[Rules] Stratification

[afs]

Stratification is used to gives conditions on the order rules are evaluated so that the output is predicable. Negation depends on order of execution.
https://en.wikipedia.org/wiki/Datalog#Negation

The spec should define stratification and provide an algorithm to calculate the layer numbering.

This can be done for rule but if applied to rule elements we may be able to define evaluation for slighter more generally.

For example. NOT can defined to require conditions dependencies on triple patterns with variables that are used in the NOT pattern, and not for unrelated variables.

Recursion of unrelated triple patterns could then be allowed. This is equivalent to rewrite, replacing the rule with several rules. On the other hand, it might not be a good idea to over-complicate the conditions for something that can be written, albeit in a different form.

[ajnelson-nist]

@afs : Are you picturing stratification be based on numbered ordering attached to each rule, like done in SHACL-AF? Or, could there be some allowance for dependencies between rules?

[liviorobaldo]

Hello! Stratification should also apply to aggregates. However, it is still not clear to me how the grammar should be modified to allow rules that produce aggregate data.

[afs]

@afs : Are you picturing stratification be based on numbered ordering attached to each rule, like done in SHACL-AF? Or, could there be some allowance for dependencies between rules?

@ajnelson-nist No need to make the user provide the ordering - and imports would interfere with user-defined ordering.

https://w3c.github.io/data-shapes/shacl12-rules/#dfn-rule-element-dependency has some text about rules depending on other rules.

[ajnelson-nist]

@afs : Are you picturing stratification be based on numbered ordering attached to each rule, like done in SHACL-AF? Or, could there be some allowance for dependencies between rules?

@ajnelson-nist No need to make the user provide the ordering - and imports would interfere with user-defined ordering.

https://w3c.github.io/data-shapes/shacl12-rules/#dfn-rule-element-dependency has some text about rules depending on other rules.

Thank you. I feel the matter has been considered, at least. I see the floating TODO "formalize", which I suppose is partially this Issue?

Is Element-level dependency something that the rule engine is expected to detect without user action?

[afs]

Is Element-level dependency something that the rule engine is expected to detect without user action?

Yes.

I'm expecting that no user action is required for a well-formed SHACL rule set of the imports closure.

Some engines may not offer everything. Implementing only semi-positive (NOT can only depend on the input graph, not inferred triples) does not require stratification.