@inducer, the fact that the build for this commit failed is somewhat mysterious, as this was a documentation only commit. I cannot reproduce the failure on my machine yet. Other than some sort of nondeterminism in the code, do you have any idea what could be happening?

Could be a minor FP issue, e.g. mismatched < vs <=. I do believe that the failure is indicative of a latent bug in the code somewhere.

Hmm, looks like there is at least one bug here. The new guiding box finder doesn't do the right thing when the child box fails to exist. I think that may be the reason we are running into the build failure, investigating now.

Figured it out. This happens when the ball center is not within the extent of the tree. Filing as a bug.