On Jul 26, 6:27 pm, Clifford Heath <n... (AT) spam (DOT) please.net> wrote:

Yes in theory but not always in practise. E.g. SHA-2. A "decent"
hash function with n output values requires about sqrt(n) evaluations
for a successful birthday attack. In certain application domains
(e.g. content addressable file stores) the chance of a collision can
be made small enough for a suitable hash to be regarded as unique for
all practical purposes.

Google in academic sites for the research on "minimal perfect hashing"
"perfect hashing" and "databases" in the literature. It is a fairly
active area that I have not looked at in along time. But as I recall,
there was a proof that a perfect hash always exists. That mean a hash
would alway require one probe to retrieve a record from a file, unlike
tree structured indexes which can require several probes.

On Jul 24, 3:23*am, Karsten Wutzke <kwut... (AT) web (DOT) de> wrote:
Apparently he didn't kick hard enough. :-) You've found an answer to
your "how", but I believe (he can confirm) that Bob's point was that
you skipped "what". Same applies to the developers of the hash
functionality to which you linked. I needed read no further than
this:

public class Person {
String name;
int age;
boolean smoker;
....
}

The wrong direction is the effort to do your data managment in the
application, rather than in the database. *That* is the wrong
direction, and the best of all solutions you come up with will be
fabulously sophisticated hacks which will work fine for your current
task at hand, impress the hell out of any equally naive developers
with whom you work, and leave you with a glass-brittle application
that nobody will want to breathe wrong on for the next 10 years.

Seeing enough of it can make you as jaded as, well, Bob.