Steven Allen
Steven Allen
I agree but also, please be careful and avoid any massive breaking changes (e.g., leave an alias behind and make sure the change doesn't break much).
So, we'd have to be *very* careful here. All of our services (DHT, bitswap, identify, etc.) will break if we start dialing ourself. Somewhat off topic... We may need some...
> but it does not seem crazy that any cluster-application using RPCs ends up trying to contact itself for some things In my experience, self-RPCs tend to cause more trouble...
@hsanjuan so, I take it this is blocking you?
Pointers: 1. Add a swarm and libp2p option to enable dialing self. 2. Add a special immortal loopback conn to the connection set on init. Given the fact that the...
How should `network.ClosePeer(self)` behave? Should we always have an implicit connection to ourself.
SGTM. Maybe we should just throw away `SetConnHandler` and use notifications?
> The rationale is that protocol handlers rely on the assumption that incoming streams are never local. Honestly, we should fix these services. Really, I think the main issue is...
My main concern here is multiple generic values. IMO, we should support them (although we can also just "fail cleanly" for now and do that later).
Note: [eglot-booster](https://github.com/jdtsmith/eglot-booster) avoids this by setting a buffer-local variable on the jsonrpc request buffer and checking it in the advice. That could be replicated in the lsp advice fairly easily....