Re: bin/57820: devel/rcs fails on 10.0_RC2 works on 9.3

[George Georgalis <george%galis.org@localhost>]

RVP, thank you for very carefully looking at this problem.

On Sun, Jan 14, 2024 at 4:17 PM RVP <rvp%sdf.org@localhost> wrote:
>
> I can tell you why this error happened: you didn't install the `text.tar.xz'
> tarball which contains `soelim'. Because soelim is missing, you get this
> error:
>

I suspected it was a date GNUism and had no clue why it would come up.
Obviously, there was an unchecked error in the rcs build, no rcsfreeze.1
man page...

Excellent discovery to identify missing soelim! I went through the full
regression of installing NetBSD and bootstrapping pkgsrc a dozen
times, first not expecting an error, then simplifying my bootstrap
narrative to eliminate irregularities with my deployment. I thought,
I had reproduced the problem with a nb10rc2-noX11 install, but
that is why I repeated the deployment with notes to the list, for
additional eyes to review.

my scripting is loaded with:
   command || { echo "command error" 1>&2 ; return 1 ;}
It saves a lot of headache. In makefiles, I simply make prodigious
use of comments, they are displayed when make stops on error.

> (Don't know why groff is not brought in on NetBSD--it is on Ubuntu
> 23.10--when pkgsrc doesn't find a suitable soelim. But, this is
> one for the pkgsrc folks to figure out...)

I agree entirely. I've worked out building from a minimal install
in the past, and I'm not one to read all the doc before starting,
but this is the only sets doc I've seen...
http://netbsd.org/docs/guide/en/chap-exinst.html#inst-distsets

...I'm using pkgsrc to leverage the separation of os and package
management for the express purpose of having a uniform sw
deployment across Darwin, NetBSD, and Linux. That means
I'm minimally deploying OS for the purpose of bootstrapping
pkgsrc (and nominal other things).

I've made the assertion that automatic dep tests would prevent
this rcs problem. In retrospect I'm not sure how I might better
navigate a minimal install?

Is there an atf workflow that qualifies my netbsd install for pkgsrc?
...rather than expecting 'everything' installed if the OS is netbsd.
Maybe pkgsrc dep tests should be run if the os is netbsd?
And, any other os? vis-a-vis detecting netbsd doesn't mean
the text set was installed.

If my goal is minimal OS, how might I determine the sets required
for a given task? Do the sets have definitions for what they contain
or should not contain? It is natural that pkgsrc is just checking if
the OS is NetBSD if there is no definition of what sets provide.

I do not know "exactly" what PREFER_PKGSRC does, but I'm
using it because I would like to avoid using the base os wherever
possible. Sometimes I can figure out what is missing when there
is breakage, but if the root cause is not installing everything, or
a non /usr/pkg LOCALBASE, that would not be anticipated.

I have re-deployed nb10rc2 as a no-X11 flavor into a single
local partition /dev/sd0a and bootstraped pkgsrc-2023Q4
into /usr/pkg/pkg-2023Q4-NetBSD_10.0_RC2_amd64
and the devel/rcs problem is resolved!

Not sure about the nb10rc2 base make on nfs observation,
but I'll start a new thread if that returns as I resume using
nfs.

Thanks!
-George

-- 
George Georgalis, (415) 894-2710, http://www.galis.org/