My understanding of Dain's cmp code is that any SQLException will result in the tx being set rollback only, and basically all work discarded.
In the new local jdbc wrapper, I've done something about as drastic: if there is any SQLException from any operation, the connection is discarded and the connection handle you get is then useless (unless you hold onto it over a method boundary, then it can get reassociated on the next call). My interpretation of the jca spec is that this is the expected behavior for a jca adapter. However,... 1. This breaks at least some code in the test suite, the cts bmp unit test basically tests for the existence of its table by looking for an exception. The solution is to always discard a connection handle after any exception. How much user code is likely to break? Does anyone know what the j2ee spec says about SQLExceptions (a reference would be great) If this makes JBoss more spec compliant, do we care if it breaks preexisting code? 2. I wrote the wrapper so it would be easy to write an extension for a specific database that could look at the SQLException content and only discard the connection if it was actually unusable. If we do this, what happens to the original SQLException? Do we turn it into a warning? Maybe this idea of db-specific exception handling won't work. Any opinions? thanks david jencks _______________________________________________ Jboss-development mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/jboss-development
