* [] is now [ ]
* {} is now { }
* commas in arglists go at end of line not beginning
In short, I ran the whole thing through nixfmt-rfc-style but only
accepted about 30% of its changes. I might grow accustomed to more
of it over time
When s6-rc stops a service, it also stops everything that
depends on it. but when it starts a service it starts only
that service, so we have to go through the other services
depending on it and figure out if they should be started too.
- uncontrolled services that are not dependent on a controlled service
- controllers
- _not_ controlled services or any other service that depends on one
Instead of treating the trigger as the "main" service and the
triggered service as subsidary, now we treat the triggered
service as the service and the trigger as "subsidary". This
needs some special handling when we work out which services
go in the default bundle, but it works better for declaring
dependencies on triggered services because it means the
dependency runs after the triggered service comes up, not
just when the watcher-for-events starts
Good: this means it's not hanging holding the s6 dataase lock.
Bad: it's the ugliest implementation and doesn't deserve to be preserved
(tbf the ugliness is not new)
We call s6-rc -u -p default to restart/start the base services
on a rebuild, otherwise services that are only in the new
configuration won't come up. However, this stops any service
started by a trigger. So, workaround is to restart the trigger
service and expect it to restart the services it manages if they're
needed
