John Cowan <xxxxxx@mercury.ccil.org> writes:

> William D Clinger scripsit:
>
>> There is no reason for clients or implementors of SRFI 126 to think
>> (h x 0) has anything to do with (h x 1), and so on.
>
> This reminds me that I'd like you to weigh in on the subject of
> hash-function clients being able to supply salt to hash functions to
> discourage DoS attacks based on the known behavior of standard hash
> functions.  The salt helps to populate the initial state of the hash
> algorithm, and is supplied by the framework on a per-process-instance
> basis, or more fine grainedly if desired (such as a per-hashtable basis).
>
> This salt could be supplied as yet another argument to the hash function,
> or by a parameter that the hash function is free to inspect but not to
> mutate (R7RS parameters can't be mutated anyway).  It would be an error
> to even inspect this parameter outside a hash function.
>
> Note that user-written hashes need access to the salt too, for we can
> often assume that the bad guy has the source of the program.

FWIW draft 9 of SRFI-126 will probably contain:

- `(hash-salt)` *procedure*

Returns a global and constant salt value for use in hash functions.
This is a random value for every run of the program, except when the
environment variable `SRFI_126_HASH_SEED` is set to a non-empty string
before program startup.

Taylan