Ryan Brush

Agreed that simply wrapping this with something like AlphaRootsWrapper is our best bet here. Note that the underlying LinkedHashMap implementation already allocates at least one object per map entry, so...

Sorry for being slow here. This is on my to-do list, but unfortunately between traveling and other demands I haven't gotten back to it. I'll see if I can pick...

The immediate issue is fixed in PR #397 here. I am wondering if this goes far enough, though. For instance, the tuned-group-by function would be prone to a similar issue,...

Alright, I went ahead and merged the immediate fix and logged #398 to consider the tuned-group-by -- which I agree isn't an immediate need. I'm also happy to push a...

@alex-dixon I agree that an alpha release is probably warranted here to make it easy to try out. There doesn't seem to be a measurable impact on complex rules (since...

@WilliamParker The one I found that shows the most change is the old clara-benchmark project: https://github.com/rbrush/clara-benchmark. The most pronounced benchmark is the clara.simple-query-large, when run like this (and modifying the...

Unfortunately I don't see a clear way to eliminate the overhead here. As a test I just had the wrapper call the Java .equals method to separate the cost of...

I saw the same results @WilliamParker saw when switching to .hashCode. Nice find! I hadn't thought that would be the most significant item. @mrrodriguez I agree with your comments in...

Alright, I took a pass at this and created PR #399 for comment. This brings the penalty at least into the mid single digits for this microbenchmark, and likely negligible...

Sounds great, Luc! We'll probably want to wait until 1.9 is released before merging to master, but having this ready in a branch we can can merge with 1.9 drops...