State lingers across notebook loads about rcloud HOT 6 CLOSED

@s-u, I was hoping to use some kind of environment trickery to make this work "nicely". Is there a way to replace the default environment and then simply replace it with a new one when we switch notebooks? My simple tests with attach and detach didn't do anything useful.

from rcloud.

Ah, we don't re-connect between loads? I was assuming each notebook has its own session (that is the Rserve design). This is a bit tricky (because .GlobalEnv is special so just creating a new environment per notebook will break any code that uses .GlobalEnv directly), but solvable -- the question is that we want to do: do we flush the previous state or do we keep all states that have been loaded so far? Obviously, the latter is convenient if you switch between two notebooks, but can use a lot of memory. The former is more "clean" because loading a notebook has always the same effect: clean slate - regardless if it's your first or second try in a session.

from rcloud.

BTW: for other reasons (the graphics device and knitr internal state among them) I would actually suggest creating a new socket. In general, due to some memory management features (such as grow-only symbol table), it's typically more efficient to not re-use the same R session for subsequent tasks. Also package management across notebooks will be a pain (unloading packages is not always reliable).

from rcloud.

The main problem with reloading everything is that the authorization dance is slow and full of redirects.

Is there any chance we can do something on the server side?

from rcloud.

Once you have the tokens, you can simply create the socket, no client-side dances anymore (dances required to get the tokens initially). This is all server-side and it only checks against the sessions server or the local database which is fast. You would only need the dance if we let the tokens expire at which point it's probably not bad to get new ones anyway.

from rcloud.

The only state that's lingering now is loaded packages and graphics state. I'm closing this for now, and we'll handle it better sometime in the future.

from rcloud.