<mark_weaver>I've found that several some kinds of errors in the OS config can result in a kernel panic with not enough diagnostic information to figure out what's going on, unfortunately. in my case, I have mispelled some identifier in a service definition.
<davexunit>I'm waiting on a VM build, let's see if this shows the same issue.
<mark_weaver>btw, there's an assymmetry between 'users' and 'groups'. With 'users', you just specify the list of users you care about and root (and others?) get added automatically if needed. With 'groups' you should cons the groups you care about to %base-groups.
<civodul>yes, we need %base-users with "nobody", for instance
<mark_weaver>to my mind, the first approach that comes to mind is to not load any user code at all into PID 1, but only in subprocesses.
<civodul>the first step (for both problems actually) is to emit one .scm file per service
<civodul>the problem is that it's PID 1 that needs to know the service DAG
<mark_weaver>why does PID 1 need to know the DAG, and even if it's true, why is that a problem?
<mark_weaver>also, if we're serious about using DMD instead of systemd, I think that someone(TM) will have to get serious about making DMD kick ass in the ways that systemd does, e.g. things like service activations based on socket listeners, automounts, etc.
<fchmmr>"experiment with guix" is an example of an entry on that list.
<daviid>right, amasing wrk! like guix, guixsd ... who would have thought a while ago guile would turn itself at the center of all this ... amazing really! really cool, it's like getting out of decadent century :)
<fchmmr>it's better to focus on current hardware, and try to get hardware actually assembled that is freedom respecting (that last part is a goal of libreboot. I want it to be pre-installed by OEMs in the future. I'm only 2% certain that this will occur, though)