Add some sort of mechanism for resolving manager for a given model type

[pederhan]

This kind of speaks for itself. It more or less re-implements the path/id resolution implemented on the managers, but does it in a very brittle and brute-forcey manner:

mreg-api/mreg_api/events.py

Lines 65 to 122 in e240515

	@classmethod
	def new(cls, obj: MregModel) -> Self:
	"""Create a reference to an API object.
	Uses the object's endpoint to determine the ID field to use.
	Cannot fail. Logs and defaults to str(obj) for value on failure.
	"""
	from mreg_api.models.models import CNAME
	from mreg_api.models.models import Atom
	from mreg_api.models.models import ForwardZone
	from mreg_api.models.models import ForwardZoneDelegation
	from mreg_api.models.models import HInfo
	from mreg_api.models.models import Host
	from mreg_api.models.models import HostGroup
	from mreg_api.models.models import Location
	from mreg_api.models.models import NameServer
	from mreg_api.models.models import Network
	from mreg_api.models.models import ReverseZone
	from mreg_api.models.models import ReverseZoneDelegation
	from mreg_api.models.models import Role
	# TODO: refactor when we have a better way to get the external ID
	# field from an object without needing the endpoint.
	MODEL_FIELDS = {
	"name": (
	Host,
	HostGroup,
	CNAME,
	ForwardZone,
	ReverseZone,
	ForwardZoneDelegation,
	ReverseZoneDelegation,
	Role,
	Atom,
	NameServer,
	),
	"network": (Network,),
	"host": (HInfo, Location),
	}
	for field_name, models in MODEL_FIELDS.items():
	if isinstance(obj, models):
	id_value = getattr(obj, field_name)
	id_field = field_name
	break
	else:
	try:
	id_value = str(obj.id) # pyright: ignore[reportAttributeAccessIssue, reportUnknownMemberType, reportUnknownArgumentType]
	id_field = "id"
	except AttributeError:
	id_value = str(obj)
	id_field = "id"
	try:
	return cls(type=obj.__class__.__name__, value=str(id_value), field=id_field)
	except Exception:
	logger.exception("Failed to instantiate ObjectRef from %s", obj)
	return cls(type=obj.__class__.__name__, value=str(obj))

It also mentions using the object's endpoint, which is outdated and no longer applies.

[pederhan]

I had a thought about this that may be insanely silly and dumb, but if we had an interface for resolving managers by model, we could also do this:

client = MregClient(...)

# We get some sort of MREG object from somewhere:
host = function_that_returns_hosts()

client.manage(host).rename(host, "newname")

The main benefit would be things that operate on multiple resource types ambiguously, such as roles and atoms:

role_or_atom = get_policy("blabla")

client.manage(role_or_atom).set_description(role_or_atom, "new description!")

In an ideal world, we would bind the model we pass in to the manager somehow, but that veers into the realm of complex and overly dynamic code that is hard to reason about and maintain:

label = get_label("blabla")

client.manage(label).set_description("new description!")

I don't even want to begin to think about having to add generics that bind to existing methods, and trying to create an interface that satisfies every single manager generically.

Implementation

I see it as something like this:

class MregClient:
    # ...
    @overload
    def manage(obj: Atom) -> AtomManager: ...
    def manage(obj: MregModelT) -> ResourceManager[MregModelT]: ...

I don't know if this type of overload is actually valid, though.