[R] month() should allow integer inputs

[asfimport]

Conclusion: we will implement this in the R bindings - month will allow integer input: month(int) will return int as long as int is between 1 and 12.

In R, more specifically in {}lubridate{}, month() can be used both to get and set the corresponding component of a date. This means month() accepts integer inputs.

suppressPackageStartupMessages(library(lubridate)) 
month(1:12)
#>  [1]  1  2  3  4  5  6  7  8  9 10 11 12
month(1:13)
#> Error in month.numeric(1:13): Values are not in 1:12

Solving this would allow us to implement bindings such as semester() in a manner closer to {}lubridate{}.

suppressPackageStartupMessages(library(dplyr))
suppressPackageStartupMessages(library(lubridate))

test_df <- tibble(
  month_as_int = c(1:12, NA),
  month_as_char_pad = ifelse(month_as_int < 10, paste0("0", month_as_int), month_as_int),
  dates = as.Date(paste0("2021-", month_as_char_pad, "-15"))
)

test_df %>%
  mutate(
    sem_date = semester(dates),
    sem_month_as_int = semester(month_as_int))
#> # A tibble: 13 × 5
#>    month_as_int month_as_char_pad dates      sem_date sem_month_as_int
#>           <int> <chr>             <date>        <int>            <int>
#>  1            1 01                2021-01-15        1                1
#>  2            2 02                2021-02-15        1                1
#>  3            3 03                2021-03-15        1                1
#>  4            4 04                2021-04-15        1                1
#>  5            5 05                2021-05-15        1                1
#>  6            6 06                2021-06-15        1                1
#>  7            7 07                2021-07-15        2                2
#>  8            8 08                2021-08-15        2                2
#>  9            9 09                2021-09-15        2                2
#> 10           10 10                2021-10-15        2                2
#> 11           11 11                2021-11-15        2                2
#> 12           12 12                2021-12-15        2                2
#> 13           NA <NA>              NA               NA               NA

Currently attempts to use month() with integer inputs errors with:

Function 'month' has no kernel matching input types (array[int32])

Reporter: Dragoș Moldovan-Grünfeld / @dragosmg
Assignee: Dragoș Moldovan-Grünfeld / @dragosmg
Watchers: Rok Mihevc / @rok

Related issues:

• [R] Implement bindings for lubridate::semester (is depended upon by)

PRs and other links:

• GitHub Pull Request #12482

_{Note: This issue was originally created as ARROW-15701. Please see the migration documentation for further details.}

[asfimport]

Joris Van den Bossche / @jorisvandenbossche:
Can you clarify what the expected output would be for the months kernel with integer input (so eg for month(1))?

For being able to set the component (such as the month) of a timestamp, that could be useful I think (Python also has this concept as a "replace" method, eg datetime.replace(month=12)), but for Arrow I think that would be a separate kernel.

[asfimport]

Dragoș Moldovan-Grünfeld / @dragosmg:
I think month(a) should return a when a is between 1 and 12 and error otherwise. 

[asfimport]

Rok Mihevc / @rok:
Month is currently just an extraction kernel for timestamps and dates, e.g.: month(timestamp[unit]) -> int32. Arrow does not have a "month array" concept so month(int) is expected to fail. Are you expecting to do month(int) -> duration?

[asfimport]

Jonathan Keane / @jonkeane:
This also should be accomplishable in the R bindings if we want to do it there quickly to unblock things before we decide/wait for the C++ implementation.

Basically, somewhere in

arrow/r/R/dplyr-funcs-datetime.R

Lines 108 to 119 in 1b9e76c

	register_binding("month", function(x, label = FALSE, abbr = TRUE, locale = Sys.getlocale("LC_TIME")) {
	if (label) {
	if (abbr) {
	format <- "%b"
	} else {
	format <- "%B"
	}
	return(Expression$create("strftime", x, options = list(format = format, locale = locale)))
	}
	Expression$create("month", x)
	})

check the input type and do casting or follow a slightly different path and return an expression that creates the thing we need.

[asfimport]

Dragoș Moldovan-Grünfeld / @dragosmg:
Not necessarily month(int) -> duration, but in R, in general (and lubridate in particular), quite a bunch of functionality goes through something like month (which is a slot in the list returned by POSIXlt). In my case the implementation of semester extraction goes through month(), and also supports integer inputs (in addition to dates and date times).
I haven't started looking into the duration type in lubridate - there lubridate has dmonths which could be used to create a duration object corresponding to a timespan of x months.

[asfimport]

Joris Van den Bossche / @jorisvandenbossche:

I think month(a) should return a when a is between 1 and 12 and error otherwise.

In that case I suppose that is also easily achievable on the R side? (without complicating the kernel in C++)

[asfimport]

Dragoș Moldovan-Grünfeld / @dragosmg:
I think so too - that was the reason behind having the title of the ticket as a question. If this is relevant mostly for R, we could implement it in the binding for month(). Maybe I should have made it a bit clearer that it is a question rather than an affirmation.

[asfimport]

Dragoș Moldovan-Grünfeld / @dragosmg:
I will edit this issue to refer to R only and I'll look into solving it in the R bindings. Thanks all for your input.

[asfimport]

Jonathan Keane / @jonkeane:
Issue resolved by pull request 12482
#12482