as.data.table should remove extra attributes from extended data.frames

[jangorecki]

Spotted when investigating #5698

x = iris
class(x) = c("tbl","data.frame")
attr(x, "t1") = "a"
y = as.data.table(x)
attr(y, "t1")
#[1] "a"       ## should be NULL
class(y)
#[1] "data.table" "data.frame"

We should assume creator of custom class provides correct data.frame method that clear any extra attributes

as.data.frame.tbl = function(x) {
  attr(x, "t1") = NULL
  class(x) = "data.frame"
  x
}
z = as.data.frame(x)
attr(z, "t1")
#NULL

So then consistency can be retained, as described in #5698

[iagogv3]

Unmotivating the use of data.table...

Many times, one has to use haven to read data with SPSS, Stata or excel formats, which obviously produces tibbles. Fortunately, they can be transformed to data.table's and, for the moment, these transformations keep attributes, particularly variable labels. IMO, removing that information is not a big thing.

Although I use setDT instead of as.data.table...

[MichaelChirico]

I think that removal of attributes by as.* methods is the norm in R... users wishing to retain attributes should do so explicitly.

It's somewhat inconvenient but I think it's more the expected behavior from experienced R users.