问题
I created a doubly-linked list, and the benefits of a sentinel node were clear - no null checks or special cases at list boundaries.
Now I'm writing a red black tree, and trying to figure out if there is any benefit to such a concept.
My implementation is based on the last two functions in this article (top down insertion/deletion). The author uses a "dummy tree root" or "head" to avoid special cases at the root for his insertion/deletion algorithms. The author's head node is scoped to the functions themselves - seemingly due to it's limited usefulness.
One other article I came across mentioned using a permanent root above the head as an "end" for iteration. This seems interesting, but I tried it and couldn't see any real benefit over using NULL as an end iterator. I also found several articles that used a shared sentinel to represent all empty leaf nodes, but this seems even more pointless than the first case.
Can anyone elaborate on how a sentinel node would be useful in a red black tree?
回答1:
Red-black tree implementations almost always use one black sentinel for all the leaves.
It saves a lot of null checks when you can check color without checking for null first.
This doesn't work in implementations that use parent pointers, of course. In those implementations leaf sentinels usually aren't used because you'd need to allocate a different sentinel for each leaf position, and that's a waste of much memory.
来源：https://stackoverflow.com/questions/45739238/benefit-of-a-sentinel-node-in-a-red-black-tree