Thanks for working on this!

I wonder if we can get away with just implementing it with EVP_MAC. Given that OpenSSL versions older than 3.0 have reached EOL (see https://openssl-library.org/policies/releasestrat/index.html), we might not need to support older versions of OpenSSL for a completely new feature.

Having said that, I wonder if we want to further make the EVP_MAC wrapper more generic, rather than limiting it to just CMAC. EVP_MAC supports additional algorithms we may be interested in adding, such as KMAC and keyed BLAKE2. We'll eventually rewrite OpenSSL::HMAC around EVP_MAC, too, so it would be nice to avoid duplicated code.

Perhaps we can implement an algorithm-agnostic OpenSSL::MAC that also provides low-level access to those OSSL_PARAMs, along with a specialized CMAC class built around it.

What do you think?

That will be the best solution if we can stop supporting EOL versions of OpenSSL and all versions of LibreSSL, at least about new features. We are also supporting LibreSSL, and I think it has never had EVP_MAC functions.

I thought about almost the same approach when I started working on this issue, but I didn't know if we could soon stop supporting live versions or add features that wouldn't work in some environments. That is why I made that hybrid code.

I will try to implement OpenSSL::MAC and some subclasses if I am assured that we can stop supporting LibreSSL somehow.

In that case, we should add those subclasses under OpenSSL::MAC (e.g. OpenSSL::MAC::CMAC ) and leave OpenSSL::HMAC as it is, shouldn't we?

We also have a choice to define OpenSSL::MAC only when it's compiled against an OpenSSL or LibreSSL release which supports it. I didn't realize that LibreSSL doesn't support it yet, but I expect it'll catch up soon.

When the version number it will be implemented is unknown, we can check it at compile time in ext/openssl/extconf.rb.

In that case, we should add those subclasses under OpenSSL::MAC (e.g. OpenSSL::MAC::CMAC ) and leave OpenSSL::HMAC as it is, shouldn't we?

That sounds reasonable to me. We can't break OpenSSL::HMAC even if an old version of OpenSSL or LibreSSL is used, so rewriting it on top of the new interface will be a future task.

#891 is a follow-up to this pull request.