PERFORMANCE: Avoid progress updates if no one is listening

[HenrikBengtsson]

A quick thought: I wonder if it's possible for the progress publisher to request that subscribers to confirm they're interested in the progress updates or not. If not, the publisher won't have to waste time creating and signaling progression conditions that will not be listened too.

[HenrikBengtsson]

Yes, when within the same R processes, a subscriber can signal back using something like:

library(progressr)
foo <- function() { 
  do_progress <- FALSE
  progress_confirm <- function() do_progress <<- TRUE
  progress("setup", confirm = progress_confirm)
  if (do_progress) progress("hello")
  42
}

Then:

> withCallingHandlers(foo(), progression = function(p) { if (!is.null(p$confirm)) p$confirm(); print(conditionMessage(p)) })
[1] "setup"
[1] "hello"
[1] 42

If not confirmed, there are no updates either:

> withCallingHandlers(foo(), progression = function(p) { print(conditionMessage(p)) })
[1] "setup"
[1] 42

[HenrikBengtsson]

The same setup-confirm communication model could possibly also work for other types of communication channels, e.g. socket connections.

[HenrikBengtsson]

And now only one handler needs to confirm, i.e. confirm and then call invokeRestart("muffleProgression").

[HenrikBengtsson]

Also, confirmation can/should only be performed if done from within the same R session (e.g. session uuid)

[HenrikBengtsson]

This requires also a way to handle the case when subscribers have no way to communicate back to the publisher, i.e. when there's one-way communication.

Another example is the relaying of conditions by the future framework: when future collects conditions = "progression" (a) it needs to confirm that it receives them, or (b) the publisher needs a way to know that it will not receive a confirmation in this case. For the latter, maybe this can be done via a global R option which is set for some of the future backends (types of backend workers) but not others?