futuremice(): Best practices for undoing future::plan() changes

[HenrikBengtsson]

Hello, in

mice/R/futuremice.R

Lines 146 to 168 in ded22e4

	future::plan(future.plan,
	workers = n.core
	)
	# begin future
	imps <- furrr::future_map(
	n.imp.core,
	function(x) {
	mice(data = data,
	m = x,
	printFlag = FALSE,
	seed = seed,
	...
	)},
	.options = furrr::furrr_options(
	seed = TRUE,
	globals = globals,
	packages = packages
	)
	)
	# end multisession
	future::plan(future::sequential)

you're resetting the future plan to plan(sequential) when done. Note that the user might have another type of plan set for other reasons. Instead, set your custom plan as:

 with(future::plan(future.plan, 
   workers = n.core 
 ), local = TRUE)

and drop the future::plan(future::sequential) call at the end.

This is the recommended approach to undo plan changes, cf. https://future.futureverse.org/articles/future-7-for-package-developers.html#avoid-changing-the-future-backend

[thomvolker]

Dear @HenrikBengtsson thanks for the suggestion! This has been on the to-do list for a while already, but we never actually proceeded to change this because we are working to allow mice to have a parallelizable sampler by default, which would make a separate futuremice wrapper obsolete (in #701). This is certainly something we have to keep in mind, so let's leave the issue open for now.