Use LintedData for checking of generated RDF files

[jmkeil]

Hi. I develop the tool LintedData that can be used to automatically check RDF files in a CI pipeline for common problems, like the xsd:float precision problem (#77). @mgrub indicated, that you might be interested to use it for the SI reference point RDF representation. If you want to give it a try, we could setup a pipeline job that checks the generated RDF files. The documentation contains a GitHub Actions job to check RDF files with LintedData. What do you think about it?

[Crparedes]

Thank you.
Your tool is very useful and well documented.

It could be convenient to discuss the current issues before includding linteddata in our GitHub workflow, but I appreciate the kind proposal to set up a pipeline job.

Running it from the CLI flags some issues:

• NamespaceHijackingCheck

  • (Hundreds) mostly from our prefixes with https://si-digital-framework.org/. However, we do own the domain.
  • Some come from prefixes of GitHub's urls that used as part of PROV: objects and subjects that are being taken as RDF resources instead of literals. I remember @laurenth974 mentioned an issue in our graphs related to this.

• OntologyNotAvailableOnWebCheck

  • being flagged for ontologies that do have landing pages: si:, units:, constants:, prefixes:, quantities:.
  • and of course for ontologies that do not at the moment cipm:, cctf:, cgpm:, bodies

• UntypedClassCheck and UntypedPropertyCheck

  • It is flagged for many predicates of the A-Boxes, but they are defined in the T-Boxes whose prefixes are declared. Is it that we are missing explicitly owl:importing them in the A-Boxes graphs?
  • It is flagged for some external classes (e.g. skos:prefLabel, dcterms:license)
  • It is also flagged in the T-Box of si.ttl for some of our classes that are explicitly defined (e.g. si:UnitProduct). Those are not being flagged when linting the single full serialization file.
  • Some of our class and properties that are being flagged are indeed not being declared (e.g. si:SIDecisionScope) or being declared as skos:concept (i.e. rb:Decision and 8 more in rb:)

• VersionIriExistenceCheck for all graphs

• IriTooLongCheck for some timestamped prov:activities

• LeadingOrTrailingSpaceCheck for many strings

• MissingDescriptionCheck for many owl:Classes

• Some MissingDomainCheck and MissingRangeCheck for some owl:ObjectPropertys

• MultipleLabelsCheck for 'rb:hasEvent' (an @fr tag got swapped by a @en)

• MissingLicenseForOntologyCheck for cipm:, cctf:, cgpm:, bodies

• DifferentNamingConventionsCheck, we are not using PascalCase for the class si:nonSIUnit. Is this important?

Aditionally, for the single full serialization file:

• Several MultipleDomainsCheck and a MultipleRangesCheck, probably because the si: T-Box is duplicated in the bodies: T-box and some domains and ranges that use blank nodes with owl:oneOf are being added independently in the full serialization
• Several MultipleOntologyDeclarationsCheck

The reports are below:

• IndividualFiles_LintedData.md
• FullSerializedSingleFile_LintedData.md

[jmkeil]

• NamespaceHijackingCheck
  • (Hundreds) mostly from our prefixes with https://si-digital-framework.org/. However, we do own the domain.
  • Some come from prefixes of GitHub's urls that used as part of PROV: objects and subjects that are being taken as RDF resources instead of literals. I remember @laurenth974 mentioned an issue in our graphs related to this.

Determining the proper own namespace turned out to be quiet complicated. This check is currently under reconsideration (see LintedData#205, LintedData#206). Maybe I should have disabled it by default until this is solved.

• OntologyNotAvailableOnWebCheck
  • being flagged for ontologies that do have landing pages: si:, units:, constants:, prefixes:, quantities:.

I had a look into this. The reason for failure is, that the RDF files (after content negotiation) are returned with the header Content-Type: application/octet-stream. One could argue in both directions, whether this should be flagged here or not, as it is not a fault of the ontology itself. But it clearly is a misconfiguration of the server and might break client software to properly consume the RDF.

• UntypedClassCheck and UntypedPropertyCheck
  • It is flagged for many predicates of the A-Boxes, but they are defined in the T-Boxes whose prefixes are declared. Is it that we are missing explicitly owl:importing them in the A-Boxes graphs?

Yes. The prefix declaration is solely a instruction to the parser how to interpret the IRIs. Reuse of other ontologies must be declared with owl:imports.

• UntypedClassCheck and UntypedPropertyCheck
  • It is also flagged in the T-Box of si.ttl for some of our classes that are explicitly defined (e.g. si:UnitProduct). Those are not being flagged when linting the single full serialization file.

Are you referring to line 716? This is flagged for https://si-digital-framework.org/constants/ (line 708).

• IriTooLongCheck for some timestamped prov:activities

There is no hard role for a maximum length. However, longer IRIs can become quiet unhandy in some situations and should therefore not be minted thoughtlessly. We selected a default max length of 36 characters, which is already quiet long for the readability of SPARQL queries, but does not flag UUIDs. However, if you decide, a higher length is necessary here, this is completely fine.

• DifferentNamingConventionsCheck, we are not using PascalCase for the class si:nonSIUnit. Is this important?

The handling of abbreviations in PascalCase is treated differently. What is in favor of si:nonSiUnit instead of si:nonSIUnit is the slightly increased readability, as the word endings can be spotted on the lexical level of perception instead of the syntactical level of perception. On the other hand, it might be unexpected. What is most important is consistent handling over all names. si:nonSiUnit could be set as an ignored exception in the LintedData configuration. Implementing a generic rule for abbreviations in LintedData would require a validation on the semantic level, instead of the lexical level and might hide typos, of unintended upper case letters in a word.