Selecting -NULL

[lionel-]

https://stackoverflow.com/questions/57590111/tidy-evaluation-adding-an-argument-to-a-function-that-prevents-gathering

[hadley]

I'm not sure this is needed given that you can use integer():

tidyselect::vars_select(names(mtcars), integer())
#> named character(0)
tidyselect::vars_select(names(mtcars), -integer())
#>    mpg    cyl   disp     hp   drat     wt   qsec     vs     am   gear 
#>  "mpg"  "cyl" "disp"   "hp" "drat"   "wt" "qsec"   "vs"   "am" "gear" 
#>   carb 
#> "carb"

^{Created on 2019-09-11 by the reprex package (v0.3.0)}

[lionel-]

This would be nice for NULL defaults in function arguments. We have done similar changes elsewhere to make it easier to interface with NULL defaults (facets in ggplot IIRC).

[hadley]

I think making -NULL work is that way would be in conflict with tidyverse/design#24, and I don't think this case is compelling enough to violate the principle (given that there are two existing work arounds).

If we just want to solve the default value problem, I'd be in favour of:

nothing <- function() integer()

And adding a couple of tests to ensure that this behaviour is locked in.

[lionel-]

I was thinking about the principle of treating NULL as an absent input.

Though I guess this principle would favour making this work actually:

vars_select(letters, -one_of(NULL))

Edit: hmm that wouldn't help OP since that's for character vectors, not quoted inputs.

[lionel-]

I like nothing() but is it worth it to use up this symbol for such a corner case? Maybe it could also be used for teaching about vectors in R, e.g. letters[nothing()] creates a prototype.

Let's keep thinking about this, in the meantime I will recommend your solution of using integer() for default arguments. I'll also make one_of(NULL) be equivalent to one_of().

[hadley]

Sounds good to me.