Robert Kennedy writes:

> rsync seems fundamentally ugly to me because it depends on good luck
> of checksums to work. When there are checksum collisions, it can
> fail silently and heinously.

I don't know about luck. The strength of the checksum is discussed in
Andrew Trigell's PhD thesis:

       http://samba.org/~tridge/phd_thesis.pdf

Section 3.2.4 discusses the selection of the strong checksum, which is
128-bit MD4.  That it is 128-bit MD4 is also documented in the manual
page for rsync: see the --csum-length option.

> Not a common occurrence, but still a fundamental weakness IMHO.

Theoretical at this point, unless you know of a data set where you can
get 16 byte MD4 checksums to collide for different inputs?

Regards,

Giles