muendlein

> @muendlein might be worth increasing the number of searches between timings for a possibly more reliable measurement. Does the last commit help? I'll likely land this anyway eventually since...

> > After observing that the gap still exists even if the code is never reached, I'm unfortunately out of ideas. > > One idea is to set an empyrical...

> @muendlein I've seen similar behavior before, and my guess is that it's because of v8 inlining. Over a certain threshold of complexity or size, v8 stops inlining the function,...

With the latest commit the gap has been reduced but is still there. ``` after last commit before this PR 1000000 rectangles 1000000 rectangles Indexing: 182.94 +- 11.81 ms Indexing:...

@mourner Inlining `addLeafSegment` into `addAllLeavesOfNode` does not have a performance impact (I just pushed the commit). What exactly is your idea about the heuristic approach to check the relative query...

> @muendlein so, if I do [fb78a2e](https://github.com/mourner/flatbush/commit/fb78a2e1223820996c3b4850e7bf449596f63852) and then put `if (false && minX

@mourner I think you are asking the right question about what is the better choice here. Given the varity of datasets & query usecases, I think 1 solution fits all...

> 👍 to exposing an option if the regression is unavoidable. "mouse hover single data point in a scatterplot of 300k points" is something i would prefer not to regress....

@mourner I just tested the simplest heuristics approach and it seems like we are back fighting the compiler. Approach (note: dataArea is already calculated during finish): `const isLargeQuery = ((maxX...

@mourner As some time has passed, I'm wondering if you already had time to play around?