Gary Chambers wrote in a message to All:

 GC> Can someone quickly calculate the CRC-32s of "UAL" and
"AAL"
 GC> (no quotes), and tell me what you get?  For some strange
 GC> reason, I was under the [mis]conception that it was
 GC> impossible to have the same CRC for different values. 
 GC> Stumped...

Not only is it quite possible to have the same CRC for different values,
but it is an absolute necessity since there are only 2**32 possible CRC-32
values. In fact, it is relatively easy to apply basic linear algebra
concepts in order to modify a data block and keep the same CRC, since CRC
is not designed to resist this.

If you need specifically a one-way hash with a guaranteed distribution,
then you need to look at the cryptographic hashes such as MD5 or SHA.  It
is also worth noting that cryptographic hashes produce much larger ranges
-- 128 bits in the case of MD5 and 160 bits in the case of SHA -- than any
common CRC.

It would surprise me if two distinct three-byte strings had identical
CRC-32, but I see no reason that it would be mathematically impossible, and
I can even see ways to construct such strings if they exist.
 
-- Mike


---