Buffer in mimc should reason on p-digits, not block size

[ThomasPiellard]

In the mimc implementations, this for loop

for i := 0; i < nbChunks; i++ {
        copy(buffer[:], d.data[i*BlockSize:(i+1)*BlockSize])
        x.SetBytes(buffer[:])
        r := d.encrypt(x)
        d.h.Add(&r, &d.h).Add(&d.h, &x)
    }

might create collision, as the modulus is not BlockSize long but usually shorter.

What should be done is to take the value in the buffer, decompose it base p where p is the size of the snark field, and encrypt each digit one after the other, using the previous hash as initial state.

[realuncle]

@ThomasPiellard Any update regarding this issue?

[ThomasPiellard]

Hi @realuncle no currently the issue hasn't been fixed. I wonder if it's not better to directly enforce the Write() function in the mimc implementation to not accept anything which overflows r, the snark modulus. Otherwise if we pass a gigantic slice of []byte for instance it would force us to do the r-basis decomposition, which would be quite inefficient.