Nearest neighbour not in `locate_within_distance` neighbourhood about rstar HOT 5 CLOSED

I believe this bug is fixed in master, but the fix has not made its way into the latest release on crates.io. Is there any chance a new release could be prepared? Given that this bug effectively breaks nearest-neighbor lookup, it seems to me that bringing a fix out ought to be prioritized.

from rstar.

Version 0.9.0 has been released and contains the fix. Thanks everyone for the contributions!

from rstar.

It seems like the Envelope implementation for AABB has a bug. This particular line squares the min and max values before comparing them. However, that can cause the actual min to be considered as the max. Moving the squaring lines below the to below the compare-and-swap operation seems to fix the test case.

@aschampion can you review what I've mentioned above and see if that makes sense. From git log, it seems like you changed the logic while optimizing the computation of min_max_dist_2 (from O(d^2) to O(d)).

from rstar.

Thanks for getting in touch @rmanoka.

I'm not sure that's right -- we care about min and max absolute value, so squaring them first will not affect which is min and which is max. And doing the comparison first as in #64 will be incorrect in cases where the max is negative.

The issue reported here may be due to the consistency issue reported in #40. I'm not at my workstation, but will rebase and update #40 tomorrow as requested and check whether it resolves this test case.

from rstar.

@aschampion Thanks for clarifying; my suggestion was indeed incorrect.

from rstar.