Inherited slots and slot_usage

A class can get the metaslots of a slot from several places: its own slot_usage, the slot_usage of its parent classes and mixins, and the slot definition. This page shows how LinkML-Scala combines two kinds of them, where the LinkML specification, LinkML-Python 1.12.1 and LinkML-Scala do not all agree:

• Boolean metaslots, like required and multivalued, which are true or false. LinkML-Scala follows the specification here, and LinkML-Python works differently.
• Boolean expressions: any_of, all_of, exactly_one_of and none_of, which are lists of conditions. LinkML-Scala works differently from both the specification and LinkML-Python here.

Boolean metaslots

The LinkML specification combines boolean metaslots, like required and multivalued, with OR when it derives a slot. So a slot_usage can make a slot required or multivalued, but setting either to false has no effect when the slot or a parent class already sets it to true. LinkML-Scala does the same. LinkML-Python 1.12.1 uses the value from the closest slot_usage instead.

classes:
  Base:
    slots:
      - name
      - tags
  Child:
    is_a: Base
    slot_usage:
      name:
        required: false
      tags:
        multivalued: false
slots:
  name:
    required: true
  tags:
    multivalued: true

Slot of Child	LinkML-Python 1.12.1	LinkML-Scala
name	Optional	Required
tags	Single-valued	Multivalued

So {} is a valid Child for linkml-validate, but not for the JSON Schema, SHACL, Pydantic or TypeScript that LinkML-Scala generates. To make a slot optional or single-valued in some classes, leave it optional or single-valued in the slot definition, and turn it on in the classes that need it.

Boolean expressions

LinkML-Scala combines the any_of, all_of, exactly_one_of and none_of lists of a slot so that a value of the class is also a valid value of each of its parent classes and mixins (Liskov substitution).

Expression	LinkML specification	LinkML-Python 1.12.1	LinkML-Scala
all_of	Joins the lists	Keeps only the most specific list	Joins the lists
none_of	Joins the lists	Keeps only the most specific list	Joins the lists
any_of	Joins the lists	Keeps only the most specific list	Keeps the most specific list, and requires the other lists too (see below)
exactly_one_of	Joins the lists	Keeps only the most specific list	Keeps the most specific list, and requires the other lists too (see below)

Joining the lists is correct for all_of and none_of: each member is one more check. For any_of and exactly_one_of, neither joining nor keeping one list is correct:

• Joining any_of lists allows more values. A subclass can’t narrow down the ranges.
• Keeping only the most specific any_of list allows a subclass to add ranges that a parent class doesn’t allow.
• An exactly_one_of with fewer members can allow more values: a value that matches two inherited members is not valid for the parent class, but it is valid for a list with only one of these members.

So LinkML-Scala keeps the most specific list, and adds each other list to all_of, as a member with this list as its any_of or exactly_one_of. An any_of list is not added if it has all the members of the most specific list, because then it allows all the values that the most specific list allows.

classes:
  Base:
    slots:
      - value
  Child:
    is_a: Base
    slot_usage:
      value:
        any_of:
          - range: integer
          - range: boolean

slots:
  value:
    any_of:
      - range: integer
      - range: string

Value	LinkML specification (any_of of integer, string, boolean)	LinkML-Python 1.12.1 (any_of of integer, boolean)	LinkML-Scala (both any_of lists)
42	Valid	Valid	Valid
“abc”	Valid	Invalid	Invalid
true	Valid	Valid, but not valid for Base	Invalid

For Child.value, LinkML-Scala uses this derived slot:

any_of:
  - range: integer
  - range: boolean
all_of:
  - any_of:
      - range: integer
      - range: string

The SHACL generator emits all of it. The OWL generator gives each class its own any_of ranges; OWL subclassing requires the ranges of the parent classes too. The JSON Schema and Pydantic generators do not support boolean expressions yet.