as_tibble.data.table() method?

[ccsarapas]

(I originally requested this at tidyverse/tibble/issues/1555, but they suggested requesting here instead.)

Would you consider adding a data.table method to as_tibble()? On extremely cursory testing, it may be as simple as adapting as_tibble.data.frame() with an extra line to remove the .internal.selfref attribute:

library(tibble)
library(data.table)
data(mpg, package = "ggplot2")

# two identical tibbles
mpg2 <- copy(mpg)
identical(mpg, mpg2)
#> [1] TRUE

# after converting to data.table and back to tibble, no longer identical
setDT(mpg2)
mpg2 <- as_tibble(mpg2)
identical(mpg, mpg2)
#> [1] FALSE

# identical again after removing ".internal.selfref" attribute
attr(mpg2, ".internal.selfref") <- NULL
identical(mpg, mpg2)
#> [1] TRUE

^{Created on 2023-09-22 with reprex v2.0.2}

[jangorecki]

Adding such method means extending dependencies of DT. We rather would like to reduce them, also Suggested dependencies. In this case there are many recursive dependencies as well.

Is there a reason why as_tibble.data.frame method doesn't work?
going via data.frame should work after looking at as.data.frame.data.table and what is under

data.table/R/data.table.R

Line 2799 in 8803918

setattr(x, ".internal.selfref", NULL)

---

I believe there reason is presented by this minimal example, which doesn't need tibble to reproduce:

library(data.table)
data(mpg, package = "ggplot2")

mpg2 <- copy(mpg)
identical(mpg, mpg2)
#> [1] TRUE

setDT(mpg2)
mpg2 = as.data.frame(mpg2)
identical(mpg, mpg2)
#> [1] FALSE

[MichaelChirico]

Isn't this the Enhances: dependency, and isn't the direction reversed, i.e., {data.table} would provide as.data.table.tibble() and {tibble} would provide as_tibble.data.table()?

[jangorecki]

So the problem is that initial dataset to which you want to compare to isn't just data.frame. If we use normal data.frame it works as expected.

iris2 = copy(iris)
identical(iris, iris2)
# [1] TRUE

setDT(iris2)
iris2 = as.data.frame(iris2)
identical(iris, iris2)
# [1] TRUE

if as_tibble would go via as.data.frame then all should work seamlessly, but it must be dispatching to second class attribute "data.frame" instead.
Defining the following should be sufficient

as_tibble.data.table = function(x) as_tibble(as.data.frame(x))

because as.data.frame.data.table method handles removing its attributes.

I believe this problem is not only related to data.table but any package that defines class extending data.frame and uses custom attributes. as_tibble will assume it is standard data.frame and not go via as_tibble.default. Therefore to resolve problem at the root it should be rather something like that

as_tibble.data.frame = function(x) {
  if (!identical(class(x), "data.frame")) return(as_tibble(as.data.frame(x)))
  ... # continue old as_tibble.data.frame body
}

---

I noticed data.table would suffer from the same issue as tibble so I filled #5699 so problem can be addressed nicely without needing any package to add dependencies.

[jangorecki]

Indeed proposed one line change works as expected: https://github.com/Rdatatable/data.table/pull/5700/files

[TimTaylor]

@jangorecki - do you want to raise this suggestion over at {tibble}? (I'm happy to do so if you'd prefer not to but didn't want to jump in over the top).

[AngelFelizR]

Thanks @TimTaylor

@ccsarapas I think that this issue is ready to be closed.