Re: [ANN] LuaJIT-2.0.0-beta1

[Florian Weimer <fw@...>]

* Rob Kendrick:

> On Wed, 04 Nov 2009 20:33:28 +0100
> Florian Weimer <fw@deneb.enyo.de> wrote:
>
>> * Rob Kendrick:
>> 
>> > One approach I've used in the past is to use an (extremely) trivial
>> > PRNG to decide which byte to hash next, using the current hash
>> > value as the seed, and the length of the string as the initial seed.
>> 
>> See the other thread---it's pretty easy to force collisions with
>> longer strings.
>
> To force, or accidentally come by?  If the former, it's pretty easy
> whatever hash you use,

Sure?  Try lookup3.c from Bob Jenkins.

> unless it's a cryptographic one, which you don't want at all in this
> context.

You don't want a slow function.  But why not something resistant to
collisions?

Those cryptographic hash functions try to solve a much more difficult
problem: absence of collisions without any secret.  For some opaque
implementation, the "without any secret" part is certainly overkill.