Discourage <<- ?

[MichaelChirico]

{lintr} has long discouraged <<- as a default "undesirable operator":

https://github.com/r-lib/lintr/blob/50d9430173a9ed07570956e2a1de01c82fea9d93/R/zzz.R#L218-L223

Somehow I'd thought the Tidyverse guide also makes this ruling, but just found myself checking & realized there's no opinion.

I think avoiding <<- is good practice. The picture-perfect ideal usage looks like this:

w <- e <- NULL
tryCatch(
  foo(),
  warning = \(c) w <<- c,
  error = \(c) e <<- c
)

Here the target of <<- is maximally close in the code to where <<- is used. In any other case, it starts requiring a large mental load on the code reader to keep track of in which environment the LHS of <<- actually ends up getting assigned, and mistakes can cause really unexpected and hard-to-debug behavior (x <<- ..., where we meant y <<- ..., might wind up wiping out something important in some other environment).

These days I strongly encourage authors to be sometimes painfully explicit instead:

w <- e <- NULL
env <- environment()
tryCatch(
  foo(),
  warning = \(c) env$w <- c,
  error = \(c) env$e <- c
)

The code is slightly messier (we often wind up needing many more env$ qualifiers) but often easier to understand and maintain.

PS
The guide used to discourage pipe "in-place" assignment operator %<>%:

style/pipes.Rmd

Line 129 in 8de4bbc

The magrittr package provides the `%<>%` operator as a shortcut for modifying an object in place. Avoid this operator.

This advice was thrown out along with the move to only use the native pipe |> (#225). I believe the motivation for that advice would be similar to the proposed ruling here.