[R] Implement bindings for lubridate calculation functions

[asfimport]

Reporter: Nicola Crane / @thisisnic

Related issues:

• [R] lubridate functions for 9.0.0 (is a child of)

_{Note: This issue was originally created as ARROW-14820. Please see the migration documentation for further details.}

[asfimport]

SHIMA Tatsuya / @eitsupi:
Hi, I was wondering if it was possible to convert timestamp to date in Arrow's dplyr pipeline.

Currently, the following seems to work the same as.Date and as_date, but it does not be able to set a timezone.
It would be great if bindings for these functions could be provided.

https://lubridate.tidyverse.org/reference/as_date.html

library(arrow, warn.conflicts = FALSE)
library(dplyr, warn.conflicts = FALSE)
library(lubridate)
#>
#> Attaching package: 'lubridate'
#> The following objects are masked from 'package:base':
#>
#>     date, intersect, setdiff, union

df <- tibble::tibble(
  col1 = lubridate::as_datetime("2010-08-03 00:50:50", tz = "Europe/London"),
)

df |>
  arrow_table() |>
  mutate(
    col2 = cast(cast(col1, arrow::timestamp()), arrow::date32()), # lubridate::as_date()
    col3 = cast(col1, arrow::date32()) # as.Date()
  ) |>
  collect()
#> # A tibble: 1 × 3
#>   col1                col2       col3
#>   <dttm>              <date>     <date>
#> 1 2010-08-03 00:50:50 2010-08-02 2010-08-03

[asfimport]

Dragoș Moldovan-Grünfeld / @dragosmg:
Hi @eitsupi,

Thanks for filing the ticket. Once PR 12433 is merged you will be able to use date() to extract the date component of a timestamp. We aim to cover as.Date() as part of the  same pull request.

The snippet of code below should work once we merge the PR. Please note base::as.Date() by default assumes you want the date in UTC :

library(dplyr)
library(lubridate)
library(arrow)

df <- tibble::tibble(
  col1 = lubridate::as_datetime("2010-08-03 00:50:50", tz = "Europe/London"),
)

# lubridate
df %>% 
  mutate(x = date(col1),
         y = as.Date(col1))
#> # A tibble: 1 × 3
#>   col1                x          y         
#>   <dttm>              <date>     <date>    
#> 1 2010-08-03 00:50:50 2010-08-03 2010-08-02

# arrow
df %>% 
  arrow_table() %>% 
  mutate(x = date(col1),
         y = as.Date(col1)) %>% 
  collect()
#> # A tibble: 1 × 3
#>   col1                x          y         
#>   <dttm>              <date>     <date>    
#> 1 2010-08-03 00:50:50 2010-08-03 2010-08-02

[asfimport]

SHIMA Tatsuya / @eitsupi:
Thank you for working on this, I wasn't aware of the function date(), but I think this could satisfy a lot of use cases once it is implemented.

I also think that if the usage of cast() is documented as I have written, users will be able to use cast() to achieve more complex cases.

[asfimport]

Dragoș Moldovan-Grünfeld / @dragosmg:
@eitsupi  I opened a separate ticket (ARROW-15814) to improve the documentation for {}cast(){}. I would also like to encourage you to contribute if you think that's something you might want to do. 
On the timezone aspect, that is slightly trickier, Arrow currently transforms the timestamp into UTC and any timezone is store as metadata only. As such, casting with a different timezone (as the base R behaviour) is not immediately possible. There are several tickets (for example, ARROW-15806) looking into this and hopefully this will be solved sometime in the near(ish) future.