as.period fails for grouped data

[julianbarg]

as.period fails with a nondescript error when applied to a grouped dataframe.

df <- data.frame(group = c("a","a","b"),
                 p1 = as.Date(c("2020-10-23", "2020-10-24", "2020-10-25")), 
                 p2 = as.Date(c("2020-10-26", "2020-10-27", "2020-10-28")))

#This succeeds
df %>%
  mutate(interval = as.period(interval(start = p1, end = p2)))

# This throws the error message
df %>%
  group_by(group) %>%
  mutate(interval = as.period(interval(start = p1, end = p2)))

Error message:
Error: Problem with mutate() input interval.
x Internal error in vec_proxy_assign_opts(): proxy of type double incompatible with value proxy of type integer.
ℹ Input interval is as.period(interval(start = p1, end = p2)).
ℹ The error occurred in group 2: group = "b".

[vspinu]

@DavisVaughan you have added support for vectrs. Is this issue expected?

[DavisVaughan]

I think this is a lubridate bug? I think that year/month/day/hour/minute components of a Period object should always be integer, and the second component should be numeric. But that is not the case. Creating a period object from as.period() maintains this property, but creating one from period() or a helper like days() does not. That causes the problem.

library(lubridate)
#> 
#> Attaching package: 'lubridate'
#> The following objects are masked from 'package:base':
#> 
#>     date, intersect, setdiff, union
library(vctrs)

x <- as.Date("2020-10-25")
interval <- interval(x, x + 1)
period_from_interval <- as.period(interval)

period <- days(3)

get_fields <- function(x) {
  list(
    year = x@year, 
    month = x@month, 
    day = x@day, 
    hour = x@hour, 
    minute = x@minute, 
    second = x@.Data
  )
}

# this feels correct
vapply(get_fields(period_from_interval), class, character(1))
#>      year     month       day      hour    minute    second 
#> "integer" "integer" "integer" "integer" "integer" "numeric"

# this feels wrong?
vapply(get_fields(period), class, character(1))
#>      year     month       day      hour    minute    second 
#> "numeric" "numeric" "numeric" "numeric" "numeric" "numeric"

# the shared empty period object is just created by:
# `delayedAssign("lubridate_shared_empty_period", period())`
vapply(get_fields(lubridate:::lubridate_shared_empty_period), class, character(1))
#>      year     month       day      hour    minute    second 
#> "numeric" "numeric" "numeric" "numeric" "numeric" "numeric"

# we expect the same types as `lubridate_shared_empty_period`, so this errors
vec_c(period_from_interval, period_from_interval)
#> Error: Internal error in `vec_proxy_assign_opts()`: `proxy` of type `double` incompatible with `value` proxy of type `integer`.

# this has the same type as `lubridate_shared_empty_period`, so no error
vec_c(period, period)
#> [1] "3d 0H 0M 0S" "3d 0H 0M 0S"

^{Created on 2020-11-05 by the reprex package (v0.3.0.9001)}

Maybe these should be integers (except seconds)

lubridate/R/periods.r

Lines 449 to 450 in b9025e6

	out <- list(second = 0, minute = 0, hour = 0, day = 0,
	week = 0, month = 0, year = 0)

And day math here might should use 7L

lubridate/R/periods.r

Line 457 in b9025e6

out$day <- out$day + 7 * out$week

An even more robust way to handle this is probably to handle it in the initialization function for Period. Let validObject() first do its checks to make sure the values are integer-ish, then after that actually convert the components to integer

lubridate/R/periods.r

Line 174 in b9025e6

validObject(.Object)