Re: git: ce348fe5cfc3 - main - amd64 & i386: enable VIMAGE in MINIMAL

On Sat, Feb 3, 2024 at 12:07 PM Kristof Provost <kp@freebsd.org> wrote:

> On 3 Feb 2024, at 11:58, Gleb Smirnoff wrote:
> > We just need at least one supported working kernel without VIMAGE (and
> > probably other useful stuff, that is not useful to everyone).
> >
> Strong agree on this point.
>
> There are enough users out there who do not want VIMAGE (well, at least
> one, but it’s enough of one to care), so we need to ensure it keeps
> building. It’s very easy to forget about the non-VIMAGE case, and if we
> don’t have any kernel configs without we’re going to keep breaking it and
> not noticing until you run into it.
>
> That was even raised in the GitHub review.
>

MINIMAL is not a CI image to test things. It's a replacement for GENERIC
that loads what one can. Its contents need to reflect that. That's why I
did not give the issue weight. However,we likely need at a minimum
LINT-NOVIMAGE like we have LINT-NOINET today. That's easy enough to
arrange, and I'm happy to do it to cover the CI aspects of things.

It kinda shows, though, that we may want to have more kernels with
carefully thought out options to account for the different needs.  I'll try
to write up my thoughts on what they should be. I'm leaning to have an vm
vs embedded vs server split as well as a load-it-all vs compile-it-in
split, regardless of what they are named. And there's some base for all
three and then extensions from there (vm and embedded may be the same, and
server may be a superset of those two). So I'll take an action item to come
up with a concrete proposal for these things, along with good definitions
so people enhancing the system in the future have good guidance on wher to
do that.

Warner