rust-orphan-rules VS getrandom

Well, unless someone comes up with better, compatible rules, the orpan rules are gonna stick around.

Additionally there are other issues with rust currently. Compile time code (ala constexpr) is not up to par with C++20 (not really close). The const generic aren't as powerful as C++20 which added non primitive non type template parameters (though with you stuck with C++14, it actually is significantly better than what you have, again, if you're going to use C++, just use 20). Generics accepting closures is a bit more of an ordeal in rust, compared to C++. Also C++'s Duck Typed templates allow for some uncharacteristically strong typing compared to what is expressible in Rust generics currently. Now, duck typed templates do have major downsides, for example the entire feature of concepts is completely irrelevant in rust, but required for sane DTT type bounds, but they also have major upsides. Rust currently doesn't have "negative trait bounds", basically "This objected does not implement this trait, or std::enable_if> or the equivalent concepts implementation. Rust also doesn't have trait specializations, basically template specialization. Do note all features I've talked about to this point have nightly options, they just are at various stages of being stable/complete. Another issue is the orphan rule, though this is kind of a problem in C++ too in some respects, and that's unlikely to change drastically, since there are legitimate reasons for it's existence. For a lot of code none of these things are big deals, others they are, which is why you find inconsistent feed back on these issues.

Not that simple... hence why Orphan rule is still in-place. The struct wrapper was implemented in Rust as a temporary safe work-around. However, they are making progress on a solution: https://github.com/Ixrec/rust-orphan-rules/issues/1

That's still not an entirely complete explanation because there's more nuanced situations which aren't completely foreign but are foreign enough that if allowed, two crates could write the same impls. Some of the definitions are still unofficial as far as am I'm aware. For the best reference I’ve seen so far see this for more details.

In our crate the compiler doesn't know when calling MyTrait methods on MyStruct whether to use the implementation defined in crate 3 or crate 4! Rust has a set of orphan rules to prevent this situation from happening.

Sadly because of the rusts orphan rule you cannot implement a Trait on a Type where you do not own one or the other. So, apart from upstream contributions your only options are either a new Trait or a new Type.

If anyone is looking for additional background about orphan rules, check out https://github.com/Ixrec/rust-orphan-rules

The crypto source in Go is great, no complaints there. Lints like gosec even recommend using it when generating crypto entropy. Go did a good job here, and I expect Rust will do the same sometime after getrandom reaches 1.0 so the API questions are settled, plus whatever makes sense for the future-proofing the standard library needs.

I would wait for the getrandom crate to reach 1.0, which will answer many of the questions around what an API like this can look like, and then maybe the standard library discussion will be on firmer footing because at least we'll know what API we want to immortalize. Rushing that now just to save people importing a small crate does not seem to be the way to go.

I'd caution against using /dev/random directly, and instead recommend using getrandom. It's effectively the same thing on Haiku and Redox, but is cross-platform and will upgrade to better sources on various platforms as available (such as using the getrandom() call on Linux and Android, or getentropy() on macOs, if avaialable).

getrandom

It's a default, but overwritable behavior, see the #[path] attribute. You still have to create N files for each supported platform, but at the top level you will see only one module. On of the crates which uses this approach in practice is getrandom.

IIUC the main problem which prevents from moving HashMap & co to alloc is lack of API to get system entropy which is required for DOS protection. Ideally we would have a #[global_allocator]-like functionality for retrieving system entropy. Relevant issue: https://github.com/rust-random/getrandom/issues/21