`nested_join()` to preserve the class of the nested data?

[huizezhang-sherry]

In the example below, if I nest dt2, the rowwise_df class is preserved in each list column element, I'm wondering if nest_join() should have a similar behavior?

# set up `dt1` with an id column and some random numbers
set.seed(123)
dt1 <- tibble::tibble(
  id = c(1,2,3),
  a = rnorm(n = 3)
)

# set up `dt2` as a rowwise dataframe with `id2` nested inside `id1` column 
set.seed(123)
dt2 <- tibble::tibble(
  id = rep(c(1,2,3), each = 3),
  id2  = rep(c("a", "b", "c"), 3),
  b = rnorm(n = 9)
) |>
  dplyr::rowwise()

# nesting `dt2` gives each element in data as a `rowwise_df`
dt2 |> tidyr::nest(data = -id)
#> # A tibble: 3 × 2
#>      id data                
#>   <dbl> <list>              
#> 1     1 <rowwise_df [3 × 2]>
#> 2     2 <rowwise_df [3 × 2]>
#> 3     3 <rowwise_df [3 × 2]>

# but `nest_join()` doesnt not preserve the `rowwwise_df` class 
dt1 |> dplyr::nest_join(dt2, by = "id")
#> # A tibble: 3 × 3
#>      id      a dt2             
#>   <dbl>  <dbl> <list>          
#> 1     1 -0.560 <tibble [3 × 2]>
#> 2     2 -0.230 <tibble [3 × 2]>
#> 3     3  1.56  <tibble [3 × 2]>

^{Created on 2022-06-13 by the reprex package (v2.0.1)}

[DavisVaughan]

All of the *_join() functions have the property that they return a data frame with the same type as x, the left hand side input. That is mentioned in the return value docs of ?left_join:

An object of the same type as x.

I think we want to retain that property here

[huizezhang-sherry]

I agree with retaining the type of left hand side input with *_join(). But here I'm asking what the type inside the list-column should be, i.e. with the last line, should the output be

#> # A tibble: 3 × 3
#>      id      a dt2             
#>   <dbl>  <dbl> <list>          
#> 1     1 -0.560 <tibble [3 × 2]>
#> 2     2 -0.230 <tibble [3 × 2]>
#> 3     3  1.56  <tibble [3 × 2]>

or

#> # A tibble: 3 × 3
#>      id      a dt2             
#>   <dbl>  <dbl> <list>          
#> 1     1 -0.560 <rowwise_df [3 × 2]>
#> 2     2 -0.230 <rowwise_df [3 × 2]>
#> 3     3  1.56  <rowwise_df [3 × 2]>

[DavisVaughan]

Oh I see, I misunderstood. Hmm, I think we need to think about this a little more.

[hadley]

Minimal reprex:

library(dplyr, warn.conflicts = FALSE)

df1 <- tibble(id = 1:3, a = 4:6)
df2 <- rowwise(tibble(id = c(2, 3, 3), b = 7:9))

out <- df1 |> dplyr::nest_join(df2, by = "id", multiple = "all")
class(out$df2[[2]])
#> [1] "tbl_df"     "tbl"        "data.frame"

^{Created on 2022-07-26 by the reprex package (v2.0.1)}