If this happens right now, what do users do? They probably take some other computer and Google and find that you have to choose the previous kernel, or edit the kernel cmdline. In the new world, they Google and find that you have to hold Control and choose the previous kernel, or edit the kernel cmdline.

That said, we need to start building proper error reporting to detect these cases. It doesn't have to be perfect; catching 80% of everything is much better than catching none.


On Tue, Mar 12, 2013 at 12:19 PM, Nicolas Mailhot <nicolas.mailhot@laposte.net> wrote:

Le Mar 12 mars 2013 16:10, Peter Jones a écrit :
> On Mon, Mar 11, 2013 at 12:58:05PM -0400, Matthias Clasen wrote:

> The idea would be to have a positive indication from systemd that
> we've gotten to some pre-defined point on the previous boot (say,
> starting your login manager), and not to show you any menu unless the
> previous boot didn't get that far.

This assumes nothing can go wrong after the login manager is started (for
example, the login manager hitting a selinux denial when it tries to use
new features exposed by the new kernel), and that the system is able to
detect a running, but useless login manager (input or gfx broken by new
kernel)

Did anyone check the X guys were ok with a setup where they had no longer
any room for error? They heavily depend on users being able to boot on the
previous kernel when there is a driver problem.



--
  Jasper