why compression doesn't perfectly even out entropy

[Perry E. Metzger]

"Jon Leonard" writes:
> Perry's right: giving up any statistical information is too much.
>  
> A slightly contrived example of why tossing out duplicated bytes is bad:
>  
> Suppose that a military organization is using this almost one-time-pad
> system, and my spies tell my they've fallen into the habit of sending
> "attack" and "defend" as their only 6-byte messages.  This isn't a problem
> with a real one-time pad (except for traffic analysis...), but this lets
> me determine the message 3.8% of the time!

This could actually be used for traffic analysis in many instances;
you could succeed in extracting small amounts of information from the
passing data.

Any amount of leakage can in some instances be too much...

.pm