Vladimir Sitnikov
Vladimir Sitnikov
I guess this changes behaviour, so the only way we can make it is via connection property, so we release a version that adds a property and keeps old behaviour,...
Does the change impact `ParameterMetaData`? If so, I think we need adjust the tests accordingly. An alternative (additional?) option could be adding the newly added parameters to the test matrix:...
In practice driver does know the primary is available (e.g. in case somebody else established a workable connection to primary). At the same time, the driver could piggy-back on `.isValid()`...
> Out of curiosity why does the other test in this package (SocketFactoryTestSuite) not follow this convention? Frankly speaking, I have never used "test suites" to launch a subset of...
@stuartwdouglas, sorry for the late reply, and I agree `Class.forName(...)` without classloader is problematic. However, I think the lookup order should be `Class.forName(String)` first, and only then ask TCCL. The...
Here's a similar story in Kafka (they highlight there are issues with TCCL in OSGi): https://cwiki.apache.org/confluence/display/KAFKA/KIP-60+-+Make+Java+client+classloading+more+flexible Yet another alternative is to implement `socketFactoryObject` property (not sure regarding the name though)...
>if you stop and restart the driver from the old ClassLoader hangs around forever breaking things Do you have a reproducer? Does the issue persist if you use `org.postgresql.Driver.unregister()` across...
Just in case, I am OK with having an option between `regular-first` and `TCCL-first` (I would prefer TCCL-last by default). However, if you have a full reproducer, then it would...
The change is not compatible with the previous behavior. However, I believe the incompatibility is subtle enough so we can ship it without a connection property. The database uses `search_path`...
@arons , can you please clarify which issue does this change solve? Please see https://github.com/pgjdbc/pgjdbc/pull/2835#issuecomment-1466364745 I am afraid, it is hard to review the change without understanding the problem it...