catch empty datasets passed to `unite`

[bcjaeger]

Hello, I stumbled on this by mistake and thought I'd open a PR given the error message.

It seems that unite could make an assertion about the shape of its input to prevent this.

---

Brief description of the problem: unite encounters an internal error if the input data is empty.

library(tibble)
library(tidyr)
options(error = traceback)
tibble(x=NULL) %>% unite(col='tmp', everything())
#> Error in `df_append()`:
#> ! `after` must be a whole number, not an integer `NA`.
#> ℹ This is an internal error that was detected in the tidyr package.
#>   Please report it at <https://github.com/tidyverse/tidyr/issues> with a reprex
#>   (<https://tidyverse.org/help/>) and the full backtrace.
#> Backtrace:
#>     ▆
#>  1. ├─tibble(x = NULL) %>% unite(col = "tmp", everything())
#>  2. ├─tidyr::unite(., col = "tmp", everything())
#>  3. └─tidyr:::unite.data.frame(., col = "tmp", everything())
#>  4.   └─tidyr:::df_append(after = after)
#>  5.     └─tidyr:::check_number_whole(after, min = 0L, max = n, .internal = TRUE)
#>  6.       └─tidyr:::.rlang_types_check_number(...)
#>  7.         └─tidyr (local) .stop(x, what, ...)
#>  8.           └─tidyr:::stop_input_type(...)
#>  9.             └─rlang::abort(message, ..., call = call, arg = arg)

^{Created on 2024-09-25 with reprex v2.0.2}

[DavisVaughan]

This actually works in the dev version of tidyr and gives

# A tibble: 0 × 1
# ℹ 1 variable: tmp <chr>

unite() has a guarantee that it will always generate the col column, regardless of how many columns you provide it, and we uphold that now even in the empty data frame case

[bcjaeger]

Nice - thank you!