phenest wrote:We need to know what you've already tried, so we don't repeat anything.
Apart from the list of things I've already mentioned?
Well, there's pages and pages of near-miss search results that relate to using a keyboard (typically Microsoft branded) that has media keys, which appear to cause the mouse driver to believe a mouse click has started but not completed. This doesn't apply: the Zoostorm keyboard I'm using is very Windows-vanilla, with no media keys, just the usual Windows / Menu keys. FWIW, I've got the numeric keypad keys mapped to the virtual desktops in what would be a 3x3 arrangement if only Debian would allow me to consider them in anything but a 1x9 vertical strip. (But that's a grouch for another day.)
I had tried doing the modprobe dance (individually
modprobe -r usbhid, hid_generic and hid, in that order, then
modprobe them back, in reverse order), over an SSH session. Without the
modprobe -r hid, this made no difference. With it, the mouse refused to play at all (cursor included), even after I tried the lock-unlock dance. Another lock-unlock dance, and it came back, working properly. I think I offended its feelings.
There were many other tests (e.g.
service gdm restart) that succeeded in nuking the window manager and all its windows completely (as I'd expected). This is not a desirable outcome.
It's not a swap thrash.
top shows I'm running wholly within the 8GB RAM - in fact, I've over 2GB free - and not touching the 16GB available swap. There are no slow-downs during the loss of mouse clicks: keyboard response remains snappy.
Switching virtual desktops changes only what's on the screens.
orythem27 wrote:Maybe next time it "crashes", you could ssh in and use `journalctl -r` to try to locate any recent message that looks like an abnomaly to you. This might help you obtail more keywords to be put in to search engines.
Thanks - this is relevant. I found the following lines in the journal:
Code: Select all
May 19 10:27:26 machine-name gnome-session[6862]: Gjs-Message: JS LOG: pushModal: invocation of begin_modal failed
May 19 10:26:46 machine-name gnome-session[6862]: Another user connected to the server, forcing the disconnection of the current connection.
May 19 10:26:46 machine-name gnome-session[6862]: ERRINFO_DISCONNECTED_BY_OTHER_CONNECTION (0x00000005):
The Gjs-Message line also appears in a number of other Debian bugreps (#767280, #768911 for instance) referencing similar mouse (mis)behaviour - interesting, but with no useful tips. I don't even have gnome-do installed, so that's not the source. It's also unclear whether the message is related to the cause or the effect of the misbehaviour. An interesting discussion relating to Gnome-Shell at
https://github.com/simonthechipmunk/jum ... w/issues/1 may be relevant, but it's hard for me to see how I could apply its conclusions. It does appear to be a Gnome problem that's been extant for maybe four years, reported not just on Debian but also Fedora (
http://forums.fedoraforum.org/archive/i ... 99502.html), but without any resolution I've been able to find thus far. Interestingly, I've not found one of these bugreps that also mentions the keyboard sometimes going away as well, but I'm sure that there's something in this.