dplyr::if_else changes datetime (POSIXct) values - r

Question

I'm working with a dataset that has a lot of timestamps. There are some invalid timestamps which I try to identify and set to NA. Because if_else() forces me to have the same data type in both arms, I'm using as.POSIXct(NA) to encode such missing values.
Interestingly, the results differ when I invert the test (and change the true and false argument) in if_else().
Here is some code to illustrate my problems:
x <- tibble(
A = parse_datetime("2020-08-18 19:00"),
B = if_else(TRUE, A, as.POSIXct(NA)),
C = if_else(FALSE, as.POSIXct(NA), A)
)
> x
# A tibble: 1 x 3
A B C
<dttm> <dttm> <dttm>
1 2020-08-18 19:00:00 2020-08-18 19:00:00 2020-08-18 21:00:00
Any idea, why C is two hours later?
Follow-up:
Based on the great answers below, I think a more readable solution should perhaps generate a missing datetime object with parse_datetime(NA_character_) and use this in the code instead of as.POSIXct().
R> NA_datetime_ <- parse_datetime(NA_character_)
R> x <- tibble(
A = parse_datetime("2020-08-18 19:00"),
B = if_else(TRUE, A, NA_datetime_),
C = if_else(FALSE, NA_datetime_, A)
)
R> map(x, lubridate::tz)
$A
[1] "UTC"
$B
[1] "UTC"
$C
[1] "UTC"

Answer 1

At First, you need to know that parse_datetime() returns a date-time object with an tzone attribute default to UTC. You can use lubridate::tz(x$A) and attributes(x$A) to check it.
From the document of if_else(), it said the true and false arguments must be the same type. All other attributes are taken from true. Hence, in part C of your tibble:
C = if_else(FALSE, as.POSIXct(NA), A)
as.POSIXct(NA) doesn't have a tzone attribute, so A's tzone is dropped and reset to the time zone of your region. Actually, C is not two hours later. The three columns have equal time but unequal time zones. To fix it, you can adjust as.POSIXct(NA) to own a tzone attribute, i.e. replace it with
as.POSIXct(NA_character_, tz = "UTC")
Note: You must use NA_character_ instead of NA because the tz argument in as.POSIXct() only works on character objects.
Finally, revise your code as
x <- tibble(
A = parse_datetime("2020-08-18 19:00"),
B = if_else(TRUE, A, as.POSIXct(NA_character_, tz = "UTC")),
C = if_else(FALSE, as.POSIXct(NA_character_, tz = "UTC"), A)
)
# # A tibble: 1 x 3
# A B C
# <dttm> <dttm> <dttm>
# 1 2020-08-18 19:00:00 2020-08-18 19:00:00 2020-08-18 19:00:00
Remember to check their time zones.
R > lubridate::tz(x$A)
[1] "UTC"
R > lubridate::tz(x$B)
[1] "UTC"
R > lubridate::tz(x$C)
[1] "UTC"

Answer 2

This is a timezone problem :
lubridate::tz(x$A)
[1] "UTC"
lubridate::tz(x$B)
[1] "UTC"
lubridate::tz(x$C)
[1] ""
This is due to the way if_else <- function (test, yes, no) works : it uses the attributes of the yes argument which for C is NA.