The Plan, phase 1

[emilazy]

It’s finally time.

This PR eliminates user activation, addressing #96 by following the first phase of the plan from #1016 (comment), fulfilling a goal I started drafting patches for in 2023 before I was even a nix-darwin maintainer.

I strongly recommend reviewing this one commit‐by‐commit. See the commit messages for much more detail and for things I’m still uncertain about. I’ve been running this on my own system for a while now, but it could definitely use a staged roll‐out before we actually hit the merge button. We should also probably also reach out to authors of third‐party modules that will be affected once it’s mostly baked, though at least anything that breaks should mostly break loudly. (Adding text to our existing activationScripts.{checks,userDefaults,userLaunchd,homebrew}.text definitions is an exception – those will run as root now. But I can’t think of a particularly graceful way to handle that case, and anyway it’s not exactly something we fully support in the first place.)

Closes: #96

[spikespaz]

Can we have an option to disable primary user, so that dogfooding finds problems with not having a primary user sooner?

[spikespaz]

I just realized the default is null, so never mind.

[emilazy]

Yeah, leaving it at the default will ensure your system doesn’t depend on any primary user and give a hopefully‐helpful error message if you set anything that requires it.

[spikespaz]

in practice, it may be useful for us to
nest this into users.users.*.homebrew instead, at the expense of
being an unsupported setup if used to its full potential. Since
that would be a breaking change to the inteface anyway, I think
adding homebrew.user for now is acceptable.

I understand the motive here, but are you sure it's not better to make more breaking changes in one go? I'm worried that going through another great migration would agitate some.

[emilazy]

I previously thought that too – that we should figure out the correct interfaces and write good migration documentation and tools and get all the breaking changes over with in one go – but the end result was that this work stalled out for years because there’s a lot of API design and implementation work to get things to a good state: we need to decide which things should be global or per‐user; we need to decide what should stay in nix-darwin and what should be deduplicated with Home Manager; for things like our system.defaults options, they’re currently not fully declarative in the way that Nix‐managed options should be, so if we’re breaking everyone’s configuration then we ought to make that important breaking change at the same time; and so on.

If we tried to do it all at once, this work would have spent even longer in development hell and be unreviewably big. For the longest time i wasn’t sure how to resolve that, but after discussing on Matrix I realized that introducing a system.primaryUser option, though strange from an interface design perspective, would allow us to minimize the breaking changes in the first phase: the only required change for the vast majority of users should be having to set system.primaryUser = "username"; in their configuration, which I think is an acceptable level of fuss for the improvement in internal architecture and coherent model it gives us, and allows people who want a truly user‐agnostic system manager to opt for that world now at the expense of reduced functionality. That takes the pressure off any major interface renovations and lets us work on them independently and batch them up as we see fit.

In the case of Homebrew, homebrew.user defaults to the primary user, so it’s purely a bonus for people who want to use Homebrew without setting system.primaryUser (or to use a different user entirely for Homebrew, I suppose).