/**
* At the highest level, we adhere to the classic b-tree insertion algorithm:
*
* <p>1. Add to the appropriate leaf 2. Split the leaf if necessary, add the median to the parent
* 3. Split the parent if necessary, etc.
*
* <p>There is one important difference: we don't actually modify the original tree, but copy each
* node that we modify. Note that every node on the path to the key being inserted or updated will
* be modified; this implies that at a minimum, the root node will be modified for every update,
* so every root is a "snapshot" of a tree that can be iterated or sliced without fear of
* concurrent modifications.
*
* <p>The NodeBuilder class handles the details of buffering the copied contents of the original
* tree and adding in our changes. Since NodeBuilder maintains parent/child references, it also
* handles parent-splitting (easy enough, since any node affected by the split will already be
* copied into a NodeBuilder).
*
* <p>One other difference from the simple algorithm is that we perform modifications in bulk; we
* assume @param source has been sorted, e.g. by BTree.update, so the update of each key resumes
* where the previous left off.
*/
public <V> Object[] update(
Object[] btree, Comparator<V> comparator, Iterable<V> source, UpdateFunction<V> updateF) {
assert updateF != null;
NodeBuilder current = rootBuilder;
current.reset(btree, POSITIVE_INFINITY, updateF, comparator);
for (V key : source) {
while (true) {
if (updateF.abortEarly()) {
rootBuilder.clear();
return null;
}
NodeBuilder next = current.update(key);
if (next == null) break;
// we were in a subtree from a previous key that didn't contain this new key;
// retry against the correct subtree
current = next;
}
}
// finish copying any remaining keys from the original btree
while (true) {
NodeBuilder next = current.finish();
if (next == null) break;
current = next;
}
// updating with POSITIVE_INFINITY means that current should be back to the root
assert current.isRoot();
Object[] r = current.toNode();
current.clear();
return r;
}
