Missing predicate between `solid:TypeIndex` and `solid:TypeRegistration`

[srosset81]

Hello,

I would like to implement Type-Indexes for the ActivityPods project. (For information, in other projects with public data, I have used VOiD endpoints and it also worked well, but it cannot be used for Pods because it uses a single /.well-known/void URL. Here's an example)

However, it seems a semantic link is missing on the current spec:

<>
  a solid:TypeIndex ;
  a solid:ListedDocument.
                    
<#ab09fd> a solid:TypeRegistration;
  solid:forClass vcard:AddressBook;
  solid:instance </public/contacts/myPublicAddressBook.ttl>.
                    
<#bq1r5e> a solid:TypeRegistration;
  solid:forClass bk:Bookmark;
  solid:instanceContainer </public/myBookmarks/>.

There is no predicate linking the solid:TypeIndex resource with the various solid:TypeRegistration resources. Since for ActivityPods we use a triple store for storage (and not the filesystem), each resource must have its own URI. And in the example above, we have no way to find the solid:TypeRegistration resources by reading the solid:TypeIndex resource.

A simple predicate like solid:hasTypeRegistration would solve this problem.

Thanks for reading.

[jeff-zucker]

The answer to this question is wider than the type indexes. Suppose I have a document of type schema:oOfferCatalog and it has a bunch of schema:Offers ... do we need to explicitly say, these offers are in this catalog? My $0.02 - from a writing point of view, yes, it's good practice to be explicit. From a reading point of view, no, it's good practice to accept things that make sense. OTOH, this might be entirely too loose an interpretation.

[lecoqlibre]

In my understanding, all the triples are linked by the graph (document) they belongs to (named graph).

@srosset81 If you are storing the fragment part in the TripleStore, a workaround could be to return all the triples who share the same RDF subject without the fragment?

[jeff-zucker]

@lecoqlibre, I wouldn't tie it to the document/graph - there could be statements with many types of subjects. But if a document says "I am a catalog of things of type X", then I would think it is safe to assume that statements whose subject is of type X are in the catalog. Things with other types of subjects would not be.

[lecoqlibre]

@jeff-zucker yes absolutely, the workaround won't work for statements with other subjects but at least provide a way to support a "basic" TypeIndex with a TripleStore. If we assume that a TypeIndex resource could contain other things, then adding a condition on the type will allow to retrieve only the registrations.

I wouldn't tie it to the document/graph. [...] Things with other types of subjects would not be.

Should we use the term "dataset" instead? So the other types of subjects would be part of the dataset but not part of the document?

I avoid using the term "dataset" because it collides with the concept of RDF dataset which is a collection of RDF graphs.

[jeff-zucker]

I am in favor of the proposal to have a specific predicate. I wouldn't consider an app broken that was forgiving and assumed that link, but it is better to have the link explicit.

[lecoqlibre]

OK just to be clear @jeff-zucker you would be in favor to add a predicate saying that the TypeIndex document contains things of type solid:TypeRegistration? So you are not saying that the TypeIndex should link to all of its registrations, right?

Because I think @srosset81 was suggesting to add a solid:hasTypeRegistration linking to all the solid:TypeRegistration it has like:

<>
  a solid:TypeIndex ;
  a solid:ListedDocument;
  ex:hasType solid:TypeRegistration; # jeff-zucker's proposal?
  solid:hasTypeRegistration <#ab09fd>, <#bq1r5e>. # srosset's proposal?
                    
<#ab09fd> a solid:TypeRegistration;
  solid:forClass vcard:AddressBook;
  solid:instance </public/contacts/myPublicAddressBook.ttl>.
                    
<#bq1r5e> a solid:TypeRegistration;
  solid:forClass bk:Bookmark;
  solid:instanceContainer </public/myBookmarks/>.

[jeff-zucker]

I'm in favor of @srosset81's proposal to link all the items as you show as a recommendation to developers whose apps are creating or adding to a type index. The question is what does one do with a type index that doesn't have that predicate (such as every type index in existence at the moment). In terms of reading, apps can, I think, make the assumption that if there is a thing of type solid:TypeRegistration in a document labeled as a solid:TypeIndex that the solid:hasTypeRegistration relationship can be inferred. There is also the question of a "correcting app" - one that can recognize a broken typeIndex and fix it. In order to fix it, the app would have to make that assumption in order to add the hasTypeIndex triples.

[jeff-zucker]

My comment above speaks in terms of documents but the same applies to graphs. If the graph is defined as a collection/list of things of type X, and there are things of type X in the graph, readers of the graph can assume that those things are in the collection/list. Perhaps I am being too lenient.

