Re: Anyone working on an automated install?

[Aleksej Saushev <asau%inbox.ru@localhost>]

Josh Branning <lovell.joshyyy%gmail.com@localhost> writes:

> Again, thanks everyone. I guess I have my answer. Nobody is really
> working on automating sysint using a configuration file, because a
> script does the job, even if it can get complicated for certain setups.
>
> Doing a quick search online, there have been a few projects intending to
> do what I'd like (but not specifically). All of these projects
> unfortunately seem to be abandoned, maybe because they were aiming to do
> too much at once.
>
> I've been looking at the source code for sysinst, but I cannot make no
> head-nor-tail of it - as previously said, my coding skills aren't great.

Coding skills are not critical here.

You really do not need to look at sysinst for this task,
this part seems to be the one you don't understand.
What sysinst does is interface, so that you could look yourself at what
partitions are and change them, what sets exist and which to install,
what services exist and which to run at startup.
Neither problem is not going be solved during unattended installation.
You know ahead what partitions you need, which sets to install, which
services to setup. You really do not need extra level of indirection
that is harder to program than shell.

> Though, even if I did succeed in modifying sysinst to take a
> configuration file, I'd still have to find a way to send it to you guys,
> so that it'll get accepted and of course, it'd have to be ported
> successfully to all other architectures.
>
> TBH, I think this one is a no-go, I'll probably end up having to install
> it on each machine manually, as there are some partitions on some
> machines that I want to keep.

Try to think more about your hypothetical sysinst automation while
you're doing that.

For instance, if you wish to keep some partitions on some machines,
then you have to do the following: detect those machines, conditionalize
code that creates partitions, create code that detects partition layout
unknown ahead and creates partitions in some non-trivial way.
The latter prompts support for conditionals and perhaps something more
(like non-local variables to pass information through stages).
This means that your hypothetical automation support should be almost
full-blown programming language. But we know such languages already,
we have few of them: sh, AWK, Lua. It makes more sense to use those
than to grow something new.

> Though, of course, it's a feature I'd definitely like to see in NetBSD
> in future.


-- 
HE CE3OH...