CSVW reference formulation instances

[pmaria]

See: kg-construct/rml-lv#34 (comment)

In RML-LV it is possible to switch reference formulations when defining fields voor a logical view, for example when you have nested CSV data nested within a JSON document. CSVW would help a great deal to be able to specify how to handle the csv data (delimiter, header, etc).

A proposal for this is to allow instances of rml:CsvwReferenceFormulation to configure CSVW aspects for that specific reference formulation, similar to namespaces and rml:XPathReferenceFormulation.

[]
  rml:logicalSource [
    rml:source [
      rml:root rml:MappingDirectory
      rml:path "data.csv" ;
    ] ;
    rml:referenceFormulation [
      a rml:CsvwReferenceFormulation ;
      rml:option [
        rml:null "" ; 
      ] ;
      rml:csvwTableSchema [
        a csvw:TableSchema ;
        csvw:columns ( "id" "name" "age" ) ;
      ] ;
      rml:csvwDialect [
        a csvw:Dialect ;
        csvw:delimiter ";" ;
        csvw:quoteChar "\"" ;
        csvw:escapeChar "\\" ;
        csvw:commentChar "#" ;
        csvw:header false ;
        csvw:skipRows 1 ;
      ] ;
    ] ;
  ] ;

[s-minoo]

Isn't CSVW (as defined in the io-registry specs) automatically usable wherever there's a rml:referenceFormulation?
Or is there something explicit that I need to define in the CSVW reference formulation spec?

[pmaria]

As I understand it, the current definition of CSVW assumes that a CSVW description can be used as an rml:Source. At least, that was how it originally was described in RML-IO, and I believe that is still the definition currently (although the examples are not entirely clear on this at the moment).

This proposal is to allow the CSVW description to, alternatively, be defined on an instance of the reference formulation, instead of as the source. This will allow us to use CSVW on nested fields in RML-LV, and possibly in other use cases like provided sources (#10)

[s-minoo]

I'll need to update the others too then, because my understanding is that all the definitions under reference formulation (see the screenshot) should all be usable wherever there's an rml:referenceFormulation.

[pmaria]

I'm not sure I follow. Except for the XPath reference formulations, all other reference formulations currenty only describe their usage with a single individual instance of the respective reference formulation (the default reference formulation). So for example rml:JSONPath:

[]
  rml:logicalSource [
    rml:referenceFormulation rml:JSONPath ;
    rml:source [ ... ] ;
    rml:iterator "$[*]"
  ] .

The same for the currenty described CSVW reference formulation:

[]
  rml:logicalSource [
    rml:referenceFormulation rml:CSVW ;
    rml:source [
      a csvw:Table ;
      # CSVW definitions
   ] ;
    rml:iterator "$[*]"
  ] .

What this issue proposes is different:

[]
  rml:logicalSource [
    rml:referenceFormulation [
     a rml:CSVWReferenceFormulation;
     # CSVW definitions
   ]
    rml:source [ .... ] ;
    rml:iterator "$[*]"
  ] .

[s-minoo]

Yes I definitely agree that that CSVW (or any other reference formulation spec) should be used as part of rml:referenceFormulation and not as part of rml:source. That is also the default use case that comes to my mind when reading the reference formulation specs.

But I do not see where the current definitions of any reference formulation spec state that they're used as part of rml:source.

Since you said that XPath reference formulation states explicitly that it should be used as part of rml:referenceFormulation (not rml:source), is there a section or a paragraph in the XPath reference formulation spec that I could use as an inspiration for other specs?