class Client

def call(request, task: Task.current)
	Console.logger.debug(self) {"#{request.method} #{request.path} #{request.headers.inspect}"}
	
	# Mark the start of the trailers:
	trailer = request.headers.trailer!
	
	# We carefully interpret https://tools.ietf.org/html/rfc7230#section-6.3.1 to implement this correctly.
	begin
		write_request(request.authority, request.method, request.path, @version, request.headers)
	rescue
		# If we fail to fully write the request and body, we can retry this request.
		raise RequestFailed
	end
	
	if request.body?
		body = request.body
		
		if protocol = request.protocol
			# This is a very tricky apect of handling HTTP/1 upgrade connections. In theory, this approach is a bit inefficient, because we spin up a task just to handle writing to the underlying stream when we could be writing to the stream directly. But we need to maintain some level of compatibility with HTTP/2. Additionally, we don't know if the upgrade request will be accepted, so starting to write the body at this point needs to be handled with care.
			task.async(annotation: "Upgrading request...") do
				# If this fails, this connection will be closed.
				write_upgrade_body(protocol, body)
			rescue => error
				self.close(error)
			end
		elsif request.connect?
			task.async(annotation: "Tunnneling request...") do
				write_tunnel_body(@version, body)
			rescue => error
				self.close(error)
			end
		else
			task.async(annotation: "Streaming request...") do
				# Once we start writing the body, we can't recover if the request fails. That's because the body might be generated dynamically, streaming, etc.
				write_body(@version, body, false, trailer)
			rescue => error
				self.close(error)
			end
		end
	elsif protocol = request.protocol
		write_upgrade_body(protocol)
	else
		write_body(@version, request.body, false, trailer)
	end
	
	return Response.read(self, request)
rescue => error
	self.close(error)
	raise
end