Re: Compute Hash Code

On Wed, 2004-12-08 at 15:23 -0800, Tony Carolla wrote:
> Even with MD5, there are possibilities of different input values
> generating the same key.  I was just reading an article the other day
> that demonstrates this weakness.  I am not sure if I can send out
> links on this list, but if I can, the link is here:
> 
> http://www.doxpara.com/md5_someday.pdf
> 

Right, any kind of checksum routine _can_ produce an identical result
for two different inputs, unless the checksum is at least the size of
the data.  That's been the case forever. 

Simple example, there are only ten possible signatures in a mod-10 check
digit calculation.  It's still a useful algorithm for short numbers, and
transposing or dropping digits will likely cause a checksum error.

In any case, with md5, the chances of a dupe in any reasonably sized set
of documents is negligible.  I would not hesitate to use an md5sum for a
key into a table of products.  

However, I might be inclined to try a GUID (globally unique id).  The
guid is a long string of hex digits that includes a component of the
system id, the date and time, so it is *extremely* unlikely that a dupe
could ever be accidentally created.  It's not really a checksum, though,
just a unique number.  Don't know if that makes it useless for the task
at hand, though.

Regards,
Rich