There is something I can't understand in transaction management in
a CMT environment. When demarcating a transaction the declarative way
in a session bean (a method name 'A' with REQUIRED attributes for example).
If this method A uses a data access layer that is developped by hand (I mean with java classes getting their own database connexion).
If I need call differents method of my dat access layer, each of them getting a connection from the pool, how can I be sure that the connection
will be always the same to manage the transaction correctly ? It is written somewhere in a spec that when a transactional context exist, it must be associated with one connection to the database, and this association must remains until a commit or a rollback is done ? wht's the matter if there different transactional resources involved in a (distributed) transaction ?