`redundant_ifelse_linter()` ignores `missing` arg

[jmbarbone]

The redundant_ifelse_linter() ignores dplyr::if_else()'s missing arg and suggests a correction which would change the results.

packageVersion("lintr")
#> [1] '3.0.2.9000'
text <- "
list(
  dplyr::if_else(c(TRUE, FALSE, NA), TRUE, FALSE),
  dplyr::if_else(c(TRUE, FALSE, NA), TRUE, FALSE, TRUE),
  dplyr::if_else(c(TRUE, FALSE, NA), TRUE, FALSE, FALSE),
  dplyr::if_else(c(TRUE, FALSE, NA), 1, 0),
  dplyr::if_else(c(TRUE, FALSE, NA), 1, 0, 2)
)
"

lintr::lint(text = text, linters = lintr::redundant_ifelse_linter())
#> <text>:3:3: warning: [redundant_ifelse_linter] Just use the logical condition (or its negation) directly instead of calling if_else(x, TRUE, FALSE)
#>   dplyr::if_else(c(TRUE, FALSE, NA), TRUE, FALSE),
#>   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#> <text>:4:3: warning: [redundant_ifelse_linter] Just use the logical condition (or its negation) directly instead of calling if_else(x, TRUE, FALSE)
#>   dplyr::if_else(c(TRUE, FALSE, NA), TRUE, FALSE, TRUE),
#>   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#> <text>:5:3: warning: [redundant_ifelse_linter] Just use the logical condition (or its negation) directly instead of calling if_else(x, TRUE, FALSE)
#>   dplyr::if_else(c(TRUE, FALSE, NA), TRUE, FALSE, FALSE),
#>   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#> <text>:6:3: warning: [redundant_ifelse_linter] Prefer as.numeric(x) to if_else(x, 1, 0) if really needed.
#>   dplyr::if_else(c(TRUE, FALSE, NA), 1, 0),
#>   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#> <text>:7:3: warning: [redundant_ifelse_linter] Prefer as.numeric(x) to if_else(x, 1, 0) if really needed.
#>   dplyr::if_else(c(TRUE, FALSE, NA), 1, 0, 2)
#>   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
eval(parse(text = text))
#> [[1]]
#> [1]  TRUE FALSE    NA
#> 
#> [[2]]
#> [1]  TRUE FALSE  TRUE
#> 
#> [[3]]
#> [1]  TRUE FALSE FALSE
#> 
#> [[4]]
#> [1]  1  0 NA
#> 
#> [[5]]
#> [1] 1 0 2

^{Created on 2023-03-30 with reprex v2.0.2}

[MichaelChirico]

I find that code quite hard to read... I would prefer skipping a lint if missing= is supplied by name, rather than positionally. But it's not good to recommend incorrect code even in the positional case.

I would also prefer coalesce(LOGICAL, DEFAULT) or tidyr::replace_na(LOGICAL, DEFAULT) over ifelse(LOGICAL, TRUE, FALSE, DEFAULT).

The case is less clear for coalesce(as.integer(LOGICAL), DEFAULT) vs. ifelse(LOGICAL, 1L, 0L, DEFAULT)

[MichaelChirico]

Also noting that dplyr::if_else() does type promotion via missing=:

dplyr::if_else(c(TRUE, FALSE, NA), TRUE, FALSE, 5)
# [1] 1 0 5

(well, sometimes:)

dplyr::if_else(x > 5, 1, 0, NA_character_)
# Error in `dplyr::if_else()`:
# ! Can't combine `true` <double> and `missing` <character>.
# Run `rlang::last_trace()` to see where the error occurred.