Boolean expressions and linkml:Any

[ialarmedalien]

The LinkML docs say:

Boolean constraints

The following LinkML constructs can be used to express boolean constraints:
any_of
all_of
none_of
exactly_one_of

These can be applied at the class or slot level.

from https://linkml.io/linkml/schemas/advanced.html#boolean-constraints

and

Unions as ranges

any_of can be used to express that a range must satisfy any of a set of ranges.
[…]
Note that the range of vital_status is declared as Any, which is further constrained by the any_of expression.
Currently, it is important to always have a range declaration (even if it is Any), because LinkML constraint semantics are monotonic (i.e. new constraints can be specified but existing ones cannot be overridden - see slots for more on this).

from https://linkml.io/linkml/schemas/advanced.html#unions-as-ranges

However, personinfo.yaml (e.g. in linkml/tests/test_transformers/input/personinfo.yaml), everyone’s favourite test staple, contains the following:

> slots:
>   type:
> [...]
>     slot_usage:
>       type:
>         any_of:
>           - range: FamilialRelationshipType
>           - range: NonFamilialRelationshipType
>         required: true

with no linkml:Any anywhere to be seen.

When is linkml:Any needed?

• Only for range declarations?
• Any time you use one of the boolean constraints on a class or slot?
• Depends on what package(s) you’re using and whether or not it got implemented?

[cmungall]

Yes, the docs need to be clearer on this. Strictly speaking, assuming that the default_range is string, and type inherits it, then this is non-monotonic.

This should be avoided for two reasons:

1. clarity
2. the json-schema generator will make unsatisfiable subschemas

Reason 2 may not be super-apparent. It is formally encoded in the compliance tests

linkml/tests/test_compliance/test_boolean_slot_compliance.py

Lines 148 to 152 in 5863045

	if framework == JSON_SCHEMA:
	# if use_default_range and not is_valid:
	# https://github.com/linkml/linkml/issues/1483
	if is_valid and use_default_range:
	expected_behavior = ValidationBehavior.FALSE_POSITIVE

Which still lack a nice report

Now, it could be argued that the expectation is for any_of to work more like python or json-schema rather than OWL, and I think this is perfectly reasonable. If there is consensus, then we can codify the following:

• default_range is only applied if there is no explicit range of a slot AND the slot is not used in an any_of or exactly_one_of context

Formally, we would treat the default_range application as a pre-processing step, so the full model semantics is still monotonic.

Comments on the above welcome

Note that regardless, the recommendation is still to be explicit for now to avoid nasty json-schema surprises (either via linkml:Any, or in the case of personinfo, since it's a union of two classes, using a common superclass as the range preserves monotonicity)

[ialarmedalien]

I've implemented fairly comprehensive tests for range retrieval based on the currently implementation of the range and slot_range_as_union functions in SchemaView.py:

linkml/linkml-runtime#454

re: default_ranges - I had some thoughts about them here:

SchemaView slot ranges, default ranges, and monotonicity

I would like to add in an induced_range-type function that just returns the useful information -- discards linkml:Any or None if there is any_of or exactly_one_of specified for the range, or returns the default range if there is no range specified for the slot.

The endgame here is having a function that can work out the possible types for a slot, which would be used by generators that map linkml to other formats.

[Enro]

For the sake of record keeping, I link to a related issue (that I misfiled in cookiecutter repo): linkml/linkml-project-cookiecutter#144

[ialarmedalien]

Do you have a class with name Any and the class URI linkml:Any in your schema?

We really need a minimal, complete example to be able to troubleshoot further.