14 Jul
2022
14 Jul
'22
8:42 p.m.
there might be some merit to building the index on the fly rather than incrementally. basically, you want the biggest full subtrees between the writes, where these full subtrees are taken from the content of the last flush. the reason to have a full subtree is to limit the depth and sibling node count together. however, interestingly since some leaves end up on internodes, one could construct a subtree with the same depth that has more leaves than a full subtree would. in practice, the full subtree approach limits the sibling node count, because the extra subtrees that are in excess of the full count can be merged together with future writes, to make another equally-sized full subtree. :S what does that mean in a depth-focused approach