Why does this create an "NAs introduced by coercion" warning

Question

Curious why the following produces an "NAs introduced by coercion" warning

# Example dataframe
df <- tibble(
  session = c("a",2),
)

df %>%
  mutate(sessionNum = case_when(
    session == "a" ~ 1,
    TRUE ~ as.numeric(session)
  ))

I thought there'd be no need for coercing anything into an "NA" as "a" is covered by the first case_when.

Even the following dataframe produces the warning!

# Example dataframe
df <- tibble(
  session = c("a"),
)

Answer 1

It seems that dplyr's if_else() and case_when() evaluate the RHS regardless if the condition is TRUE or not. According to Hadley Wickham, this is needed for type stability.

https://github.com/tidyverse/dplyr/issues/5321

Since the RHS is being evaluated for all values you are coercing a character type into a numeric type (i.e., as.numeric(c('a', '2'))) resulting in the warning.

Also noted here: https://github.com/tidyverse/dplyr/issues/5341

Answer 2

I believe it's because case_when() is vectorized on the result.

Taking a slightly simpler example:

library(dplyr)
session <- c("a", 2)

session
#> [1] "a" "2"

case_when(
  session == "a" ~ 1,
  .default = as.numeric(session)
)
#> Warning in vec_case_when(conditions = conditions, values = values,
#> conditions_arg = "", : NAs introduced by coercion
#> [1] 1 2

This is roughly equivalent to the base R code:

session <- c("a", 2)
condition <- session == "a"
result_if_true <- 1

result_if_false <- as.numeric(session)
#> Warning: NAs introduced by coercion


result <- vector(length = length(session))

result[condition] <- result_if_true[condition]
result[!condition] <- result_if_false[!condition]
result
#> [1] 1 2

So, even though as.numeric("a") is not present in the final result, it is computed in an intermediate step.