In this case I feel changing the LRUCache to support generics was an integral part of the changeset to prevent possible regressions with cache usage (as I was changing the cache key from String to another type). Without it, it would be possible to use the incorrect object type in one of the many code paths, by adding generics we benefit from compile time checks.
Often getting that CLA signed involves getting at least the legal department involved, especially in larger, more corporate organizations. In those cases, indefinitely maintaining a patch may sound like a good alternative.
I didn't benchmark the difference between cache usage with precalculating the hash code in a constructor vs calculation in the hash code itself but string concatenation then taking the hash code of that was the real performance killer.
The link is right down the bottom but the GitHub page also contains the benchmark code and results which seemed more interesting https://bugs.mysql.com/bug.php?id=88021
Hey johnou, thanks I saw this, my comment originally had been a response to https://news.ycombinator.com/item?id=15449672, which I interpreted as referring to a different bug, and was then moved and edited by an admin ... anyways thanks for your contribution and my colleagues will review and probably (as said: I don't know enough Java and thus can't judge it) apply it :)
