Python examples segfault easily

[mmaker]

The file python/kyber1024/kyber1024_params.py seems meant to be edited by developers.

However, if for instance I set the m vector to m=3 or m=10, the python script segfaults!
It seems there is some mishandled parameter validation in the C library.

Concretely, I'd like to know what is the time for proving knowledge of many secret keys sharing the same A term but with different secrets. Is there an easier way to check it without crashing the library?

[p-steuer]

The dimension of A in kyber1024.py was computed incorrectly. It accidentally worked with the given example but would crash with different choices. This is fixed with commit d603cc7.

Setting m >= n makes no sense, since m x n is not the dimension of the matrix from the kyber public key, but of the corresponding linear relation. Concretely, let (A1,t) be the kyber public key and A2 the identity matrix. Then
t = A1 * s + e ---> t = A * (s,e)^t, with A=[A1,A2], so A1 dimension is m x (n - m), and A2 is m x m.

For proving multiple kyber secrets with one proof each, one only has to change the left hand side t of above linear relation when setting the statement, saving a little setup time.

For proving multiple kyber secrets in one proof, for a statement, one can tensor A with an identity of appropriate dimension.
However, Id assume the fact that A is reused does not give you much compared to using different As.