Justin B Rye wrote on Jun 10: > Sorry, no. This dist-upgrade has been the trickiest I've seen for a > decade or so, and most of the glitches aren't really documented. > > #1) X starts needing xserver-xorg-input-libinput, which will be pulled > in by default if you've got xserver-xorg-input-all installed, but not > otherwise (symptoms: input stops working). There's a new item about > libinput, but it basically fails to address this issue.
In particular it fails to mention the crucial detail below. > #2) independently, startx stops working with similar symptoms, but > nothing I've found makes it work again, regardless of what else I > install (in particular, xserver-xorg-legacy makes no difference). > The only solution I've found is to switch to lightdm (which might as > well happen before the dist-upgrade). If nobody else is seeing this > then I don't even know what to report a bug against. After confirming that I was also getting these symptoms on a third machine with less antiquated hardware, I've finally tracked down what I was missing: the user starting the session needs to be a member of group "input". Or root, of course, which is why it's not needed with lightdm; but it's a new requirement with libinput that doesn't seem to be documented anywhere - certainly not in release-notes section 5.3.3. (In fact it's so undocumented that even the page for libinput on the ArchLinux wiki never mentions the word "group"! I can see mentions of the input group in old touchpad HOWTOs written in the days when libinput was an obscure alternative approach, but there's no hint it would be needed with my oldfashioned psaux mouse and keyboard.) > #3) after an upgrade, ifupdown is marked as automatically installed > and no longer held in by dependencies, so things like aptitude want > to uninstall it as junk. Fortunately I'd noticed this was likely to > happen before I started doing upgrades, so I've never had my network > crippled by this. I suggested some text about this in > https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=864166#57 > > #4) at the end of the dist-upgrade needrestart restarts networking, > which fails - the wifi interface gives errors as if I'd defined two > gateway addresses. You can tell needrestart not to do this, but > fortunately it's fixed by a reboot, and a reboot was the next step > in the dist-upgrade anyway. However, it's easy to imagine ways this > could combine with the other items on this list to be a real problem. > I can't reproduce the issue outside the context of a dist-upgrade, so > again I don't know where the bug is exactly, and if nobody else is > seeing it then I don't know if it warrants being mentioned in the > release-notes. Again, I can confirm this as absolutely reproducible, but it only happens if you've got needrestart installed, which according to popcon is under 2% of users (and hopefully the kind that can cope). And under normal circumstances, "systemctl restart networking.service" still works correctly on Stretch. > It's possible that #2 amd #4 only hit me because I'm doing my trial > upgrades on an old machine, and I'm not using a mainstream desktop > environment (just fvwm). But I haven't heard anybody claiming to have > done any bug-free upgrades, so it's hard to be confident. I've seen no sign of other sufferers from #2 or #4 on debian-user, so maybe it doesn't matter. -- JBR with qualifications in linguistics, experience as a Debian sysadmin, and probably no clue about this particular package
