[jira] [Commented] (ARROW-14297) [R] smooth out integer division to better match R

["Dewey Dunnington (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/ARROW-14297?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17441129#comment-17441129 ] 

Dewey Dunnington commented on ARROW-14297:
------------------------------------------

The place where we’re currently doing this conversion is here: <https://github.com/apache/arrow/blob/master/r/R/expression.R#L212-L223> 
Runnable version of the above: 

{code:R}
library(arrow, warn.conflicts = FALSE)
library(dplyr, warn.conflicts = FALSE)
library(testthat, warn.conflicts = FALSE)

a <- c(1:4, NA_integer_)
a_arrow <- Array$create(a)

expect_identical(as.vector(a_arrow %/% 2), a %/% 2)
#> Error: as.vector(a_arrow%/%2) not identical to a%/%2.
#> Objects equal but not identical

# this actually works, but on accident as in R the type of the denominator
# is the one that determines the type of the output 
expect_identical(as.vector(a_arrow %/% 2L), a %/% 2L)

expect_identical(as.vector(a_arrow %/% 0L), a %/% 0L)
#> Error: as.vector(a_arrow%/%0L) not identical to a%/%0L.
#> 4/5 mismatches (average diff: NaN)
#> [1] 2147483647 - NA == NA
#> [2] 2147483647 - NA == NA
#> [3] 2147483647 - NA == NA
#> [4] 2147483647 - NA == NA

b <- -a
b_arrow <- Array$create(b)
expect_identical(as.vector(b_arrow %/% 2), b %/% 2)
#> Error: as.vector(b_arrow%/%2) not identical to b%/%2.
#> 2/5 mismatches (average diff: 1)
#> [1]  0 - -1 == 1
#> [3] -1 - -2 == 1
expect_identical(as.vector(b_arrow %/% .2), b %/% .2)
#> Error: as.vector(b_arrow%/%0.2) not identical to b%/%0.2.
#> Objects equal but not identical
{code}


> [R] smooth out integer division to better match R
> -------------------------------------------------
>
>                 Key: ARROW-14297
>                 URL: https://issues.apache.org/jira/browse/ARROW-14297
>             Project: Apache Arrow
>          Issue Type: Improvement
>          Components: R
>            Reporter: Dragoș Moldovan-Grünfeld
>            Assignee: Dewey Dunnington
>            Priority: Major
>             Fix For: 7.0.0
>
>
> Behaviour of integer division from Arrow differs from R. 
> {code:r}
> > a <- c(1:4, NA_integer_)
> > a_arrow <- Array$create(a)
> > 
> > expect_identical(as.vector(a_arrow %/% 2), a %/% 2)
> Error: as.vector(a_arrow%/%2) (`actual`) not identical to a%/%2 (`expected`).
> `actual` is an integer vector (0, 1, 1, 2, NA)
> `expected` is a double vector (0, 1, 1, 2, NA)
> > 
> > # this actually works, but on accident as in R the type of the denominator
> > # is the one that determines the type of the output 
> > expect_identical(as.vector(a_arrow %/% 2L), a %/% 2L)
> > 
> > expect_identical(as.vector(a_arrow %/% 0), a %/% 0)
> Error: as.vector(a_arrow%/%0) (`actual`) not identical to a%/%0 (`expected`).
> `actual` is an integer vector (2147483647, 2147483647, 2147483647, 2147483647, NA)
> `expected` is a double vector (Inf, Inf, Inf, Inf, NA)
> > expect_identical(as.vector(a_arrow %/% 0L), a %/% 0L)
> Error: as.vector(a_arrow%/%0L) (`actual`) not identical to a%/%0L (`expected`).
>   `actual`: 2147483647 2147483647 2147483647 2147483647 NA
> `expected`:         NA         NA         NA         NA NA
> > 
> > b <- -a
> > b_arrow <- Array$create(b)
> > expect_identical(as.vector(b_arrow %/% 2), b %/% 2)
> Error: as.vector(b_arrow%/%2) (`actual`) not identical to b%/%2 (`expected`).
> `actual` is an integer vector (0, -1, -1, -2, NA)
> `expected` is a double vector (-1, -1, -2, -2, NA)
> > expect_identical(as.vector(b_arrow %/% .2), b %/% .2)
> Error: as.vector(b_arrow%/%0.2) (`actual`) not identical to b%/%0.2 (`expected`).
> `actual` is an integer vector (-5, -10, -15, -20, NA)
> `expected` is a double vector (-5, -10, -15, -20, NA)
> {code}
> expect_dplyr_equal() uses expect_equal() which does not enforce type.
> This is a follow-up issue to [ARROW-13800|https://issues.apache.org/jira/browse/ARROW-13800] 



--
This message was sent by Atlassian Jira
(v8.20.1#820001)