Clearly a lot of work went into this so kudos for that.
I do think it missing the point a tiny bit though. For me the primary use of Ngrok is to automatically get sub a domain with TLS and this seems to be outsourcing that to Caddy. And also I need to run a server. If I was going to run a server and run caddy I'd probably just use SSH -L directly.
Also SSH -L is a TCP/IP level tunnel so I'm missing something regarding why the HTTP/2 multiplexing is necessary vs just using the tunnel as is.
SSH does some kind of lower-level multiplexing. Multiplexing is unavoidable here because the local server must maintain simultaneous TCP connections that must be somehow distinguished between each other whereas the tunnel is just a stream of binary data.
I understand multiplexing is necessary. My point is SSH tunnels already multiplex. If two people connect to the same port on the tunnel machine then that will be two seperate sockets on the destination machine. Since they are already two separate sockets what is the point of adding HTTP/2 protocol awareness at the tunnel level? Just let the endpoints handle it.
Oh looks like we misunderstood each other. The fact is that h2tunnel is not using SSH, because my goal was a minimal Node.js native solution. So I had to look for a built-in alternative for multiplexing and found HTTP2.
I do think it missing the point a tiny bit though. For me the primary use of Ngrok is to automatically get sub a domain with TLS and this seems to be outsourcing that to Caddy. And also I need to run a server. If I was going to run a server and run caddy I'd probably just use SSH -L directly.
Also SSH -L is a TCP/IP level tunnel so I'm missing something regarding why the HTTP/2 multiplexing is necessary vs just using the tunnel as is.