We calculate based on array length and hash of the elements, with some bitshifting etc.

Based on MRI's implementation it appears we're missing a few things:

• Murmur calculation of the hash
• Pseudo-random salt (MRI uses rb_ary_hash pointer address)

I can take this up tonight, looks like a nice bug get started.

Here is the change, https://gist.github.com/Who828/056848941c6efee4044e
Let me know if that's what you expected before I send the PR.

The PR looks good, but I wonder if we should also add some kind of pseudo-random salt like MRI. Could be a one-shot random number from almost anywhere. Or we could skip it.

@enebo, @tduehr: Yay or nay? Do we need it?

I think it's a good idea to have one per runtime. I'm not entirely sure the full impact, there may be some issues relating to shared objects between runtimes.

#2443 gets us using proper hashing algorithm but does not include salting. I'll file a separate issue for that, since I think we need a bit more discussion about what to use for salt.

Had to revert the patch because it has other issues.

The index and realLength passed into the hash was not for the bytelist but for the array. That was one thing I noticed. The other issue I saw too was [nil, false].hash nil.toString().getBytes() seems to return null?

Looks like this is fixed in 9.1.10.0:

[].hash
=> 5632391355580947474
[2].hash
=> 4399786657182928470

Woot, thanks!