On 2019-01-12 13:33:35 +0000, Colin Watson wrote: > On Sat, Jan 12, 2019 at 02:11:25PM +0100, Vincent Lefevre wrote: > > On 2019-01-12 10:36:02 +0000, Colin Watson wrote: > > > The ID changing is presumably the systemd bug you mentioned, and that > > > seems to be the grave part of this. > > > > Actually, I think that the fact that grub-pc changes the configuration > > and loses the previous one is a grave bug too (even though it is > > caused by the udev bug). There should have been a way for the user to > > keep the previous configuration so that nothing gets broken due to > > temporary issues. > > I'm not at all sure that I agree. The configuration in question exists > in order to tell the postinst which devices it should run grub-install > on. If the device does not exist, it obviously isn't possible to run > grub-install on it. > > I can see the argument that it might be convenient to have configuration > that effectively says "install to this device if it exists, otherwise > continue anyway"; but such a configuration may mean that the GRUB image > on disk that you might in fact attempt to boot from will end up being > incompatible with the rest of /boot/grub/ and thus cause a failure to > boot even though the postinst pretended everything is fine! > Disregarding this kind of configuration error can have grave > consequences of its own. In any case, even if it might be convenient to > have such a configuration, I won't accept that part of the issue as a > grave bug.
I meant that the issue was that the user may not have remember or even known where GRUB was installed (this was my case, and I could find the information just because I keep track of the changes of config.dat). This makes the problem unfixable without risking to erase some data. > Note that temporary issues of this kind are extremely rare; this is the > first of its kind that I can recall hearing about. By contrast, it's > quite common for people to accidentally end up with a boot sector that > they thought was being updated when it in fact wasn't. It makes a lot > more sense for the GRUB maintainer scripts to prioritise dealing with > the latter situation than the former. So I'm happy to try to improve > the way the maintainer scripts responded here, but not in a way that > results in silently ignoring missing devices. > > If the dialog box hadn't been cut off in a way that made it non-obvious > that GRUB needed you to select devices to install to, I don't think you > would have ended up in this situation. I don't think so. I think that I would have chosen to let the new version of GRUB uninstalled for the moment (just like what I actually did), because I did not know where to install it. Same loss of information. The only difference would have been that the message would have been clearer. -- Vincent Lefèvre <vinc...@vinc17.net> - Web: <https://www.vinc17.net/> 100% accessible validated (X)HTML - Blog: <https://www.vinc17.net/blog/> Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)
