Add unnest_tree()

[mgirlich]

Closes #1384.

@DavisVaughan Would you be interested in adding this to tidyr? If so, what about the inverse operation nest_tree()?

[mgirlich]

Removing the vctrs_list_of class has huge impact on performance:

After removing:

devtools::load_all("~/GitHub/tidyr/")
#> ℹ Loading tidyr
test_tree <- readr::read_rds("~/GitHub/tidyr/test-tree.rds")
bench::mark(
  unnest_tree = unnest_tree(
    test_tree,
    id,
    children
  )
)
#> Warning: Some expressions had a GC in every iteration; so filtering is disabled.
#> # A tibble: 1 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 unnest_tree    170ms    180ms      5.27    3.97MB     7.02

^{Created on 2022-08-25 with reprex v2.0.2}

Before removing:

#> Warning: Some expressions had a GC in every iteration; so filtering is disabled.
#> # A tibble: 1 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 unnest_tree    26.9s    26.9s    0.0372     508MB     7.62

^{Created on 2022-08-25 with reprex v2.0.2}

[mgirlich]

It is not quite as bad as the list_of performance hit but still a lot. Without unclassing this takes

devtools::load_all("~/GitHub/tidyr/")
#> ℹ Loading tidyr
test_tree <- readr::read_rds("~/GitHub/tidyr/test-tree.rds")
bench::mark(
  unnest_tree = unnest_tree(
    test_tree,
    id,
    children
  )
)
#> Warning: Some expressions had a GC in every iteration; so filtering is disabled.
#> # A tibble: 1 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 unnest_tree    425ms    428ms      2.34    3.89MB     7.01

^{Created on 2022-08-25 with reprex v2.0.2}

[DavisVaughan]

I won't be able to take a look for a while, but it seems interesting at first glance

BTW I just merged #1387 so you should be able to rebase/merge-main so the checks can pass (except some weirdness with the pkgdown one right now)

[mgirlich]

Thanks for the quick feedback. I'm fine with waiting for a review, already good to know that you consider it interesting. Some questions for the future:

• Must all children have the same columns?
• Should it really be an error if column children does not exist at top level?
• What if child has column level_to, parent_to, or ancestors_to?