On Aug 10, 2008, at 5:11 PM, Konstantin Käfer wrote:
32 bit allow for more than 4 billion unique IDs. If any site goes over it, they can easily change their columns (or rethink their data model).
I would definitely accept random 32 bit identifiers as a step in the right direction. That being said, on any 64 bit machine, the 32 bit identifier is represented internally as a 64 bit identifier anyway. It is totally conceivable to me that the set of drupal sites in existence could quickly encounter collisions with a 32 bit identifier. Again, a driving use case is "two drupal sites that were founded independently of any knowledge of each other would like to merge their databases". Whenever you're using a hash function to generate identifiers, having a key space that is close to the size of the number of keys dramatically increases the chance that a collision will occur. By using a 64 bit identifier, we minimize that chance. Thanks for your feedback, ~ethan fremen