Seam grew out of the frustration of the Hibernate team with the statelessness typical of the previous generation of Java application architectures. The state management architecture of Seam was originally designed to solve problems relating to persistencein particular problems associated with optimistic transaction processing. Scalable online applications always use optimistic transactions. An atomic (database/JTA) level transaction should not span a user interaction unless the application is designed to support only a very small number of concurrent clients. But almost all interesting work involves first displaying data to a user, and then, slightly later, updating the same data. So Hibernate was designed to support the idea of a persistence context which spanned an optimistic transaction.
Unfortunately, the so-called "stateless" architectures that preceded Seam and EJB 3.0 had no construct for representing an optimistic transaction. So, instead, these architectures provided persistence contexts scoped to the atomic transaction. Of course, this resulted in many problems for users, and is the cause of the number one user complaint about Hibernate: the dreaded LazyInitializationException. What we need is a construct for representing an optimistic transaction in the application tier.
EJB 3.0 recognizes this problem, and introduces the idea of a stateful component (a stateful session bean) with an extended persistence context scoped to the lifetime of the component. This is a partial solution to the problem (and is a useful construct in and of itself) however there are two problems:
The lifecycle of the stateful session bean must be managed manually via code in the web tier (it turns out that this is a subtle problem and much more difficult in practice than it sounds).
Propagation of the persistence context between stateful components in the same optimistic transaction is possible, but tricky.
Seam solves the first problem by providing conversations, and stateful session bean components scoped to the conversation. (Most conversations actually represent optimistic transactions in the data layer.) This is sufficient for many simple applications (such as the Seam booking demo) where persistence context propagation is not needed. For more complex applications, with many loosly-interacting components in each conversation, propagation of the persistence context across components becomes an important issue. So Seam extends the persistence context management model of EJB 3.0, to provide conversation-scoped extended persistence contexts.