Hello everyone,
I hope this can clarify my proposal for ISSUE-18 presented at the last meeting.
The proposal is based on the following two ideas:
1) Each TriplesMap is associated with exactly one relation.
2) Relations can be of two kinds:
- RefereciableRelations which are those that have a name in the DB schema (i.e. tables and views);
- QueryResultRealtions which are the runtime relations generated from the result of a SELECT SQL query.
The following UML illustrates the relationship between the TriplesMap and the Relation.
The following R2RML demonstrate the usage of the proposed schema:
<TriplesMap1>
a rr:TriplesMapClass;
rr:relation <MyTable>;
...
<TriplesMap2>
a rr:TriplesMapClass;
rr:relation <GetFailStudents>;
...
<TriplesMap3>
a rr:TriplesMapClass;
rr: relation <GetFailStudents>;
...
<MyTable>
a rr:ReferenciableRelationClass;
rr:tableName "my_table";
rr:tableOwner "my_schema".
<GetFailStudents>
a rr:QueryResultRelationClass;
rr:SQLQuery "SELECT id, name FROM my_schema.students WHERE grade < 50".
The advantages that I see on this proposal are:
- Abstraction of the relation attributes to its own semantic object.
- Separation of concerns between the TriplesMaps and the Relation
- Enable reuse of the relation objects by multiple TripleMaps because now every relation has its own URI.
The disadvantages that I see on this proposal are:
- Changes the structure of the current R2RML specification.
- Introduces complexity to the R2RML language.
The names of properties and classes could (or should) be changed to others that are more user-friendly and meaningful.
Cheers,
Alex
