Hello, I could reproduce your issue.
Halim Sahin, le lun. 14 juin 2021 11:08:08 +0200, a ecrit: > when a running desktop crashs or a user does > sudo systemctl lightdm restart > > a11y including orca etc works as expected. > Only java apps which use atk-wrapper is completely inaccessible. > > Then removing all files in /run/user/1000/* works or a full reboot. > > My Question is: which files are needed by java-atk-wrapper? > > It could be an issue of atk-wrapper if it can't determine some > interfaces and use obsolete sockets etc. Well, it's quite involved. Apparently when running lightdm restart or the session crashes, the dbus session bus does not actually terminate. And thus the at-spi bus does not terminate either... But the only way that jaw uses to determine whether an at-spi bus is running is the venerable AT_SPI_BUS X property, which went away with the X session. C applications go through the at-spi bus which they happen to find through the session bus. It's removing /run/user/1000/bus (the session bus) that "fixes" jaw by making dbus recreate a new session bus, and thus at-spi create a new at-spi bus, and set its address in the AT_SPI_BUS X property, and then jaw can find it. So as usual with bugs there are actually two issues: - Bad dbus cleanup on lightdm restart or session crash. That is a bit surprising, I though all the systemd etc. stuff was precisely meant to keep that kind of thing sane... - jaw still relying on the AT_SPI_BUS X property (but possibly some other toolkits do that as well, but on wayland it'll have to go through a session bus anyway). Samuel
