To implement data access code in our application we need some framework to wrap around jdbc (ORM is not our choice, because of scalability).
The coolest framework I used to work with is Spring-Jdbc. However, the policy of my company is to avoid external dependencies, especially spring, J2EE, etc.
So we are thinking about writing own handy-made jdbc framework, with functionality similar Spring-jdbc: row mapping, error handling, supporting features of java5, but without transaction support.
Does anyone have experience of writing such jdbc wrapper framework?
If anyone has experience of using other jdbc wrapper frameworks, please share your experience.
Thanks in advance.
We wrote our own wrapper. This topic is worthy of a paper but I doubt I'll ever have time to write it, so here are some key points:
we embraced sql and made no attempt to hide it. the only tweak was to add support for named parameters. parameters are important because we do not encourage the use of on-the-fly sql (for security reasons) and we always use PreparedStatements.
for connection management, we used Apache DBCP. This was convenient at the time but it's unclear how much of this is needed with modern JDBC implementations (the docs on this stuff is lacking). DBCP also pools PreparedStatements.
we didn't bother with row mapping. instead (for queries) we used something similar to the Apache dbutil's ResultSetHandler, which allows you to "feed" the result set into a method which can then dump the information wherever you'd like it. This is more flexible, and in fact it wouldn't be hard to implement a ResultSetHandler for row mapping. for inserts/updates we created a generic record class (basically a hashmap with some extra bells and whistles). the biggest problem with row mapping (for us) is that you're stuck as soon as you do an "interesting" query because you may have fields that map to different classes; because you may have a hierarchical class structure but a flat result set; or because the mapping is complex and data dependent.
we built in error logging. for exception handling: on a query we trap and log, but for an update we trap, log, and rethrow an unchecked exceptions.
we provided transaction support using a wrapper approach. the caller provides the code that performs transaction, and we make sure that the transaction is properly managed, with no chance of forgetting to finish the transaction and with rollback and error handling built-in.
later on, we added a very simplistic relationship scheme that allows a single update/insert to apply to a record and all its dependencies. to keep things simple, we did not use this on queries, and we specifically decided not to support this with deletes because it is more reliable to use cascaded deletes.
This wrapper has been successfully used in two projects to date. It is, of course, lightweight, but these days everyone says their code is lightweight. More importantly, it increases programmer productivity, decreases the number of bugs (and makes problems easier to track down), and it's relatively easy to trace through if need be because we don't believe in adding lots of layers just to provide beautiful architecture.