Signed-off-by: Stephen J Day <stephen.day@docker.com>

Printing all requests and responses to the log is quite expensive, especially when sending megabytes of data through 9P. This patch leaves the logging in the error paths and the flush path. Signed-off-by: David Scott <dave.scott@docker.com>

We take some time to cleanup the codec implementation. The need for the message to implement the Message interface has been removed to allow flexibility in what can land in the message field. Error types can be returned directly now. The messages have been moved into a separate file. We also ensure that the number of pointer types is reduced in deserialized messages, hopefully avoiding unnecessary pointer traversal. Signed-off-by: Stephen J Day <stephen.day@docker.com>

Signed-off-by: Stephen J Day <stephen.day@docker.com>

Signed-off-by: Stephen J Day <stephen.day@docker.com>

Signed-off-by: Stephen J Day <stephen.day@docker.com>

The repl and client session work fully with rminnich/go9p/ufs. The key to this was proper support for directory serialization, which is subtle. The size field for Rread was the incorrect size. Several robustness improvements are also included. Issues with the timeouts and missing error returns have been addressed. Signed-off-by: Stephen J Day <stephen.day@docker.com>

After a few connection bugs and framing issues, the transport has been refactored into a channel that manages raw message sending. More importantly, this frees up pre-session protocol version negotiation. To support this, the encoder and decoder have been folded into a Codec interface. The encoding support still needs work, but this has made the server and client code much simpler to understand. Signed-off-by: Stephen J Day <stephen.day@docker.com>