[R] Objects Written to Feather Not Restored Exactly

[asfimport]

Rather simple - write the standard 'iris' dataset to feather, then read it back.

At first glance everything looks the same, but setting 'attrib.as.set = FALSE' to identical() will return FALSE.

Using Waldo to compare, you can see that the order of attributes is different on the restored object. "class" should be the second attribute after "names" but before "row.names".

This should be a simple fix to the 'read_feather' function to set the correct order of attributes.

—

iris <- iris
arrow::write_feather(iris, file <- tempfile())
iris2 <- arrow::read_feather(file)
unlink(file)
identical(iris, iris2)
#> [1] TRUE
identical(iris, iris2, attrib.as.set = FALSE)
#> [1] FALSE
waldo::compare(attributes(iris), attributes(iris2))
#> names(old): "names" "class" "row.names"
#> names(new): "names" "row.names" "class"

Environment: R4.1.1, Ubuntu 20.04
Reporter: Charlie Gao

_{Note: This issue was originally created as ARROW-13661. Please see the migration documentation for further details.}

[asfimport]

Ian Cook / @ianmcook:
Thanks for the report!

Does this difference in the order of the attributes have any practical significance?

[asfimport]

Charlie Gao:
Strictly speaking, the order of attributes does not matter as documented in R. But how else do you ensure you get the same thing out as you put in?

The practical significance is that currently I am hashing the object before I write, storing the hash and then hashing the restored object to verify integrity. Currently I am having to deal with the attribute order in the code on my side, but I am thinking there is no good reason for the attributes to be out of order in the first place. Thanks.

[thisisnic]

Closing this as I don't think attribute order roundtrip fidelity is something we're likely to invest time in support unless it becomes a wider issue for other users. Happy to review a PR to fix this though if anyone wants to submit one.