You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
so i kind of feel like i have some concerns about our current design for the registry and connection: we currently kind of couple "looking up a service in the registry and allocating client channels" with "actually sending the service a Hello message" and I'm not sure if i actually like that
right now, for example, our I2C driver service has a Hello = (), and then it has a StartTransaction` request that takes the I2C addr you want to talk to, and then the response to that is "here's a channel to actually send read/write operations as part of that transaction"
this doesn't play nice with MGNP since if we want to do serdes based interfaces with tricky-pipe, there's no way to send the client another channel
so it seems like a better design would be one where the Hello was the "start transaction with addr" message, and then every request/response on the connection channel was the "hi please read/write these bytes" request and the "here is your buffer back" response
but, with the current design for service connection and Hello-ing, that means we would be doing the I2C service registry lookup and allocation of client channels for every i2c transaction. which i dislike.
MGNP should fix this by decoupling "hello"-ing from connecting, i think. which it doesn't currently do.
The text was updated successfully, but these errors were encountered:
so i kind of feel like i have some concerns about our current design for the registry and connection: we currently kind of couple "looking up a service in the registry and allocating client channels" with "actually sending the service a
Hello
message" and I'm not sure if i actually like thatright now, for example, our I2C driver service has a
Hello = (), and then it has a
StartTransaction` request that takes the I2C addr you want to talk to, and then the response to that is "here's a channel to actually send read/write operations as part of that transaction"this doesn't play nice with MGNP since if we want to do serdes based interfaces with
tricky-pipe
, there's no way to send the client another channelso it seems like a better design would be one where the Hello was the "start transaction with addr" message, and then every request/response on the connection channel was the "hi please read/write these bytes" request and the "here is your buffer back" response
but, with the current design for service connection and
Hello
-ing, that means we would be doing the I2C service registry lookup and allocation of client channels for every i2c transaction. which i dislike.MGNP should fix this by decoupling "hello"-ing from connecting, i think. which it doesn't currently do.
The text was updated successfully, but these errors were encountered: