Version française
Home     About     Download     Resources     Contact us    
Browse thread
[Caml-list] Hashtbl.hash and Hashtbl.hash_param
[ Home ] [ Index: by date | by threads ]
[ Search: ]

[ Message by date: previous | next ] [ Message in thread: previous | next ] [ Thread: previous | next ]
Date: -- (:)
From: Xavier Leroy <xavier.leroy@i...>
Subject: Re: [Caml-list] Hashtbl.hash and Hashtbl.hash_param
>  What kind of algorithm is used to compute the hash code of objects in
> O'Caml ?
> 
>  Hashtbl.hash (List.map (fun x -> Random.int 100)
> [1;2;3;4;5;6;7;8;9;10]);;
>  always returns 0 (Hashtbl.hash_param has the same properties) which is
> a poor result !

Yes, this is disappointing.  To understand what's going on, here is
how "Hashtbl.hash_param v count limit" works:

- v is traversed, depth-first.
- "Interesting" information found at each node is hashed, e.g.
      string node -> hash code of string
      integer     -> the integer itself
      constructor block -> integer tag of constructor
  (Some nodes have no interesting information, e.g. certain custom blocks.)  
- The hash values for each node are combined with a simple linear congruence.

Moreover, to prevent infinite descent in cyclic values, and ensure
that hashing doesn't take too long, the traversal is stopped when either
- "count" interesting nodes were found, or
- "limit" nodes (interesting or not) were traversed.

Now, for your example [1;2;3;4;5;6;7;8;9;10], the interesting nodes
and their associated hash values are
- the integers 1 to 10, with same hash values;
- and the 10 occurrences of the "::" constructor, which correspond to
  0-tagged blocks, with hash values 0.

The fly in the ointment is that the traversal is done right-to-left,
hence the hash values of interest are encountered in the following order:

  0 ......... 0 10 9 8 7 6 5 4 3 2 1

  ------------- --------------------
  the :: cells    the list contents

Hence, with count = 10, the traversal stops at the cons cells, and
doesn't even look at the list contents!  Result: a 0 hash value.

There are several ways to remedy this behavior, such as ignoring
zero-tagged blocks, or doing breadth-first traversal.

However, we need to think twice before changing the hashing function,
because this would cause trouble to users that store hashtables in
files using output_value/input_value: if the hash function changes
before writing and reading, the hashtable read becomes unusable.

Hence, a request for OCaml users: if you use hashtables whose keys are
structured data (not just strings or integers), *and* your program
stores hashtables to files, *and* it's important for you that these
persistent hashtables can be read back with future versions of OCaml,
then please drop me a line.

- Xavier Leroy
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners