twofish

[Georg Wrede]

I wonder if the Unittest for encryption and decryption libraries should, 
as test cases, contain a piece of nontrivial text (and possibly a tough 
binary file) that is originally [en/de]crypted with a _known_good_ 
implementation.

And a good explanation of how to acquire such a Known Good reference, so 
that anybody interested could try to repeat the test. (This is standard 
procedure in science, and cryptology is about the only thing where such 
(over ambitious) diligence makes sense in programming.)

How else would one know whether the code has only been debugged with its 
own counterpart, or with a universally accepted "base"? (Of course I'm 
talking corner cases, exceptional input, and the like, which usually 
slips past the original coder, IMHO, including myself.)

I think this goes for _any_ crypto SW submitted to the D community.

(This is not to say I have doubts on this particular implementation, 
this is a suggestion for a Best Practice.)