Ary Borenszweig

Oh, I see what you mean. Yes, each connection should try to reconnect itself.

I'm not sure that's a responsibility of the pool. If, say, an HTTP::Server handles a request and asks a connection from the pool, I would expect there to be an...

> because it's better to be safe than sorry Makes sense. But how can we do it in Crystal? We can have a periodic check... but what do we check?...

Another note: in the standard library we began using `tls` everywhere, because `ssl` is kind of an old name now, though I'm not sure it's applicable here.

But it's not clear that this is what we want, so first we must discuss this instead of reviewing this PR. But it's unlikely this will happen before 1.0. And...

I personally think the rule should be: - If there's a way to make the string representation not involve method calls, let's do that (this is the case of `URI::Params`,...

In Ruby methods can start with an uppercase letter, so you can do this: ```ruby require "uri" uri = URI("https://example.com") ``` In Crystal that syntax conflicts with generics. Maybe there's...

I think we can try using the `URI.new(...)` approach. It's just a few more letters compared to `URI.[]` and it doesn't involve adding a new method just for this. I...

> Again, #inspect is meant for humans, not machines and it's not a serialization, but rather a textual representation using a format that resembles a source code I think @straight-shoota...

You shouldn't document those (at least from an API point of view) because you are not supposed to directly expose them. You should wrap them with Crystal constructs, and document...