It is written in HiA on several occasions that the use of a composite class can lead to unnecessary complexity and implementing two one-to-many associations on entity classes might be easier. (See HiA p.197, 220, 225, 231).

I have a case almost exactly like the example on p. 230, a link table with extra data (comment, timestamps, etc). The hbm.xml mapping generation tool created a composite class.

When is the appropriate time to use a composite class?

The book also explains in very much detail _when_ a particular mapping is appropriate or when a different mapping might be easier. Since you do not have a question about the book per se, please post in the regular User forum.