anything else I've tried. In Fedora-Rawhide-20260128.n.0 I see that ibus-anthy went from 1.5.17-9.fc44 to 1.5.17-10.fc44 and anthy-unicode went from 1.0.0.20240502-13.fc44 to 1.0.0.20260127-1.fc44 ; I guess one of those is likely to be the cause, though ibus-anthy was just a rebuild...
Sorry, I didn't notice the CI of anthy-unicode is failed. The issue is caused because Meson build does not build the dictionaries correctly. I will revert the Meson build to Automake one until the issue is fixed.
Reported the issue to the upstream.
Fixed the Meson issue.