Re: [Monotone-devel] Re: Hash collisions resiliency

[Nathan Myers]

On Fri, Apr 15, 2005 at 11:35:53AM +0200, Peter Simons wrote:
> Jon Bright writes:
> 
>  > I still don't think you've understood it, but Nathaniel
>  > seems to have satisfied your (really rather inane)
>  > queries.
> 
> I don't think there's a need to be condescending. Tom's
> questions were not inane at all, IMHO, and I think it's
> perfectly sensible to ask them. Some people really want to
> know stuff, you see, no matter whether others believe they
> "need" to know it or not.

The reason for our bemusement is that people demanding to know how
they can recover from a hash collision display no such trepidation
over much more devastating events that will occur a billion billion 
billion billion times more frequently.  If they are merely curious,
they may study the source and figure it out themselves, and not
pester people who have real things to work on.  I can say with 
complete confidence that any professional statistician who worries 
about it should seek other employment, and anyone who employs such 
a one should do what he can to help hurry that process along.  

Of actual practical interest is what might happen were a bad packet
to be delivered while repositories are reconciling.  Are the entire 
contents of the stream hashed, or only database contents but not 
communications metadata?  Similarly, what if a bit flips on disk?
Both events are certain actually to occur in actual use, at least
to some users.  Does SQLite hash its own data blocks?  PostgreSQL
includes a 64-bit CRC on every block written to disk.

Nathan Myers
address@hidden