update firstboot examples in docs and tests

[rebmit]

Update the firstboot examples in the documentation and tests to be compatible with systemd v258.

Closes #22

[rebmit]

Seems like we should also update the workflows.

[rebmit]

Can confirm the tests build locally, likely hitting NixOS/nixpkgs#399245

[nazarewk]

Today I had more success with plain config, than with symlink variants mentioned in the docs.

This works perfectly fine without any additional configuration:

            {
              file = "/etc/machine-id";
              inInitrd = true;
              configureParent = true;
            }

[rebmit]

Today I had more success with plain config, than with symlink variants mentioned in the docs.

Yes, it does work, and for most setups this is sufficient — this setup is also shown in https://nix-community.github.io/preservation/examples.html#simple.

However, in the first-boot section, the goal is not only to make machine-id preservation work, but to ensure compatibility with systemd's first-boot semantics.

[nazarewk]

yeah, it actually fails with a mountpoint until I add that ConditionFirstBoot = true;

[rebmit]

It might be worth adding an explicit option for the tmpfiles argument, rather than requiring users to write something like boot.initrd.systemd.tmpfiles.settings.preservation."/sysroot/persistent/etc/machine-id".f.argument.

That would be more ergonomic and less coupled to the implementation details of preservation, making it less likely to break if the implementation changes in the future.

[ThinkChaos]

I agree: having to patch the generated tmpfiles rules is awkward.

I see 2 possible improvements (inclusive or'ed):

1. add an option like initialContent to preserveAt.*.files
2. have an opinionated option to setup machine-id properly: preservation.machine-id.preserveAt :: path that sets up everything (a preserveAt.${config.preservation.machine-id.preserveAt}.files entry, and patches the tmpfiles rule & systemd-machine-id-commit.service)

I think 2 would be nicer in the long term: on top of being more ergonomic than 1 (less specific knowledge needed about machine-id, and hiding implementation details better), it avoids users copy-pasting magic configuration that needs to be updated from time to time, which is easy to miss both initially and when updated.

[ThinkChaos]

The updated documentation works for me, thank you!

[ThinkChaos]

I agree: having to patch the generated tmpfiles rules is awkward.

I see 2 possible improvements (inclusive or'ed):

1. add an option like initialContent to preserveAt.*.files
2. have an opinionated option to setup machine-id properly: preservation.machine-id.preserveAt :: path that sets up everything (a preserveAt.${config.preservation.machine-id.preserveAt}.files entry, and patches the tmpfiles rule & systemd-machine-id-commit.service)

I think 2 would be nicer in the long term: on top of being more ergonomic than 1 (less specific knowledge needed about machine-id, and hiding implementation details better), it avoids users copy-pasting magic configuration that needs to be updated from time to time, which is easy to miss both initially and when updated.