Support OneToOne mapping with JoinTable

[beikov]

Consider this simple entity model

@Entity
public class Document implements Serializable {
    @Id @GeneratedValue
    private Long id;

    @OneToOne
    @JoinTable(
            name = "document_extra",
            joinColumns = @JoinColumn(name = "document_id", referencedColumnName = "id"),
            inverseJoinColumns = @JoinColumn(name = "document_info_id", referencedColumnName = "id")
    )
    private DocumentInfo documentInfo;
}

@Entity
public class DocumentInfo implements Serializable {
    @Id @GeneratedValue
    private Long id;
    private String someInfo;
}

When creating a EntityManagerFactory with the property "javax.persistence.schema-generation.database.action" set to "create" it fails. The error is a syntax error in the generated DDL statement ALTER TABLE DOCUMENT ADD CONSTRAINT DOCUMENT_U1 UNIQUE ().

There are multiple problems here that I found. Apparently a @OneToOne with a @JoinTable is not really supported yet? Anyway, the problems essentially seem to be that the @JoinTable is partly/fully ignored. This is because a join table with a different name i.e. "document_documentinfo" and different column names is generated instead.
There should be 2 unique constraints generated for the join table. One for the "document_id" and another for the "document_info_id" instead of an empty constraint for the "document" table.
The ignoring of the @JoinTable also suggests that a simple @OneToOne without a @JoinColumn annotation would fail too, although in that case a default join column name should be chosen.

The issues are DBMS independent and produce consistently bad DDL on H2, MySQL and PostgreSQL.

[andyjefferson]

As the docs ( http://www.datanucleus.org/products/accessplatform_5_0/jpa/mapping.html#one_one_relations ) say clearly enough, 1-1 with join table is not supported and minority interest use-case. [Even the EclipseLink/Wikibooks docs says that is not part of the JPA spec].

PS, this relates solely to the datanucleus-rdbms plugin so not really the place for it.

[beikov]

Must have missed that then. Thanks for the info.

[andyjefferson]

Marking as wont-fix since the work would have to be on the RDBMS plugin only (nothing in core cares about join tables and the like), and opened datanucleus/datanucleus-rdbms#180, so use that as reference. Not likely to attract my attention in some time ;-)

[chrisco484]

I've never had a need to have a one to one implemented as a join rather than as a bidirectional set of FKs. I imagine it would be slower than a simple bidirectional set FKs because it needs to involve a third table. A joined one to one is an interesting use case so I'm wondering if there is an advantage to this approach that I've missed.