Re: Installing, then installing CDE

On Tue, 18 Aug 2026, Liam Proven wrote:
> On 18/08/2026 2:03 pm, Stephen Borrill wrote:
>
> Thanks for the reply!
>
>>> I managed to install it, and _try_ to launch it with `startcde` but it
>>> shows a full-screen splash message, then dies with an error.
>> 
>> Do you have the error?
>
> No -- after a long interval it went back to a text screen with an error about 
> X11, I think. (It was getting late!)

No matter, I reproduced it myself.

>>> The default X11 WM (`ctwm`?) but CDE won't start.
>>> 
>>> I tried editing /etc/rc.conf but somehow killed my VM and now it won't
>>> boot at all. I am hoping for some pointers or guidance before I start
>>> from scratch.
>> 
>> What changes did you make to rc.conf? What happens when you try to boot it?
>
> I added a line to the bottom about starting `dtlogin`, saved, and rebooted.

rc.conf is a shell script, not a declarative configuration even though it 
looks like on.

It looks like you did one of:
- added something badly not in sh format to rc.conf meaning it could also 
not read rc_configured=YES to go multiuser.
- extracted etc.tar.xz on top of an existing install (will overwrite 
rc.conf)
- somehow corrupted rc.conf - perhaps your hypervisor is failing to flush 
data correctly


>> rc.conf will be read after init starts and runs /etc/rc after the kernel 
>> has initialised. So you cannot make a machine not boot by editing rc.conf
>
> Well, it seems I can! :-(

As the next part of my sentence said, "but you can make it not boot 
multi-user". So you have a partially booting system which means we can fix 
it :-)

> In I think my previous post on this list, I killed my bare-metal NetBSD 10 
> install by trying to update to 10.1. I have not had much luck with NetBSD so 
> far; it took multiple generations of VMs to get a vague hang of it and about 
> half a dozen attempts to achieve that bare-metal install.

IIRC that previously seemed to be down to odd filesystem corruption too, 
persumably on the same virtualisation system...

> My day job is the FOSS reporter at The Register and I want to write a story 
> about the 11 release. Since almost no Linux distros include CDE, I thought 
> that would make a great screenshot. And if I can learn how to do it in a VM, 
> I will nuke my hardware install and reinstall 11 with CDE in place of Xfce.

Agreed.

>> but you can make it fail to come up multi-user if rc errors out (e.g. if it 
>> cannot mount filesystems).
>
> I am not sure where I went wrong. I edited in vi, added a line, saved, quit, 
> and typed `reboot`.
>
> Maybe the reboot command is wrong? It said it was flushing and unmounting 
> cleanly...

No, that's exactly what I've been doing (always).

>> If you are stuck in single-user mode, you should be able to mount -a and 
>> then use vi to edit rc.conf to undo your changes.
>
> You misunderstand me. It says something is wrong with the entire file.

As martin says, this means that rc_configured=YES is not present or not 
readable in the file. You can cat /etc/rc.conf even with everything 
mounted read-only to see what is in it.

The default rc.conf does not have rc_configured=YES in, so the value in 
/etc/defaults/rc.conf are used. sysinst (the installer) adds 
rc_configured=YES when the installation completes. So if there is manual 
installation, you need to remember to set it yourself.

> On 18/08/2026 2:56 pm, Stephen Borrill wrote:
>> I don't use pkgin.
>
> :-o
>
> Why not?

I'm old-school :-)

I tend to compile my own binaries or, in this case, I have no packages 
installed and want some from a consistent set. pkg_add in base will do all 
that without needing to involve pkgin. Just uncomment one line in your 
.profile.

So on my VM I also did pkg_add firefox and that just worked too.

Aside: in CDE it uses .dtprofile not .profile which means things like PATH 
are wrong (no /usr/pkg/bin or /usr/X11R7/bin). This is documented at the 
end of the .dtprofile file it creates in your home area,

pkgin is more like apt-get, it deals with the full lifecycle of packages 
including upgrades.

>> 7) echo "dt1ogin=YES" >> /etc/rc.conf
>
> That looks like a 1 not an L in the command name. Typo?

Yes, sorry.

>> 8) reboot
>>
>> I now get the CDE login graphical login screen
>>
>> 1) Enter username
>> 2) Enter password
>> 3) Look at Starting CDE splash screen
>> 4) Full CDE desktop starts up
>> 5) Get "Action Required" dialog saying "The desktop messaging system could
>> not be started". While this is open, you can't interact with the rest of
>> the desktop. When you close it, the desktop exits and you go back to the
>> login screen
>>
>> Discover from searching that rpcbind needs to be running for ttsession
>> (ToolTalk) to work, so start that:
>> 1) echo "rpcbind=YES" >> /etc/rc.conf
>> 2) /etc/rc.d/rpcbind start
>
> Aha. How did you find this, please?

I just Googled for "cde The desktop messaging system
could not be started", I think.

-- 
Stephen