>> +(define shared-network-service-type
>> +  (service-type (name 'shared-network)
>> +                (extensions (list (service-extension etc-service-type identity)))
>> +                (compose concatenate)
>> +                (extend append)
>> +                (default-value '())))
>
> I’d encourage you to add a ‘description’ field as well.  :-)

Sure, will do.

>   1. ‘service-type-name’ exists for debugging purposes, and I think we
>      shouldn’t rely on it at all in our code.  Instead, we should
>      compare service types by identity, as in:
>
>        (eq? (service-kind service) foo-service-type)

Sure, will do.

>   2. The notion of “shared network” is very much a container (or VM)
>      thing, so somehow it still doesn’t feel right to me that (gnu
>      system) has to be aware of these special cases.
>
> I think the ‘host-database-service-type’ wouldn’t have this problem, but
> maybe it has other issues.  I guess this needs more experimentation,
> sorry for not coming up with clearer ideas!

If these services (the shared-network service, the hosts-database
service or indeed any other service) had access to the operating-system
object `os', then they would be able to operate independently without
having to be extended by `essential-services'. Is this possible somehow?
Is it a good idea to give services access to the os fields?