Tracking issue for additional array size support: `error[E0277]: the trait bound U...: ArraySize is not satisfied`

[tarcieri]

hybrid-array relies on impls of ArraySize on individual typenum::Unsigned types, e.g. U1, U2, U3, which is a deliberate tradeoff versus generic-array which is able to support any array size.

These ArraySize impls are the linkage between typenum types and core arrays, which provide the underlying storage used by hybrid-array, and with it const generic linkage and significantly less unsafe code.

Unfortunately, you will encounter the rustc error message in the issue title if you attempt to use an unsupported array size.

Per #57, we are attempting to handle such cases as one-off requests for sizes which can be granted on a case-by-case basis, and gating them under the extra-sizes. It's unclear how well this will work in practice, but we are attempting it for now.

The main reason to constrain the number of supported sizes is a compile time cost as we add more impls of ArraySize/AssocArraySize. Compile times are not currently too significant but get worse as we add support for more array sizes.

[oddcoder]

As per my proposal in #78, I will need sizes for NTRU-prime which are U653, U761, U857, U953, U1013, U1277. I will also need to add the type U1277 itself. I am not fully aware if I will need more sizes, since I am not fully familiar with the mathematics behind NTRU but unless there is an objection, Kindly expect a PR for those very soon.

[tarcieri]

@oddcoder see how sizes are defined here:

https://github.com/RustCrypto/hybrid-array/blob/master/src/sizes.rs#L617-L811

It shouldn't be too hard to add.

[tarcieri]

#81 provides a solution for user-defined types, however it requires removing the supertrait bound on typenum::Unsigned which seems like it will cause additional breakages beyond what hybrid-array will already entail.

For the v0.2.0 release I'd like to punt on that for now, but it's something we can consider for v0.3.

We can consider it for hybrid-array v0.5 after the next set of @RustCrypto releases, to avoid having to remove the typenum::Unsigned bound at the same time people are migrating from generic-array.