From: Steven E. H. <se...@pa...> - 2006-06-15 15:16:56
|
Sam Steingold <sd...@po...> writes: > "exhaust"?! > what do you mean _exactly_? There was a hidden assumption in my thinking. If the value of sxhash is used as a hash key directly, the keys in CLISP collide more often. That would matter if some hash table was using most-positive-fixnum buckets. Assuming that such spacious hash tables are rare, the sxhash key would need to be mapped to the bucket count again, in which case the spacing between sxhash keys is less important. > the contract of sxhash is suitability as a hashing function. > I don't see it violated. I didn't suggest that it was violated. I was only commenting that if the goal of a hashing function is to produce semi-unique keys for different values, the keys produced by CLISP are less unique than those produced by the other implementations we mentioned, and hence warranted discussion. As I said above, being less unique doesn't necessarily condemn the behavior of a hash table making proper use of these keys. Looking at just the keys in isolation may be too narrow a view. -- Steven E. Harris |