Re: [Classpathx-crypto] on testing (long)

[Raif S. Naffah]

Casey Marshall wrote:
> On Mon, 24 Jun 2002, Raif S. Naffah wrote:
>
>
> > > For the known-answer tests I would recommend using texts/keys where a bit
> > > is set somewhere in each byte/nybble/word, to try to cover a range, not
> > > just the first few.
> > >
> > > Other than that I would simply use a Comfortable Round Number.
> >
> > let's say 5 values for each test, which makes a total of 30 --including
> > the KAT ones.  the position of the bits for the latter, should be
> > selected at random.
>
>
> I think for the KA tests the number of tests should depend on the length of
> the key/text, by (say) using the word 0x80000000 at all the possible word
> (dynner, 32 bit) aligned offsets, e.g.
>
>    8000000000000000000000000000000000000000
>    0000000080000000000000000000000000000000
>    ...etc.
>
> just because it is seems less arbitrary and spans the space of keys/texts.

i'm not sure.  if the cipher is not correctly implemented, it's highly 
unlikely to pass some KATs and fails others.

even if this is the case, a cipher implementation should be deemed 
correct iff it passes *all* the tests!


> > [...]
> >
> > how do you want to split the work?
>
>
> I could do the TestOfXxxx classes, maybe make a general superclass that
> does the work, and subclasses that simply fill in the test vectors.

excellent;  a base test-case for the TestOfxxx ciphers would be great!

i'll do the selfTest() implementations.

ps. checkout the latest TestOfSerpent and Util before you start.


cheers;
rsn