[IronPython] Hashing in IronPython

[Michael Foord]

At Resolver Systems we are seeing IronPython use a radically lower 
amount of unique hashes for tuples - which we *believe* may be at the 
root of a performance problem with using dictionaries keyed by tuples.

CPython 2.5

 >>> hashes = set()
 >>> for i in range(1000):
...  for j in range(1000):
...   hashes.add(hash((i, j)))
...
 >>> len(hashes)
1000000


Same code on IronPython 2:

IronPython 2.0 Alpha (2.0.0.800) on .NET 2.0.50727.1433
Copyright (c) Microsoft Corporation. All rights reserved.
 >>> hashes = set()
 >>> for i in range(1000):
...  for j in range(1000):
...   hashes.add(hash((i, j)))
...
 >>> len(hashes)
32768
 >>>

We have a work around (using 2D arrays instead of dictionaries) but as 
dictionaries are heavily used in Python it is likely to be a problem for 
others as well.

All the best,


Michael Foord
http://www.manning.com/foord