Literal->getDatatype() always returns string? Why is it different from getLang?

[k00ni]

Literal interface says

https://github.com/sweetrdf/rdfInterface/blob/48d8a66bb7952e30e7382dd2a57cb3ce8cccebeb/src/rdfInterface/Literal.php#L45-L47

Does it mean that there is a default value for getDatatype? Like an empty string?

IMHO regarding the PHP side there is no difference between getDatatype and getLang(), both might be null at the beginning. But their definition differs for some reason. Can you please explain why?

[zozlak]

On the RDF side the difference is a literal without a lang tag is always different from one having a lang tag while lack of datatype means the datatype is assumed to be xsd:string (so "foo" and "foo"^^http://www.w3.org/2001/XMLSchema#string are equal; see https://www.w3.org/TR/rdf11-concepts/#section-Graph-Literal for a reference).

This creates an ambiguity because xsd:string datatype has two valid representations - lack of a datatype (in PHP context we would say null) and explicit xsd:string. To deal with this ambiguity I decided to explicitly forbid the null representation (but I also admit it should be precisely described in the method comment).

I must also admit this convention is not followed by Literal class constructor and DataFactory::literal() method. What is worse, DataFactory::literal() has different default (null) than the Literal class constructor (xsd:string). It's clear it must be synced.

I see three options to make it in sync:

• allow null for datatype everywhere (and state it's up to API user to deal with possible ambiguity between null and xsd:string datatype)
• disallow null everywhere (and explain in the comments it's to avoid the ambiguity)
• allow null in DataFactory::literal() and Literal class constructor for the sake of user convenience (just set the default consistently to xsd:string) but (also for sake of user convenience) ban null in the Literal:getDatatype() method and explain in the comment the method should return xsd:string (of course as a full URL) both when xsd:string datatype is explicit and implicit.

Personally I'm the most in favor of the last one and I dislike the first one the most. And what do you think?

[zozlak]

By the way to make such assumptions clear and easy to verify (and to save copy-pasting of tests code in implementations) I started to build a tests suite for the rdfInterface compatibility in the tests directory. The simpleRdf already makes use of them.

[k00ni]

I am working on my in-memory store implementation currently, which might be a good practice range for rdfInterface. When taking the view as an outsider, some things are not clear enough (yet), like how to use certain things or what you have to provide sometimes.

I follow your arguments here and support the third option too. Let us continue shaping the library with external projects and make it viable and fun to use. Might worth some iterations, but we are on a good way IMHO.

[zozlak]

When taking the view as an outsider, some things are not clear enough (yet), like how to use certain things or what you have to provide sometimes.

Sure thing. I didn't put much effort into documenting things so far as I would say it's still in quite early development stage. Please don't hesitate to ask when you run into such issues. They may indicate lack of documentation but also design flaws and it's important to discuss them.

As for now my plan is to:

• Extend the rdfInterface test suite so it covers all corner cases which came to my mind.
• Make sure simpleRdf and quickRdf pass all those tests
• Extend quickRdf tests so they assure other implementations (read simpleRdf) interoperability.
  • Make quickRdf pass all those tests.
  • Make sure performance isn't harmed.
• Try to port something I'm using in my everyday life to quickRdf.
  • This will for sure result in another library providing a factory of useful QuadTemplate.

Only once this is done and I have a feeling rdfInterface works well focus on documenting it well.

First two point of the list above are likely to be ready until the end of March. Can't tell about the others.

I follow your arguments here and support the third option too. Let us continue shaping the library with external projects and make it viable and fun to use. Might worth some iterations, but we are on a good way IMHO.

Ok, I will make the adjustments.

[zozlak]

By the way, I just read https://www.w3.org/TR/rdf11-concepts/#section-Graph-Literal carefully and there will be a change in the expected Literal::getDatatype() behavior:

• if literal's datatype is explicitely specified, it should be returned
• if literal's datatype is not explicitely set and the literal has a lang tag, http://www.w3.org/1999/02/22-rdf-syntax-ns#langString should be returned
• if literal's dataset is not explicitely set and the literal has no lang tag, http://www.w3.org/2001/XMLSchema#string should be returned

[k00ni]

Good hint!