Resolving url paths about rdf-ext

[ktk]

Hi Nicola,

Thanks for the feedback, credits go mainly to @bergos :)

I don't think path resolving is currently implemented, question is is this defined behavior in RDF? And this question could probably be answered by @retog

[bergos]

How did you parse the graph? For example the JSON-LD parser would resolve the relative path in the expand step. But like @ktk mentioned, first we need to know if there is a definition? Maybe @RubenVerborgh has an answer to this question.

[RubenVerborgh]

The conclusive answer is in the RDF 1.1 specification:

Relative IRIs must be resolved against a base IRI to make them absolute.

[RubenVerborgh]

And the fact that this doesn't happen seems to be an N3.js bug: rdfjs/N3.js#51

[bergos]

@nicola btw. have you seen Clownface for graph traversing? Here an example how you would fetch the preferences file:

var cf = rdf.cf.Store(new rdf.promise.Store(new rdf.LdpStore()));

cf.node('https://nicola.databox.me/profile/card#me')
  .then(function (cf) {
    console.log(cf.out('http://www.w3.org/ns/pim/space#preferencesFile').literal().shift());
  });

[nicola]

Of course I have seen it, it is very cool (I think that should be decoupled from this library as well!!).
I was going to build a similar system and I found it - I will let you know what I think after using it for a bit

[retog]

@RubenVerborgh it doesn't exactly not happen, the IRI Ref is not kept relative which would be against the specs (except the LDP spec, but that's another story). The thing is that the IRI is not made absolute the right way (i.e. according to the IRI RFC 3987 and RFC 3986): instead of https://nicola.databox.me/profile/../Preferences/prefs it should be https://nicola.databox.me/Preferences/prefs.

[nicola]

I hope that is the case, or it would be a nightmare for me to do that simple operation

[RubenVerborgh]

@retog Yes, that's what I mean: the resolving happens by just sticking both parts together, whereas (only) leading sequences of ../ in the relative URL should be correctly removed when resolving. It's a bug in N3.js, which I will fix soon.

[nicola]

I just got a funny: https://nicola.databox.me/profile/../Preferences/../

[RubenVerborgh]

Fixed in rdfjs/N3.js@1a0dc71 without performance impact on other types of IRIs. Will leave people some time to verify the correctness of the solution and then do a 0.4.4 release that includes the fix.

[RubenVerborgh]

@retog I actually found the RDF spec to be ambiguous about IRI resolution: https://lists.w3.org/Archives/Public/public-rdf-comments/2015Aug/0000.html. Until an answer is found, I don't feel comfortable landing a release.

[ktk]

@RubenVerborgh if I get that correctly the discussion is pretty much set right? Any ETA for new release?

[RubenVerborgh]

Not set, no, still actively discussing: https://lists.w3.org/Archives/Public/public-rdf-comments/2015Sep/
ETA: could still be a week.

[ktk]

Ah didn't notice that I only saw postings in August in the web interface of the list, thanks for the link.

[retog]

@RubenVerborgh iiuc the spec ambiguity only applies when the Turtle document contains non-normalized non-relative IRIs (i.e. the document contains https://nicola.databox.me/profile/../Preferences/prefs as base IRI or as a IRI for a resource). I seems however clear that when a relative IRI with trailing [./|../]* and no other ./ or ../ has to be resolved against a normal based IRI this has to result in a normal resource IRI. So in the example in this issue the resource name should be https://nicola.databox.me/Preferences/prefs rather than https://nicola.databox.me/profile/../Preferences/prefs under any plausible interpretation of the spec.

[RubenVerborgh]

@retog Yes indeed, there's no unclarity about this specific case; consequently, the fix for this case was already made 2 weeks ago. However, I'm still waiting for a confirmation for other cases, so the next release of N3.js does IRI resolution correct in all cases, not just this one.

[RubenVerborgh]

Released N3.js 0.4.4 with correct implementation of RFC 3986, which includes the cases in this thread.