Crate djb_hash[][src]

This library is a collection of simple and fast hash functions based on designs done by Daniel J. Bernstein. Since these hash functions are very simple they have several known limitations and should be use with extreme caution if used for things like key hashes in HashMap, etc. Most of them are venerable to key collisions which can lead to DOS attacks if exposed in any way to external bad actors.

After all that being said, many other languages, like Java, JS, Python, and PHP, have use versions of them internally and it was felt that there would be some utility for Rust to have the hash functions available when doing interoperable code in these cases.

Understanding hash names

The hash names are based on the mathematical functions they use to create hashes. I'll start with breaking down one here: "X33a". The "X##" is the multiplier stage, so before appending the next byte of data to be hashed the existing hash value will be multiplied by the number given which in this case is 33. This number will always be prime and to make the calculation as fast as possible only primes that are binary multiples + or - one are used. The reason is that actual multiplication is slow but bit shifting which act like multiplying in binary and addition or subtraction are much faster. For 33 it is convert to a shift of 5 (times 32) and then add the original number to equal the final 33.

Next there is the "a" part. The "a" stand for "add". After multiplying the running hash total as given above the next byte to be hashed is added to it to form the new hash total. Some hashes instead of adding use XOR (Exclusive OR) and would have an "x" where the "a" is in the example I've been using. Using XOR usually does a better job of distributing the effect of the new byte across more of the hash bits but the effect is less noticeable when using single byte on a longer hash like is done here.

Finally there are the one or more "Xxx" suffixes. The suffix "U32" means that internally a 32 bit unsigned integer hash total is used instead of the normal 64 bit one. Often even on 64 bit platforms 32 bit operations can be faster and may be better fit for an application. Rust by default always returns 64 bit hashes from its finish() function but since often only a 32 bit hash is needed I have added 32 bit versions as well. To return a 64 bit hash the internal 32 bit one will be zero extended to 64 bits in finish(). To save from converting to 64 bits then back to 32 bits when only 32 bits are needed the finish_u32() function can be used instead of the normal finish() function with all of the 32 bit hash versions.

The last "Xxx" part of the suffix denotes some additional operation or other specialization like is done in some application. A good example of this is in PHP where the high bit is always set because they use a zero hash value to detect an unset hash internally.





This trait is used by 32 bit hashes.