Simple primary keys are easy to understand ...

Example 22 from 3.8 Schemas shows:

      "foreignKeys": [{
        "columns": "countryRef",
        "reference": {
          "resource": "https://example.org/countries.csv",
          "columns": "countryCode"
        }
      }]

All nice and easy: the value from the single column countryRef is bound to the resource described by the row that has the matching value in primary key column countryCode so that one can link the two tables together.

However, Example 9 from 3.1 Property Syntax shows a composite primary key definition:

"primaryKey": [ "givenName", "familyName" ]

We know from this definition that there is a constraint such that the combination of givenName and familyName for a given row must be unique within the table.

But if I want to create a foreign key reference to a given row specified with a composite primary key, what would the value in the referencing column (cell) look like?

For example:

Table 1

Table 2

In this case I want to say that the value of person in table 2 maps on to the row with matching primary key from table 1. Somehow I need to map "person": "Jeremy Tandy" to "givenName": "Jeremy", "familyName": "Tandy".

It seems to me that we need to describe how the value in table 2 person maps onto the columns in table 1.

(I know this is a contrived example because all the github identifiers must be unique - but hey ...)

I wonder if we can use the URI Template syntax for this? e.g. for table 2 ...

      "foreignKeys": [{
        "columns": "person",
        "reference": {
          "resource": "<table 1>",
          "primaryKey": "{givenName} {familyName}"
        }
      }]

Does it even make sense to have a foreign key reference from an array of columns? I can't think where a foreign key reference would not relate from the value in a single column. Should it be permissible to provide an array of columns in a foreign key definition (e.g. foreignKeys/columns)? Should we limit ourselves to a single column?

It might even make sense to define the primary key in the same way; e.g. for table 1

"primaryKey": "{givenName} {familyName}"

(Note that this also affects how the @id in mapped RDF is generated)

What do folks think?