Question

I’ve been puzzling over this for a few days… feel free to shoot down any of my assumptions.

We’re using a Dictionary with integer keys. I assume that the value of the key in this case is used directly as the hash. Does this mean (if the keys are grouped over a small range) that the distribution of the key hash (same as the key itself, right?) will be in a similarly small range, and therefore a bad choice for a hashtable?

Would it be better to provide an IEqualityComparer that did something clever with primes and modulo mathematics to calculate a better distributed hash?

Answer 1

It’s not used directly in that the dictionary will still ask the key for its hash – but the hash value of an Int32 is just the value, so the thrust of your question is relevant, yes.

I believe that the way the .NET dictionary works doesn’t rely on hash values being uniformly distributed. It takes hash % bucketCount where bucketCount is always prime. (That’s from memory though – I could be wrong.)

You could still end up with an inefficient set of keys of course, if they happen to be spaced by the bucket count. That will always be the case though – a hash table would only ever be genuinely O(1) for all keys if they had unique hash values and the table maintained a set of buckets for every possible hash 🙂 In reality it tends not to be a problem. If you happen to know that it will be a problem, then yes, a custom IEqualityComparer<T> could help.