"Richard A. O'Keefe" <ok at atlas.otago.ac.nz> wrote:
> 	An object has a hash value (self hash).
> 	
> 	1.    How can I get hold of an object with its hash value?
> 	      anInstance _ AClassCanCreateInstanceUsingHashValue hash: 2340
> 	
> It is in the nature of hash codes that they are not unique.
> There could be many thousands of objects with that hash code.
> Hash codes are not pointers, they're just a hack to speed up table lookups.
> 
> I haven't checked:  does Squeak 3.x still use 12-bit hash codes?

YES!  For my type inferencer, which has a few sets way bigger than 2^12,
I  implement custom hash functions.  4096 is a lot for most purposes,
but ocassional cases require more.

(A cool trick would be if the VM offered optional "long identity
hashes".  Small identity sets ask for regular 12-bit hashes, and large
ones ask for the large, optional ones.  The same could be done at the
image level for regular #hash -- you could have #hashForSize: (or maybe
#hashOfSize:), and certain objects can generate better hashes for larger
collections.  String in particular might benefit!)

Overall, hashing is a funny beast to work with.  It works phenomenally
well when it works, but you have to get several things right or it can
fail horribly!  A funny problem I had once involved removing the first
element from a set and then adding an arbitrary element to the middle;
thus, over time, I had systematically skewed the set's data to towards
the end of the set.  Doh!


Lex