Is it correct that without this change, if you use Puma to terminate SSL/TLS, you aren't getting the benefit of #1227?

If that's the case, would be good to change the PR title so that info is reflected in the changelog/history – I didn't understand what this was about until I read the chat where #1227 was linked :)

Similar code and naming is used Server#cork_socket & Server#uncork_socket

Yeah I'm not a big fan of "skt" 😆 but I will 🙈

Can we test this?

I see that #1227 didn't add a test, I wonder if tests have been added later on or if this is untested? 🤔

Sorry. I wasn't clear. I should have seen it before.

The MiniSSL::Socket isn't related to IO or any Socket, it just has a few methods that make it appear IO like. But, it has a TCPSocket that it 'connects' with. So, testing whether Client#io.is_a? TCPSocket will never pass, but testing Client#io.to_io.is_a? TCPSocket will. Same thing is done at the start of cork_socket and uncork_socket.

Or, when passing in a MiniSSL::Socket or OpenSSL::Socket to methods that also handle TCPSocket or UNIXSocket , one needs to be careful for what tests/methods are with them...