Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!tut.cis.ohio-state.edu!usenet.ins.cwru.edu!mephisto!mcnc!rti!xyzzy!poirier
From: poirier@dg-rtp.dg.com (Charles Poirier)
Newsgroups: comp.sys.amiga
Subject: Re: Another DOS for AMIGA?
Summary: Hash quibble
Message-ID: <350@xyzzy.UUCP>
Date: 22 Feb 90 20:18:58 GMT
References: <1059@trlluna.trl.oz>
Sender: usenet@xyzzy.UUCP
Reply-To: poirier@dg-rtp.dg.com (  Poirier local)
Lines: 19

In article <1059@trlluna.trl.oz> aduncan@rhea.trl.oz.au (Allan Duncan) writes:
>> about 1-2 K. Also, the file names were very similar, thus increasing the
>> possibility of hash-collisions in the DOS!! The disk fragmentation was 
>
>NO, NO!  If the hash routine works, then changing a file name by just
>one bit in one character should give a completely different hash value.
>If it doesn't, then it isn't being hashed!

They may be different, but not necessarily "completely" different.
For some hash functions and collision-handling methods, things can get
bad.  I can't speak to AmigaDOS' hash algorithm, but there are many simple
hash functions where names differing by just the final letter give
hash values differing by just the difference of those letters.  This can
generate a clump of consecutive entries, which can make hash collisions
more severe.  Particularly so if the collision algorithm is "step ahead
to the next unused entry".  Again, I don't know if Amy's hashing does this,
just speaking in general.

	Cheers,
	Charles Poirier