What are differences in many to many mapping approach

Question

I want many to many relation mapping with explicit join table.

My java spring project is providing REST api in HAL format and now it has only two types of classed:

• entities defined and
• "empty" interfaces for repositories (annotated by @RepositoryRestResource).

Sidenote, dependencies are circa these:

spring-boot-starter-parent
spring-boot-starter-data-jpa
spring-boot-starter-data-rest
spring-data-rest-hal-browser
postgresql / h2
spring-boot-starter-hateoas

Relations between tables (I mean _links of rest resources. have a look at sample hal+json document and look for ea:basket for example.) are working as expected and it is working "for free", because of magic powder from spring auto-configuration and other magic included.

I am struggling now when adding new many to many dependency.

I have entities A, B and Tag. I want to have any number of Tag entities to be associated with A an B entities. I have no need to have any list/set in A and B entities (I will use jooq if I need anything more than crud).

First problem/question:

I see at least three approaches to model many to many relation with explicit join table (to be able to model my association needs) and I do not know differences. What are differences of these approaches?:

1. use embeddable composite key in join table as Vlad suggest? https://vladmihalcea.com/the-best-way-to-map-a-many-to-many-association-with-extra-columns-when-using-jpa-and-hibernate/
2. use @IdClass approach as stated here: https://stackoverflow.com/a/3588400/11152683
3. use multiple @Id in join table as shown here: https://hellokoding.com/jpa-many-to-many-extra-columns-relationship-mapping-example-with-spring-boot-hsql/

Second question: What approach is needed in my case to easily model associations and make magic powder do its work when providing HAL format in crud repositories. I mean that links of relations will be generated automagically.

Answer 1

To make many2many association work nicely with spring data rest things and provide HAL representation with correct things, you have first know a bit of JPA/Hibernate. There are two approaches (1 and 2 from question, as third is only shortcut for second one and working in Hibernate only.).

Both approaches are shown in proof of concept repository in tags branch. I use given repository to test various set-ups of project.

Approach 1, EmbeddedId. It does use hackish thing in FixConfig class in BackendIdConverter bean, where it uses bookRepository to get Book entity when it parses request id from url onto embedable id class.

Approach 2, IdClass. It is using plain Integers int its IdClass and does seem to be correct solution.

I think that first approach can be modified to work similarly as second, but I am not able to do it for now.

I would mark as solution any answer providing some insight for "why" it is like this.