I don't see this approach avoid the collision. I think if the key.hashcode is larger than table.length, there will be collision.
Updates：
Actually I refer to the HashMap#hash in JDK 1.8, and was a little confused about the benifit of spread the higher bits downward.
Now, I think I'm clear with the help of this link, the benifits is:
- We don't need to do the % calculation, but using a more faster way - bit shift.
For the collision, if the number of key is larger than the length of the table, then there will be a collision no matter what hash method is used.