[lecoqlibre]

I'm in favor of @srosset81's proposal to link all the items as you show as a recommendation

@jeff-zucker I think a recommendation won't be enough for implementations like the one of @srosset81. Like you said, with a recommendation, some TypeIndexes would not contain the solid:hasTypeRegistration predicate. Like SemApps (ActivityPods is based on SemApps) does not support graphs containing multiple subjects it just can't:

make the assumption that if there is a thing of type solid:TypeRegistration in a document labeled as a solid:TypeIndex that the solid:hasTypeRegistration relationship can be inferred.

In other words, SemApps can't read nor write graphs with multiple subjects inside. It does not support quads. There is no possibility to link triples to other triples except by making the links explicit. I give more details in the issue solid/specification#610.

So currently you can't POST nor PUT to a SemApps server a TypeIndex graph containing registrations at a time. As a workaround you could instead POST/PUT the solid:TypeIndex resource and POST/PUT each solid:TypeRegistration resources in different requests.

But even if you have managed to store the complete TypeIndex in the SemApps server (TripleStore) you won't be able to retrieve it! Because there is no way for the server to know which solid:TypeRegistration belong to the solid:TypeIndex.

I can see different options at this point:

1. Like said before, SemApps could exploit the fragment part of the URL to link the triples.
2. The TypeIndex specification should say that a solid:TypeIndex MUST declare a solid:hasTypeRegistration predicate for each registration of the TypeIndex.
3. SemApps servers should handle multiple subjects in requests, also recognize a request is about a TypeIndex and add the solid:hasTypeRegistration predicate by itself.
4. SemApps servers should handle quads.

If we develop option 2. all the other specifications we might encounter should adopt the same behavior and declare explicit links between triples otherwise SemApps like servers won't work. Also, declaring explicit links would make the resource heavier and more difficult to create. Even if it could be easier to read, some applications might want to check the TypeIndex is complete so it would read the whole document anyway (might be a recommendation to do this).

If we develop option 3. SemApps would have to implement a dedicated handler for any specification that is using implicit links. That makes a huge amount of work! Pretty impossible to do if SemApps want to be generic.

I would be curious of other opinions, from the TypeIndex team for instance. There could be some other reasons why the TypeIndex is making implicit links.

[jeff-zucker]

I wouldn't object to hasTypeRegistration being a MUST for writing a typeIndex. Since there are thousands of existing typeIndexes without it, I would also like the spec to recommend that clients be lenient when reading and make the assumption that all typeRegistrations in a typeIndex are part of that typeIndex whether or not that is explicitly stated. This would work fine on both a SemApps style that had the hasTypeRegistration statements and on one that didn't.

As you've pointed out, there is still the bigger question of whether Solid should require support for quads. It's probably been discussed before (@csarven ?)

[pchampin]

Since for ActivityPods we use a triple store for storage (and not the filesystem), each resource must have its own URI.

All triples stores that I know of are in fact quad stores, they support named graphs, and you could store the triples of each LDP RDF Source into a different graph. This would give you the same flexibility than storing them as separate files in the file system.

[srosset81]

Sorry for the late reply, unfortunately I don't have as much as time I would like to have to follow-up on GitHub discussions :-/

All triples stores that I know of are in fact quad stores, they support named graphs, and you could store the triples of each LDP RDF Source into a different graph. This would give you the same flexibility than storing them as separate files in the file system.

This is unfortunately not the case with Jena Fuseki 3.17, with which SemApps is "stuck" because we developed an extension to check WAC permissions. With this version of Fuseki, you can create as many named graphs as you want, but only those which are set in the configuration files are persisted (I don't know if they fixed this in later versions).

Even without full quad store support, SemApps could be improved to support hashed URI. It would be some work (for which we don't have resources at the moment) but it's not impossible.

However my problem is more philosophical than technical. I understand that, with a filesystem storage, it's more convenient to put everything on the same file. And that it's less work not to indicate the hashes of the TypeRegistrations. But my question is: why enforce all TypeRegistrations to be on the same file ? For me, the philosophy of Linked Data is that you can link (and dereference) data no matter where they are. So why should they absolutely be on the same file ? Why not allow servers to put them on separate resources, if they wish to ? A simple predicate like I'm suggesting above would solve this problem, and make TypeIndexes more flexible.

Concerning backward compatibility, I think that clients can be intelligent enough to keep on parsing TypeIndexes without a solid:hasTypeRegistration predicate. But IMO servers implementing TypeIndexes should use this new predicate.