Deprecating 'multiprocess' in favor of explicit 'multisession' or 'multicore'

[HenrikBengtsson]

I've decided to deprecate 'multiprocess' in the next release.

The main reason for this is that 'multiprocess' is ambiguous. For example, 'multiprocess' may cause confusion when people on Linux/macOS and MS Windows discuss issues and benchmark results surrounding futures. This is the same problem we have with doParallel::registerDoParallel().

Another reason is that 'multicore' is might be unstable depending on R environment and whether or not multi-threaded processing is used in the future expression. The latter might change out of the control of the developer and end-user, when a package dependency is updated.

The deprecating warning is produced by plan(), e.g.

> future::plan("multiprocess")
Warning message:
Strategy 'multiprocess' is deprecated in future (>= 1.20.0). Instead, explicitly specify either 'multisession'
or 'multicore'. In the current R session, 'multiprocess' equals 'multicore'.

[HenrikBengtsson]

Running revdep checks with 'multiprocess' deprecated triggers R CMD check WARNINGs for:

• dragon 1.0.4 (2020-10-03) 1.0.3 (2020-09-29)

  • package incorrectly calls plan() at the top-level causing it to be called during installation
  • issue filed (future::plan(multiprocess) to be deprecated; use 'multisession' / 'multicore' instead sjspielman/dragon#43) on 2020-10-03
  • check with dragon 1.0.4 (on CRAN as of 2020-10-25) => FAILED
    • reported that problem remains, cf. future::plan(multiprocess) to be deprecated; use 'multisession' / 'multicore' instead sjspielman/dragon#43 (comment) (2020-10-26)

• treeHMM 0.1.1 (2019-12-12)

  • package incorrectly calls plan() at the top-level causing it to be called during installation
  • maintainer has been emailed on 2020-10-03 and 2020-10-15

[HenrikBengtsson]

@DavisVaughan, with regard to your upcoming release, is my memory correct that you said you've already remove mentioning of 'multiprocess' in furrr in favor of either 'multisession' and 'multicore'?

[DavisVaughan]

I've added an explicit issue to ensure I remove it in favor of 'multisession' futureverse/furrr#169

If the globals release goes smoothly, I should be able to release furrr with this change this week

[HenrikBengtsson]

If the globals release goes smoothly, I should be able to release furrr with this change this week

Good news; I just received "Dear maintainer, thanks, package globals_0.13.1.tar.gz is on its way to CRAN."

[DavisVaughan]

furrr 0.2.0 is on CRAN! 🥳 https://cran.r-project.org/web/packages/furrr/index.html

[HenrikBengtsson]

Congrats. That's milestone.

(I didn't know you could submit before also binaries for globals 0.13.1 were available for all platforms - good to know that's possible)

[HenrikBengtsson]

After an explanation and some waiting time, CRAN accepted future 1.20.1 despite introducing a WARNING on 'dragon' and 'treeHMM'. Maintainers of those packages have been updated.