dplyr_col_select() reconstructs bare data.table objects

[romainfrancois]

# pak::pak("Rdatatable/data.table")
library(data.table)
library(dplyr, warn.conflicts = FALSE)

d1 <- data.table(x = 1)
attr(d1, "foo") <- "bar"

d2 <- dplyr:::dplyr_col_select(d1, "x")
attr(d2, "foo")
#> [1] "bar"

^{Created on 2022-02-02 by the reprex package (v2.0.1)}

as opposed to losing the attribute:

# pak::pak("Rdatatable/data.table")
library(data.table)
library(dplyr, warn.conflicts = FALSE)

d1 <- data.table(x = 1)
attr(d1, "foo") <- "bar"

d2 <- dplyr:::dplyr_col_select(d1, "x")
attr(d2, "foo")
#> NULL

^{Created on 2022-02-02 by the reprex package (v2.0.1)}

however, this only works with the dev version of data.table thanks to: Rdatatable/data.table#4909

This is related to: fndemarqui/covid19br#1

[DavisVaughan]

I am quite certain we don't want this to be data.table aware. The [i] method of data table selects rows rather than columns.

library(data.table)

# yay 32 rows, 0 cols
mtcars[0]
#> data frame with 0 columns and 32 rows

# 0 rows, 11 cols
as.data.table(mtcars)[0]
#> Empty data.table (0 rows and 11 cols): mpg,cyl,disp,hp,drat,wt...

Try this with your two examples:

library(data.table)
library(dplyr, warn.conflicts = FALSE)

d1 <- data.table(x = 1:2, y = 1:2)
attr(d1, "foo") <- "bar"

d2 <- dplyr:::dplyr_col_select(d1, "x")
d2

[romainfrancois]

oh right. So maybe this is rather something to do in:

  if (identical(class(.data), "data.frame")) {
    out <- dplyr_reconstruct(out, .data)
  }

[romainfrancois]

Because this ends up calling [.data.frame directly then: https://github.com/Rdatatable/data.table/blob/8b257b8c2f638567c2f7166f6bc3e086cf4195c2/R/data.table.R#L146

  if (!cedta()) {
    # Fix for #500 (to do)
    Nargs = nargs() - (!missing(drop))
    ans = if (Nargs<3L) { `[.data.frame`(x,i) }  # drop ignored anyway by DF[i]
        else if (missing(drop)) `[.data.frame`(x,i,j)
        else `[.data.frame`(x,i,j,drop)
    # added is.data.table(ans) check to fix bug #81
    if (!missing(i) && is.data.table(ans)) setkey(ans, NULL)  # drops index too; tested by plyr::arrange test in other.Rraw
    return(ans)
  }

[DavisVaughan]

Yea as you've seen I think the issue is basically that in non-data.table aware paths they just hand off to [.data.frame which doesn't retain attributes

[DavisVaughan]

Might be good to also mention the data.table handling in the comment and link to this PR

[romainfrancois]

mention this pr in comment