Printing error involving group_nest, rowwise , and across

[TylerGrantSmith]

I came upon the following error when attempting to debug some code. I think some combination of group_nest/rowwise/across is constructing a tibble that breaks the tbl formatter.

suppressWarnings(library(tidyverse))
tibble(x = 1:2, y = 1) %>% 
  group_nest(y) %>%
  rowwise() %>% 
  summarise(str = print(across()))
#> Error in `summarise()`:
#> ! Problem while computing `str = print(across())`.
#> i The error occurred in row 1.
#> Caused by error in `[<-.data.frame`:
#> ! replacement element 2 is a matrix/data frame of 2 rows, need 1

[hadley]

Possibly related error:

library(dplyr, warn.conflicts = FALSE)

tibble(x = 1, y = list(tibble(x = 1:2))) %>% 
  rowwise() %>% 
  summarise(x = across())
#> Error in `summarise()`:
#> ! Problem while computing `x = across()`.
#> ℹ The error occurred in row 1.
#> Caused by error in `vec_c()` at dplyr/R/summarise.R:335:2:
#> ! `value` should have been recycled to fit `x`.
#> ℹ In file 'slice-assign.c' at line 238.
#> ℹ This is an internal error in the vctrs package, please report it to the package authors.

^{Created on 2022-07-21 by the reprex package (v2.0.1)}

[DavisVaughan]

Two slightly more minimal examples

library(tidyverse)

tibble(x = list(1:2)) %>%
  rowwise() %>%
  mutate(y = across())
#> Error in `vec_slice()`:
#> ! Column `x` (size 2) must match the data frame (size 1).
#> ℹ In file 'slice.c' at line 191.
#> ℹ This is an internal error in the vctrs package, please report it to the package authors.

tibble(x = list(1:2)) %>%
  rowwise() %>%
  summarise(y = across())
#> Error in `summarise()`:
#> ! Problem while computing `y = across()`.
#> ℹ The error occurred in row 1.
#> Caused by error in `vec_c()`:
#> ! `value` should have been recycled to fit `x`.
#> ℹ In file 'slice-assign.c' at line 238.
#> ℹ This is an internal error in the vctrs package, please report it to the package authors.

I think both of these are variants of this scenario:

library(tidyverse)

tibble(x = list(1:2)) %>%
  rowwise() %>%
  mutate(y = x)
#> Error in `mutate()`:
#> ! Problem while computing `y = x`.
#> ✖ `y` must be size 1, not 2.
#> ℹ Did you mean: `y = list(x)` ?
#> ℹ The error occurred in row 1.

I'm fairly the mutate() one should error since you are trying to change the number of rows.

Maybe the summarise() one should result in a df-col for y containing tibble(x = 1:2) (hard to say)

[DavisVaughan]

library(tidyverse)

tibble(x = list(1:2)) %>%
  rowwise() %>%
  mutate(y = {assign("foo", across(), envir = .GlobalEnv); 1})
#> # A tibble: 1 × 2
#> # Rowwise: 
#>   x             y
#>   <list>    <dbl>
#> 1 <int [2]>     1

# This is what `across()` is giving us
foo
#> Error:
#> ! Assigned data `map(.subset(x, unname), vectbl_set_names, NULL)` must be compatible with existing data.
#> ✖ Existing data has 1 row.
#> ✖ Assigned data has 2 rows.
#> ℹ Row updates require a list value. Do you need `list()` or `as.list()`?

unclass(foo)
#> $x
#> [1] 1 2
#> 
#> attr(,"row.names")
#> [1] 1

# Probably from something like this,
# since we "knew" the original number of rows
new_tibble(list(x = 1:2), nrow = 1L)
#> Error:
#> ! Assigned data `map(.subset(x, unname), vectbl_set_names, NULL)` must be compatible with existing data.
#> ✖ Existing data has 1 row.
#> ✖ Assigned data has 2 rows.
#> ℹ Row updates require a list value. Do you need `list()` or `as.list()`?

^{Created on 2022-07-21 by the reprex package (v2.0.1)}

[hadley]

So the root cause seems to be that across() is generating a corrupt data frame because it's somewhere unchopping x?

library(dplyr, warn.conflicts = FALSE)

df <- tibble(x = list(1:2, 3:5)) %>%
  rowwise() %>%
  mutate(across = list(across()))

lapply(df$across, unclass)
#> [[1]]
#> [[1]]$x
#> [1] 1 2
#> 
#> attr(,"row.names")
#> [1] 1
#> 
#> [[2]]
#> [[2]]$x
#> [1] 3 4 5
#> 
#> attr(,"row.names")
#> [1] 1

^{Created on 2022-07-30 by the reprex package (v2.0.1)}

[DavisVaughan]

Note to ourselves:

I mentioned in an earlier comment that I thought these were variants of:

library(tidyverse)

tibble(x = list(1:2)) %>%
  rowwise() %>%
  mutate(y = x)
#> Error in `mutate()`:
#> ! Problem while computing `y = x`.
#> ✖ `y` must be size 1, not 2.
#> ℹ Did you mean: `y = list(x)` ?
#> ℹ The error occurred in row 1.

But I no longer believe that is correct.

across(cols, .fns = NULL) should actually be semantically equivalent to the proposed pick(cols) (#6204), so for this special case of .fns = NULL we would preserve the list-column "as is" rather than trying to access/unchop its elements.

.fns = NULL would be deprecated in favor of pick() to avoid this kind of confusion going forward.