Add a Constraint Component for uniqueness among values reached by a SHACL Path

[mgberg]

I propose the addition of a (Boolean-valued?) constraint component that would create a validation error for any repeated value among those obtained from evaluating a property path. This would presumably only be meaningful if the path was not just a single predicate or an inverse path of a single predicate. This could be used to help detect irregularities in tree-like structures.

For example, consider a genealogy graph. Suppose one wanted to create a constraint that verified a person's parents did not share a parent (i.e., that a person does not have a grandparent via more than one way). Also, it would make sense to add this constraint component to the property shape for the path ex:father|ex:mother given as an example in the current SHACL spec to verify a person was not both the mother and father of a person.

I realize that this behavior can be easily accomplished via a SPARQL Constraint or Constraint Component, but it would presumably be more convenient and efficient if it was a built-in capability. Also, I suppose this would only work if the SHACL validator returned the full list of values reached when evaluating a property path instead of the set of unique values- I'm not sure how specific implementations work in this respect.

[HolgerKnublauch]

This may be difficult to implement. SHACL engines will usually rely on existing SPARQL property path engines, which will already eliminate duplicates before they can be processed.

[mgberg]

Yeah I figured that was likely.

[TallTed]

[@mgberg] For example, consider a genealogy graph. Suppose one wanted to create a constraint that verified a person's parents did not share a parent (i.e., that a person does not have a grandparent via more than one way).

It seems worth cautioning you (and later readers) against making assumptions which have demonstrable contradictions in real world data. Analysis of several sub-sects of the Mormons, for instance, will reveal a great many persons who have one grandparent (or great-grandparent, or further) via multiple paths.

[mgberg]

[@mgberg] For example, consider a genealogy graph. Suppose one wanted to create a constraint that verified a person's parents did not share a parent (i.e., that a person does not have a grandparent via more than one way).

It seems worth cautioning you (and later readers) against making assumptions which have demonstrable contradictions in real world data. Analysis of several sub-sects of the Mormons, for instance, will reveal a great many persons who have one grandparent (or great-grandparent, or further) via multiple paths.

Well sure, I'm just using this as an example to communicate the idea.

[HolgerKnublauch]

I will close this (old) issue as this is too difficult to implement given that SHACL paths rely on SPARQL paths which already eliminate duplicates.

@mgberg please reopen if you disagree with this assessment.