Mike Matrigali wrote:
I am ok with moving latching out of the lockmanager, if it really is necessary to get the performance you are looking for. I had hoped that latching would remain a "service" provided by another module so that if there are multiple store implementation they could share a single latching implementation. I assume changes to remove latching out of lock manage will break existing in progress work to do in memory store - but I don't know for sure.
Just want to point out that the work done on the in-memory store is brushed up to a state where it can be applied to trunk. It should be possible to test if a change in the latching code would break it.
One would have to hack the test harness a little to make it use the correct back-end. I might make a patch and upload it to DERBY-686, run suites.All and report on the results. I did try it once, but only ran a few tests.
If anyone else wants to have a go, please do so, as I will be busy and away for the whole next week (JavaPolis).
-- Kristian [ snip ]
