Re: Formal objection to removing features from node shapes

Peter,

Since the WG has already decided to allow the use of sh:disjoint and sh:equals for node shapes, I think presenting a formal use case for using sh:disjoint in the node shapes would not change the specification. It could only serve as a way to inform a working group and, as a result, either confirm the WG’s opinion that this is a fairly uncommon use case for typical data modeling or not. But in either case, it is now supported.

Future proofing of SHACL, in my understanding, has been discussed in the context of SHACL SPARQL and other extensions for allowing users to extend the language beyond SHACL Core. To my knowledge, WG has never thought it could built into SHACL Core everything any user may ever want to say about a valid shape of their RDF graphs, let alone what such users may want to say in the future should RDF change.

I have asked you to re-write your formal objection in light of the above change.

Would it be correct to summarize your current formal objection as follows:

This is a formal objection to the decision of the RDF Data Shapes working group to reclose ISSUE-139 by making node shapes ill-formed if they use any
of sh:minCount, sh:maxCount, sh:uniqueLang, sh:lessThan, sh:lessThanOrEquals, or sh:qualifiedValueShape.

Node shapes that use any of these properties have suitable obvious definitions so there is no problem adding them back to SHACL.

With the resolution node shapes and property shapes have different features. This difference complicates the language, making it harder to explain to users and harder to machine-generate.  Because the language is more complex implementations become more complex and testing becomes more complex.

Removing these node shapes from the language may cause a drop in expressive power.  I  can't think of any loss in RDF per se. However, if generalized RDF (allowing literals as subjects) is considered then there would be expressivity loss.  Of course, SHACL is only defined for RDF itself, but literals as subjects has come up as a possibility for future versions of RDF.

I am working on Wiki pages to document all your formal objections and the WG responses to them in a way that doesn’t require going through all the e-mail threads. So, I need a description of the objection as it stands now after the change already made by the WG.

If this is not an accurate description of your objection, please provide one.

Thanks,

Irene

> On Mar 6, 2017, at 10:07 AM, Peter F. Patel-Schneider <pfpschneider@gmail.com> wrote:
> 
> I presented an example of the kind of expressivity that that was lost by the
> decision of the working group to remove sh:disjoint from node shapes.  If that
> is not adequate then I can easily put together a formal use case for
> consideration by the working group.
> 
> I don't view it as a given that there will no longer be any expressivity loss.
> I can't think of any loss in RDF per se, but that doesn't mean that there
> won't be any.   Is the working group prepared to be wrong twice in a row?
> 
> If generalized RDF (allowing literals as subjects) is considered then there
> still is expressivity loss.  Of course, SHACL is only defined for RDF itself,
> but there had been quite a bit of talk about future-proofing SHACL, and
> literals as subjects has come up as a possibility for future versions of RDF.
> 
> So, no, I am not willing to withdraw my objection.
> 
> Peter F. Patel-Schneider
> Nuance Communications
> 
> On 03/01/2017 07:44 AM, Irene Polikoff wrote:
>> Peter,
>> 
>> The WG has decided to restore the use of sh:disjoint and sh:equals for the node shapes, even though we do not have any use cases demonstrating that this expressivity is commonly required by the users and, thus, needs to be in SHACL CORE.
>> 
>> Given that there would no longer be any expressivity loss in SHACL CORE, would you remove the formal objection?
>> 
>> If you still formally object, please rephrase your formal objection so that it no longer relies on the loss of expressive power and the specific example you have provided.
>> 
>> Thanks,
>> 
>> Irene 
>> 
>>> On Feb 24, 2017, at 10:17 PM, Irene Polikoff <irene@topquadrant.com> wrote:
>>> 
>>> Peter,
>>> 
>>> Do you have examples of the loss of expressive power in cases of sh:minCount, sh:maxCount, sh:uniqueLang, sh:lessThan, sh:lessThanOrEquals, or sh:qualifiedValueShape constraint component?
>>> 
>>> Regards,
>>> 
>>> Irene
>>> 
>>> 
>>>> On Feb 24, 2017, at 9:57 PM, Peter F. Patel-Schneider <pfpschneider@gmail.com> wrote:
>>>> 
>>>> This is a formal objection to the decision of the RDF Data Shapes working
>>>> group to reclose ISSUE-139 by making node shapes ill-formed if they use any
>>>> of sh:minCount, sh:maxCount, sh:uniqueLang, sh:equals, sh:disjoint,
>>>> sh:lessThan, sh:lessThanOrEquals, or sh:qualifiedValueShape.
>>>> 
>>>> Node shapes that use any of these properties have suitable obvious
>>>> definitions so there is no problem adding them back to SHACL.
>>>> 
>>>> With the resolution node shapes and property shapes have different features.
>>>> This difference complicates the language, making it harder to explain to
>>>> users and harder to machine-generate.  Because the language is more complex
>>>> implementations become more complex and testing becomes more complex.
>>>> 
>>>> Removing these node shapes from the language causes a decided drop in
>>>> expressive power.  For example,
>>>> 
>>>> ex:s1 rdf:type sh:NodeShape ;
>>>> sh:targetClass ex:C1 ;
>>>> sh:disjoint ex:p1 .
>>>> 
>>>> checks that SHACL instances of ex:C1 do not have themselves as a value for
>>>> ex:p1.   This useful ability cannot be obtained through any other means.
>>>> 
>>>> So the resolution to close ISSUE-139 removes useful expressive power from
>>>> SHACL without appreciably reducing implementation or testing costs or
>>>> reducing user confusion.  All these features need to be added back to node
>>>> shapes.
>>>> 
>>>> Peter F. Patel-Schneider
>>>> Nuance Communications
>>>> 
>>> 
>>