[Hendrikto]

In general, I get your argument, but cryptography is the perfect example for something so well-specified, well-understood, and extremely widely used, that these arguments do not really apply. You are not going to have to make backwards-incompatible changes to SHA256 or Poly1305, etc. It has minimal API surface too, and is not going to be a large maintenance burden. But nearly everybody is going to need crypto at some point. It is great to have blessed and well-audited standard implementations that people can rely on, without even having to make a choice. This it is not something where you want “the community to make backwards incompatible changes at a higher pace.”.

Crypto is something any modern language should include in the stdlib, imo.

[tptacek]

Proviso: you do cryptography in the stdlib if you have the means to do it well. Go did; Filippo Valsorda has built a whole company practice on keeping that library excellent.

Certainly, that's a better outcome than just providing bindings to OpenSSL, which is what most other languages do.

[bborud]

I think I agree, but I am not sure I understand what "the means to do it well" actually means. I would think of it more as a community decision: let's focus efforts on doing X well in the standard library since X is important and people shouldn't have to ask "ok, so which third party library is the best choice right now".

Let me be the first to point out that this is not an easy thing to do since it depends heavily on the community/team/maintainer dynamics. Even agreeing on the goal or scope can be really hard. But if the Rust community is as good as people say it is, that should be doable, right?

On the other hand it wouldn't have to be definitive and exhaustive. Just a safe default. Like http muxers in Go: the one that came with the standard library was fine for a lot of uses, but people generally used third party muxers. I certainly did. And then one of the most used third party ones became shaky as it was no longer being maintained (now it is again, I think). Eventually the one in Go was improved to where I'd prefer to use that since it represents a "safe default" (and I am probably not going to need whatever extra features or performance third parties can provide).

Also note that I see myself first and foremost as an _engineer_. I care less about purity in theory than what things mean in practical terms. And in practical terms I appreciate Go for having so much useful stuff in the standard library. Stuff I kind of think we should take for granted in any serious language today.