leap second for 1969?

[mcfrank]

Hi lubridate folks, I found something odd when rounding a time series around lubridate::origin: in particular, it seems like rounding introduces some kind of a leap second between 1969 and 1970 (see plot).

Take a look at this example:

t <- seq(-10000,10000,33)/1000
t_date <- lubridate::origin + t
t_date_rounded <- lubridate::round_date(t_date, ".033s")
plot(t_date,t_date_rounded)

It doesn't seem like the ideal behavior of this rounding should be to introduce the leap second? Or maybe I'm just not understanding this functionality? Thanks for any help.

[hadley]

This is another bug fixed by the timechange package:

library(lubridate, warn.conflicts = FALSE)

t <- seq(-10000, 10000, 33) / 1000
t_date <- origin + t
plot(t_date, round_date(t_date, ".033s"))

plot(t_date, timechange::time_round(t_date, ".033 asec"))

^{Created on 2019-11-19 by the reprex package (v0.3.0)}

[DavisVaughan]

This also works correctly in clock. You have to use a clock type, which truly supports sub second precision date-times.

library(clock)
library(magrittr)

t <- seq(-10000,10000,33)/1000

# Approximately split into seconds/milliseconds
secs <- floor(t)
millisecs <- round((t - secs) * 1000)

x <- year_month_day(1970, 1, 1) %>%
  as_naive_time() %>%
  add_seconds(secs) %>%
  add_milliseconds(millisecs)

x_floor <- time_point_floor(x, "millisecond", n = 33)
head(x_floor)
#> <time_point<naive><millisecond>[6]>
#> [1] "1969-12-31 23:59:49.968" "1969-12-31 23:59:50.001"
#> [3] "1969-12-31 23:59:50.034" "1969-12-31 23:59:50.067"
#> [5] "1969-12-31 23:59:50.100" "1969-12-31 23:59:50.133"

y <- x_floor %>%
  as_duration() %>%
  as.double()

plot(t, y)

^{Created on 2021-05-25 by the reprex package (v1.0.0)}

[vspinu]

It's working correctly in devel