Only accept named nodes as predicates for metadata

[joachimvh]

Following ae06e99 I think we can remove the toCachedNamedNode function completely. Only the RepresentationMetadata uses it but in practice I think it already receives named nodes most of the time, so we can also change the interface there and always do the call with the .terms uris then.

[RubenVerborgh]

Check; I was wondering about that when I went through the code indeed. The only exception are the shorthands though, what do we do with contentType?

[joachimvh]

Would have to look at how to specifically do the change, but we can potentially just have an exception for those in the RepresentationMetadata file, but there is no need to have this function publicly available.

[constraintAutomaton]

It seems like every usage of addQuad use a NamedNode as parameters for predicate (which is the only element that called cachedNamedNodes ). Can we restrict the type to only NamedNode.

Every usage of remove uses RDF.type (of type string) which could be replace by a constant of the type NamedNode.

add in LinkMetadataGenerator uses a string. setOveride and getAll use also a string. We could move cachedNamedNodes inside the RepresentationMetadata and make the function private.

[MisterTimn]

As a start I went ahead and moved toCachedNamedNode into RepresentationMetadata and getting to know the relevant classes.

Is the desired goal to change all calls to effectively use a NamedNode or is this impossible/cumbersome in some cases? Right now I am going through the remove related function calls I think they can all indeed use it, only in tests are strings still getting used (as far as I can tell).

Any ideas where to best start?

[joachimvh]

As a start I went ahead and moved toCachedNamedNode into RepresentationMetadata and getting to know the relevant classes.

I think the shorthands object can even be removed. I'm pretty sure its contents are never used in practice.

Is the desired goal to change all calls to effectively use a NamedNode or is this impossible/cumbersome in some cases?

The first. This should always be possible. If some classes take a string as input (which will be the case for some constructors probably), it would then be their responsibility to convert the string to a named node.

The one issue/exception is the overrides object in the RepresentationMetadata constructor since the keys have to be strings there. toCachedNamedNode will have to stay for that case.

Any ideas where to best start?

Remove string as an acceptable value for predicates (and subjects) in all RepresentationMetadata functions and see what breaks. My guess is that the number of cases should be limited, except perhaps in the tests.

[MisterTimn]

Remove string as an acceptable value for predicates (and subjects) in all RepresentationMetadata functions and see what breaks. My guess is that the number of cases should be limited, except perhaps in the tests.

I have already done so for remove operations and all seems to be working (after changing some calls with strings to NamedNodes), I'll open a draft PR asap and continue work on this there.

Since I am changing function parameter types, this will be a semver.major probably?

[joachimvh]

Since I am changing function parameter types, this will be a semver.major probably?

Yes, definitely.

[joachimvh]

Done in #1261