Empty selections shouldn't error with `allow_rename = FALSE`

[DavisVaughan]

tidyselect::eval_select(character(), mtcars, allow_rename = TRUE)
#> named integer(0)

tidyselect::eval_select(character(), mtcars, allow_rename = FALSE)
#> Error: Can't rename variables in this context.

^{Created on 2020-11-04 by the reprex package (v0.3.0.9001)}

[eutwt]

Would adding && length(pos) > 0 to the ensure_named check as below cause problems elsewhere? If not I think that would be a simple way to close this issue. Seems like it doesn't cause any test failures.

ensure_named <- function(pos, vars, uniquely_named, allow_rename) {
  if (!allow_rename) {
    if (is_named(pos) && length(pos) > 0) {
      abort("Can't rename variables in this context.")
    }
    return(set_names(pos, NULL))
  }
...

[DavisVaughan]

Slightly more rlang like would be is_named(pos) && !is_empty(pos) (same idea though).

I do think that would fix it, since the main issue is how is_named() treats empty named vectors vs vectors with "" names.

library(rlang)

x <- character()
y <- "mpg"

tidyselect::eval_select(x, mtcars, allow_rename = FALSE)
#> Error: Can't rename variables in this context.
tidyselect::eval_select(y, mtcars, allow_rename = FALSE)
#> [1] 1

# Why are the above two different?
# Internally this is done to both
x <- set_names(x, names2(x))
y <- set_names(y, names2(y))

# empty, but named
x
#> named character(0)

# only "" names, which we don't interpret as being "named" in `is_named()`
y
#>       
#> "mpg"

rlang::is_named(x)
#> [1] TRUE
rlang::is_named(y)
#> [1] FALSE

[eutwt]

Yeah. The issue is that the names vector of x exists (not is.null) and doesn't have any invalid elements (is.na or == ""). It's just a length-0 vector. Its indistinguishable from a properly-named vector that's been emptied. So I think it's reasonable for rlang to say that x is a named vector.

x <- setNames(1, '')[0]
names(x)
#> character(0)

y <- setNames(1, 'y')[0]
names(y)
#> character(0)

z <- 1[0]
names(z)
#> NULL

^{Created on 2021-08-23 by the reprex package (v2.0.1)}

It's just that within the logic of eval_select, having a named empty vector doesn't imply the user is trying to rename variables. It wasn't named by the user, it's only named because it was passed to named by loc_validate.