dplyr:::replace_with function

[jlegewie]

Moving this from tidyverse/dplyr#2040

I think dplyr:::replace_with can be useful and should be exported...
if_else uses it, coalesce uses it and probably other internal functions as well. It's a useful way to recode data even though similar things can be achieved with if_else.

It should work with typed NAs though but #2038 covers that, I think.

[nickresnick]

posting the ever annoying "+1" here. Would it be great to have for things like modified dplyr::coalesce for factors.

[DavisVaughan]

Was thinking about vec_coalesce() today so I figured I'd add this in

If we don't like using the common type and common size, then I think the signature should be vec_coalesce(x, ...) to indicate that we are using the size/type of x.

This is one place it would be nice for vec_slice <- to know how to not recycle length 1 inputs and instead use them repeatedly

library(rlang)
library(vctrs)

vec_coalesce <- function(..., .ptype = NULL) {
  args <- list2(...)
  
  n_args <- vec_size(args) 
  
  if (n_args == 0L) {
    return(NULL)
  }
  
  if (n_args == 1L) {
    out <- args[[1L]]
    return(out)
  }
  
  args <- vec_cast_common(!!! args, .to = .ptype)
  args <- vec_recycle_common(!!! args)
  
  out <- args[[1L]]
  args <- args[-1L]
  
  for (arg in args) {
    is_na <- vec_equal_na(out)
    
    if (!any(is_na)) {
      break
    }
    
    vec_slice(out, is_na) <- vec_slice(arg, is_na)
  }
  
  out
}

vec_coalesce()
#> NULL

vec_coalesce(1, 0)
#> [1] 1

vec_coalesce(1, FALSE, .ptype = logical())
#> [1] TRUE

vec_coalesce(NA, 1)
#> [1] 1

vec_coalesce(c(1, NA, 2), 0L)
#> [1] 1 0 2

vec_coalesce(
  data.frame(x = c(1, NA, 3)), 
  data.frame(x = 2)
)
#>   x
#> 1 1
#> 2 2
#> 3 3

# a bit odd, but this technically makes sense
vec_coalesce(
  data.frame(x = c(1, NA)), 
  data.frame(x = 2, y = 3)
)
#>   x  y
#> 1 1 NA
#> 2 2  3

vec_coalesce(
  factor(c("x", "y", NA, "x", NA)),
  factor("MISSING!")
)
#> [1] x        y        MISSING! x        MISSING!
#> Levels: x y MISSING!

# Common size is used as the reference
vec_coalesce(
  1,
  c(1, 2, 3)
)
#> [1] 1 1 1

vec_coalesce(
  NA,
  c(1, 2, 3)
)
#> [1] 1 2 3

[hadley]

I think this is just #15? No one has supplied any examples so it's hard to know.