`rename_with()` doesn't handle factors well

[jmbarbone]

rename_with.data.frame() has unexpected results when .fn returns a factor -- or at least unexpected for me. Documentation does suggest that .fn should return a character vector, but I wouldn't expect factors to be coerced to integers first.

The coercion occurs within [<-, which could be replaced with vec_slice()?

dplyr/R/rename.R

Line 69 in 434caa2

names[cols] <- .fn(names[cols], ...)

df <- data.frame(
  a = 1,
  b = 2,
  c = 3
)

foo <- function(x) {
  factor(toupper(x))
}

dplyr::rename_with(df, foo)
#>   1 2 3
#> 1 1 2 3


# name reassignment within rename_with.data.frame()
x <- c("a", "b", "c")
x[1:3] <- foo(x)
x
#> [1] "1" "2" "3"


# vctrs seems fine
x <- c("a", "b", "c")
vctrs::vec_slice(x, 2:3) <- foo(x[2:3])
x
#> [1] "a" "B" "C"

^{Created on 2022-11-23 with reprex v2.0.2}

[hadley]

We should probably just error if you return anything other than a string.