coalesce() silently coerces NaN to NA

[thisisnic]

I was fixing some Arrow test failures which are the result of the dev version of waldo correctly distinguishing between NA_real_ and NaN, and came across this behaviour where NaN values are silently coerced by dplyr::coalesce() into NAs.

I'm unsure if this is a bug or just a difference in implementations/interpretations, but the second row in the cwx column in the example below is making me think it could be a bug?

library(dplyr)
library(arrow)

df <- tibble(
  w = c(NaN, NaN, NA_real_),
  x = c(NA_real_, NaN, 3.3),
  y = c(NA_real_, 2.2, 3.3),
  z = c(1.1, 2.2, 3.3)
)

df %>%
  mutate(
    cw = coalesce(w),
    cz = coalesce(z),
    cwx = coalesce(w, x),
    cwxy = coalesce(w, x, y),
    cwxyz = coalesce(w, x, y, z)
  )
#> # A tibble: 3 × 9
#>       w     x     y     z    cw    cz   cwx  cwxy cwxyz
#>   <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1   NaN  NA    NA     1.1    NA   1.1  NA    NA     1.1
#> 2   NaN NaN     2.2   2.2    NA   2.2  NA     2.2   2.2
#> 3    NA   3.3   3.3   3.3    NA   3.3   3.3   3.3   3.3

df %>%
  arrow_table() %>%
  mutate(
    cw = coalesce(w),
    cz = coalesce(z),
    cwx = coalesce(w, x),
    cwxy = coalesce(w, x, y),
    cwxyz = coalesce(w, x, y, z)
  ) %>%
  collect()
#> # A tibble: 3 × 9
#>       w     x     y     z    cw    cz   cwx  cwxy cwxyz
#>   <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1   NaN  NA    NA     1.1   NaN   1.1  NA    NA     1.1
#> 2   NaN NaN     2.2   2.2   NaN   2.2 NaN     2.2   2.2
#> 3    NA   3.3   3.3   3.3    NA   3.3   3.3   3.3   3.3

^{Created on 2023-04-24 with reprex v2.0.2}

[hadley]

is.na(NaN) is TRUE so I think this is probably correct?

[thisisnic]

At any rate, it's not causing anyone actual issues as far as I'm aware, so I'll close this