From the unit of work pattern i uderstand a method of doing typic transactions based on some domain repostiries (using a repository per domain object) . Example : after defining some repository objects in the UoW object , commit those repositories based on theyr state .
Also the repositories should not contain any transaction logic .
What happens when an insert() leads to a creation of a new object (auto generated id) that later on is needed by another object in the same transaction ?
Unit of work does not seem to work for this case . There could be even more specific and complex transaction where objects are generated when the UoW commit is ran .
How should the transactions be treated in this case ?
Usually ORMs like NHibernate or EntityFramework know how to handle the order of calls to DB. E.g. NHibernate's inverse is used in order to specify who is responsible for bidirectional relationship.
If you develop your own DataAccessLayer/ORM it is your responsibility to specify the invocation order. The simplest solution is 'Add all new entities' => 'Delete all deleted entities' => 'Update all dirty entities'.
Once an entity is added to DB you can retrieve as a result its @@IDENTITY/ SCOPE_IDENTITY and then update its Id using any appropriate solution.
来源：https://stackoverflow.com/questions/20622203/is-unit-of-work-a-good-pattern-for-transactions-that-will-auto-generate-new-obje