Consider making foo[[bar]] and foo[bar] never expand when bar is "simple"

[DavisVaughan]

See #183

i.e. with

private$event_callback_counts[[domain]] <- private$event_callback_counts[[domain]] - 1

it should probably never break on the [[ and expand to

private$event_callback_counts[[domain]] <- private$event_callback_counts[[
    domain
  ]] - 
  1

instead it should probably give this, even though the first line exceeds the 80 char limit

private$event_callback_counts[[domain]] <- private$event_callback_counts[[domain]] - 
  1

Related to #163 because we would not want line breaks in $ to force line breaks here!

We will have to nail down what "simple" means, probably:

• Keywords like TRUE or NULL
• Identifiers
• Numerics / integers / complexes
• etc

Having a list of what makes something "simple" will likely be useful in other places too?

It should probably also respect persistent line breaks if you manually request the expansion? That's quite an edge case though.

[DavisVaughan]

Now that we've implemented hugging in #228, I think that this may actually be an alternative type of hugging! The rule could be something like:

For [[ and [ calls with exactly 1 argument, the [[ and [ "hug" that argument, never fully expanding.

It's possible this is just a small extension to is_hugging_call()?

That would have helped in dplyr where we have

# before
{
  {
    get_current_group_id = function() {
      duplicate(private[["env_current_group_info"]][["dplyr:::current_group_id"]])
    }
  }
}

# after 
{
  {
    get_current_group_id = function() {
      duplicate(private[["env_current_group_info"]][[
        "dplyr:::current_group_id"
      ]])
    }
  }
}

# ideally
{
  {
    get_current_group_id = function() {
      duplicate(
        private[["env_current_group_info"]][["dplyr:::current_group_id"]]
      )
    }
  }
}

And also

# before
env_current_group_info[["dplyr:::current_group_size"]] <- duplicate(length(private$rows[[group]]))

# after
env_current_group_info[[
  "dplyr:::current_group_size"
]] <- duplicate(length(private$rows[[group]]))

# ideally
env_current_group_info[["dplyr:::current_group_size"]] <- duplicate(length(
  private$rows[[group]]
)) 

Note that as soon as you go past 1 argument, I do feel that not hugging is probably still better (i.e. the current behavior)

# before
env_current_group_info[["dplyr:::current_group_size", a_column]] <- duplicate(length(private$rows[[group]]))

# after
env_current_group_info[[
  "dplyr:::current_group_size",
  a_column
]] <- duplicate(length(private$rows[[group]]))

[lionel-]

Now that we've implemented hugging in #228, I think that this may actually be an alternative type of hugging!

Agreed and here is another case that feels very related: #257

[DavisVaughan]

It won't help box much though because we'd probably only apply this to single argument cases, and it seems very common to import multiple things in the [ ], like box::use(a = ./a[f, g]), in which case you'd get expansion again

[lionel-]

I only meant to point out another way the hugging notion can be generalised