Rye should respect ssh_config_options better

[rickhull]

My use case for rye is to automate things that I'm doing "manually" via SSH. I have an extensive SSH config (~/.ssh/config) that specifies which keys and users to use for which hosts. I am finding that Rye::Box is not respecting my ssh config. Rye::Box is unaware of the various keys in ~/.ssh and it (obviously) doesn't try to use them when connecting to hosts that specify those keys in ~/.ssh/config.

I did some source diving, and it looks like the only ssh config option that is respected is :port, and to some extent :user. I say "to some extent" because I found in some cases that the user from ~/.ssh/config was not being used, though I see in the source that it should be, via @rye_user. I haven't gotten to the bottom of this yet.

I also use ssh-agent extensively to hold my decrypted private keys in memory. I'm not sure if Rye will use this, and looking at Rye.sshagent_info and @@agent_env -- do these do anything? I see that @@agent_env is initialized to an empty hash -- and then never updated again anywhere.

I really like this project, and I'm eager to use it, but I'm having trouble understanding some of these design decisions. I'd be happy to collaborate and discuss more extensively.

[rickhull]

I figured out the issue with username. Rye::Box works as expected, but Rye::Set does not:

https://gist.github.com/rickhull/9c9c056f040afd11329d

[delano]

Hi Rick, thanks for reporting these. I haven't had much time lately to spend on Rye. I'm happy to pull in a patch that pass all tests but unfortunately can't be much in debugging.

[rickhull]

Honestly, I probably won't tease this apart either. But it seems clear to me that the Set does not reliably approximate a "set" of boxes. The abstraction leaks.