Closed rohanag closed 8 years ago
yeah, the zero vector check was in the wrong place, causing many good tree paths to be neglected during tree traversal.
The funny thing is, the bug affected some trees more than others, for e.g. results on the tree in the test case are unaffected by this fix. I'm not exactly sure why, and what the memory location at topNodeOffset
is supposed to contain when the node being traversed has more than MIN_LEAF_SIZE
descendants.
LGTM, looks like a bugfix to me.