Brett Okken

As @panchenko mentions, there is no leak. There is not even "unintentional object retention". Would a better approach to this be a property which caps the size of the decoder...

@sehrope, I have been testing various decoding options with jmh benchmarks off and on for the past year or so. The current implementation is still the overall best when considering...

Also, in practice the synchronization does not end up being all that impactful. A different instance is created for each connection, so there is very rarely (if ever) any contention.

> I guess with bigger values caching the buffer would not help much There might be more benefit there than you would expect. G1GC, at least in java 8, does...

Are any Connections/ResultSets being processed concurrently? /An/ explanation could be concurrent use of a Calendar object.

Based on https://github.com/pgjdbc/pgjdbc/pull/474, it seems that perhaps it would be good to have @marschall and @vlsi review.

@davecramer and @vlsi, I am somewhat baffled by the seemingly random test failures I keep seeing here. The latest is The command "./.travis/travis_install_openj9.sh" failed and exited with 4 during ....

So some of the issues were caused by travis being flaky, others appear to have been nature of the tests and more aggressively cleaning up cached statements that have gone...

@vlsi, I think this is what we discussed sharing parsed query texts and execution counts across connections. A weak reference is used for the cached result and the NativeQuery object...

@vlsi, can we get this PR merged?