Lars Wendler (Polynomial-C)
Gentoo package maintainer and bug-wrangler
Am Donnerstag 21 April 2011, 03:12:21 schrieb Donnie Berkholz:
> On 13:32 Thu 14 Apr , Kfir Lavi wrote:
> > When i run world update, I usually don't really check all the written
> > stuff.
> > If I do this, I'm sure a lot more Gentoo users do the same. So do
> > expect people rebooting the machine without checking what your have
> > wrote. This can be a major headache if you have few systems that are
> > doing auto updates. I would solve this issue by stopping the emerge
> > and getting the attention of the user. If I don't get the attention of
> > the user, no openrc will be installed. It should be something like
> > emerge -C ... 1 .2 3 4 5...
> > To conclude, you can't issue such a change without proper confirmation
> > from the user.
> I know this is the case. You're going to get literally thousands of
> people (or more) who break their Gentoo systems if that indeed is the
> consequence of not reading the migration guide and doing some action.
> From a glance over the guide, it wasn't immediately obvious what in
> there would result in a broken system. Perhaps it's the "run
> dispatch-conf" that's buried in the middle of a paragraph without enough
> emphasis? That's particularly confusing for people who use etc-update
> instead, and it *needs* to move somewhere more obvious like a separate
> code listing with big <important> tags and bold text. The line of red
> text just isn't enough, it needs to stand out even more.
> It seems like nobody's really clear on what exactly happens though,
> since I've seen people talking about this *maybe* resulting in an
> unbootable system. Has anyone tested it?
I didn't test it intentionally. The last time I accidently rebooted a system
freshly moved to bl-2/openrc without updating the config files the boot process
threw a couple of strange errors. I cannot exactly remember what kind of
errors that were but the result was a system hanging in the middle of the boot
process with a message similar to "nothing left to do in this runlevel" and I
wasn't able to log into the system.
Another problem I've once encountered after updating a system to use openrc
was no running udev daemon after boot. I first didn't notice this but X didn't
start and funny part was that X won't tell you it cannot start because the
devicenodes in /dev for the graphics card were missing. So took me nearly a
day of frustrating research until I found that the udev init script wasn't
added to the sysinit runlevel. Of course this is mentioned in the migration
guide but it should be explicitly pointed out how fatal this can be to not
have udev getting started.
I can offer to "abuse" my two stable VMs (amd64 / x86) for this to test if
there's interest in getting "exact results".
> One potential cleaner approach to the same idea Kfir suggested is to
> make it an interactive emerge with an ACCEPT_LICENSE-like feature that
> pops up something you must read and agree to.