From: Jeff Garzik <jgarzik@pobox.com>
Date: Sun, 24 Apr 2005 00:35:57 GMT
Subject: Re: Hash collision count
Message-ID: <426AE9ED.4060005@pobox.com>
In-Reply-To: <20050423234637.GS13222@pasky.ji.cz>

Petr Baudis wrote:
> Dear diary, on Sun, Apr 24, 2005 at 01:20:21AM CEST, I got a letter
> where Jeff Garzik <jgarzik@pobox.com> told me that...
> 
>>Second, in your scenario, it's highly unlikely you would get 4 billion 
>>sha1 hash collisions, even if you had the disk space to store such a git 
>>database.
> 
> 
> It's highly unlikely you would get a _single_ collision.

Agreed.


>>First, the hash is NOT unique.
>>
>>Second, you lose data if you pretend it is unique.  I don't like losing 
>>data.
> 
> 
> *sigh*
> 
> We've been through this before, haven't we?

<shrug>

In messing around with archive servers, people get nervous using 
(hash,value) based storage if there isn't even a simple test for collisions.

Someone just told me that one implementation of the Venti archive 
server[1] simply fails the write, if a data item exists with a duplicate 
hash value.  As long as git fails or does something -predictable- in the 
face of the hash collision, I'm satisfied.

	Jeff


[1] http://www.cs.bell-labs.com/sys/doc/venti/venti.html