Confirm interpretation of any_of constraints is correct

[eecavanna]

Today, if a slot has the any_of constraint defined on it, refscan completely disregards the slot's range.

refscan/refscan/refscan.py

Lines 168 to 188 in d7d0edd

	# Determine the slot's "effective" range, by taking into account its `any_of` constraints (if defined).
	#
	# Note: The `any_of` constraints constrain the slot's "effective" range beyond that described by the
	# induced slot definition's `range` attribute. `SchemaView` does not seem to provide the result
	# of applying those additional constraints, so we do it manually here (if any are defined).
	#
	# Reference: https://github.com/orgs/linkml/discussions/2101#discussion-6625646
	#
	names_of_eligible_target_classes: list[str] = []
	if "any_of" in slot_definition and len(slot_definition.any_of) > 0: # use the `any_of` attribute
	for slot_expression in slot_definition.any_of:
	if slot_expression.range in schema_view.all_classes():
	own_and_descendant_class_names = schema_view.class_descendants(slot_expression.range)
	names_of_eligible_target_classes.extend(own_and_descendant_class_names)
	else: # use the `range` attribute
	if slot_definition.range not in schema_view.all_classes(): # if it's not a class name, abort
	continue
	else:
	# Get the specified class name and the names of all classes that inherit from it.
	own_and_descendant_class_names = schema_view.class_descendants(slot_definition.range)
	names_of_eligible_target_classes.extend(own_and_descendant_class_names)

I implemented it that way because I was under the impression that the any_of constraint was used to further constrain the slot's range, beyond the constraints already placed on the slot by the slot's range. In other words, I was under the impression any_of would always be a more specific set than range (in which case, I wondered why a schema author would bother specifying a range on that slot in the first place).

I'm creating this ticket to represent the task of getting an authoritative answer to the following question:

1. Is ignoring the range in the presence of any_of, consistent with how people proficient with LinkML would interpret a LinkML schema?

Related (upstream): linkml/linkml#2103

[sierra-moxon]

"used to further constrain the slot's range" is also how I would interpret the range. However, if a slot range is further constrained by slot_usage, the more general range is still valid, just redundant. That's why some of the methods in schema view return a list of ranges, e.g. ["NamedThing," "PlannedProcess"] -- both of these would be "true" -- but "PlannedProcess" is more specific (when we're validating, we don't want a range of "OntologyClass" to validate. This means we want a method that returns the "most specific slot range in the context of this class." The induced_slot method in LinkML can be this method, but it is currently not taking into account the anyOf or oneOf constructs, and there is a discussion about whether this should be the default instead of in another method. I think the current understanding is that this anyOf/oneOf functionality should start in the induced_slot method).

[eecavanna]

Thank you, @sierra-moxon, for writing that up and also helping me understand the situation in other conversations.

[eecavanna]

I moved the code that handles any_of into a helper function in an attempt to facilitate testing it. The helper function is named get_names_of_classes_in_effective_range_of_slot (link) and is in the file refscan/lib/helpers.py. After moving it to a helper function, I implemented a test targeting that function, in an attempt to demonstrate to people how refscan currently treats any_of (when it's present on a slot). The test is named test_get_names_of_classes_in_effective_range_of_slot (link) and is in the file tests/test_helpers.py.