`filter()` removes labels from `glue` variables

[d-morrison]

Here's a reprex:

library(labelled)
library(glue)
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
test = tibble(
  a = 1 |> set_label_attribute("a1"),
  b = glue::glue("two") |> set_label_attribute("b2")
)
test$b |> get_label_attribute()
#> [1] "b2"

# filter() strips the variable label from column `b`
dplyr::filter(test)$b |> get_label_attribute()
#> NULL

# filter() doesn't strip the label from column `a`
dplyr::filter(test)$a |> get_label_attribute()
#> [1] "a1"

^{Created on 2025-01-28 with reprex v2.1.1}

I expected column b to keep its label attribute. Not sure if this is ultimately an issue with dplyr, glue, or labelled, but seems like it's in dplyr?

[d-morrison]

(tested with dplyr 1.1.4.9000, glue 1.8.0, labelled 2.14.0, and R 4.4.2 on windows 11)

[DavisVaughan]

At the lowest level vctrs is slicing these vectors

library(labelled)

x <- 1 |> set_label_attribute("a1")
y <- glue::glue("two") |> set_label_attribute("b2")

vctrs::vec_slice(x, 1)
#> [1] 1
#> attr(,"label")
#> [1] "a1"
vctrs::vec_slice(y, 1)
#> two

For unclassed atomics, we have some fallback behavior that attempts to retain the attributes during slicing.

For classed objects like glue that provide vctrs methods and know the set of attributes they require, we only keep the attributes that they have requested that we keep. This generally disregards any "extraneous" attributes.

I'd encourage using the labelled type directly if you really need to keep a label around, as it was designed for this and tells vctrs to keep the label attribute around.

library(labelled)
library(glue)
library(dplyr)

test = tibble(
  a = labelled(1:2, label = "a1"),
  b = labelled(glue::glue("two"), label = "b2")
)

str(dplyr::filter(test))
#> tibble [2 × 2] (S3: tbl_df/tbl/data.frame)
#>  $ a: int+lbl [1:2] 1, 2
#>    ..@ label: chr "a1"
#>  $ b: chr+lbl [1:2] two, two
#>    ..@ label: chr "b2"