Parametrizable mimc endianness

[hussein-aitlahcen]

Parametrizable mimc endianness

[ivokub]

Indeed, making it parametric is an option, but I think it would still be a bit confusing. We have actually stumbled a few times when hashing from bytes to fr.Element inside MiMC, for example in cases when input byte slice overflows fr.Element.

My proposal has been to instead implement a new hash interface which takes as an input fr.Element (see #448), but the problem with the approach is that we would have to have separate interfaces for every supported field or use type parameters (which has been shown to be slow and may not be most intuitive). Then, it would be up to the user to perform the hashing from bytes to fr.Element (for example by using hash-to-field) in a most convenient way.

But until that as a workaround having either optional endianness or different MiMC constructor seems reasonable. cc @ThomasPiellard and @gbotrel, what do you think?

[hussein-aitlahcen]

Indeed, making it parametric is an option, but I think it would still be a bit confusing. We have actually stumbled a few times when hashing from bytes to fr.Element inside MiMC, for example in cases when input byte slice overflows fr.Element.

My proposal has been to instead implement a new hash interface which takes as an input fr.Element (see #448), but the problem with the approach is that we would have to have separate interfaces for every supported field or use type parameters (which has been shown to be slow and may not be most intuitive). Then, it would be up to the user to perform the hashing from bytes to fr.Element (for example by using hash-to-field) in a most convenient way.

But until that as a workaround having either optional endianness or different MiMC constructor seems reasonable. cc @ThomasPiellard and @gbotrel, what do you think?

Also random question regarding this: would it be cryptographically insecure to consider MiMC as a 253bit (avoid the modulus) block-cipher hash function? If not we could have a sha2-like interface ingesting arbitrary bytes?

[gbotrel]

Technically we can do both -- for the "hash interface that takes bytes", add in the constructor an option (retro compatible change) with an enum big / little endian.
And if one days it comes to it that doesn't prevent us from doing #448 .

[ivokub]

Also random question regarding this: would it be cryptographically insecure to consider MiMC as a 253bit (avoid the modulus) block-cipher hash function? If not we could have a sha2-like interface ingesting arbitrary bytes?

It should indeed be a secure cryptographic hash function. The issue it is not usually used outside of circuits is that it is quite slow. But in circuit due to the algebraic nature it is better than SHA2.

The problem with current interface (that MiMC implements hash.Hash i.e. ingests bytes) is that we have to split the input into log(q) bit partitions, construct individual fr.Element from the partitions and then run MiMC. But the individual partitions could overflows the field. So, we would rather have to represent the input byte slice as base q big integer, but this is very memory-expensive when the inputs are a few MBs.