Weight column not removed by `uncount()` when using `.data` pronoun

[ndunnewind]

I expected the weights column to be removed when using the .data pronoun in uncount().

df <- tibble::tibble(x = c("a", "b"), n = c(1, 2))

tidyr::uncount(df, n)
#> # A tibble: 3 × 1
#>   x    
#>   <chr>
#> 1 a    
#> 2 b    
#> 3 b

tidyr::uncount(df, .data$n)
#> # A tibble: 3 × 2
#>   x         n
#>   <chr> <dbl>
#> 1 a         1
#> 2 b         2
#> 3 b         2

^{Created on 2024-11-12 with reprex v2.1.1}

[gguyver]

We (@tinarozsos and I) have looked at this issue. The problem seems to arrive in uncount.data.frame, where the weights argument is turned into a quosure and then later checked for whether it is a symbol in the context of the dataframe, using the function quo_is_symbol(). If .remove is TRUE, and quo_is_symbol() is also TRUE, then the matching column is removed.

quo_is_symbol() returns FALSE when using .data pronoun, meaning the matching column is not removed.

We propose editing the function with additional checks, but haven’t had the time to develop a working check yet. The code as currently written looks like so:

if (.remove) {
    if (quo_is_symbol(weights_quo)) {
      out[[as_string(get_expr(weights_quo))]] <- NULL
    } else {
      # remove ".data$" from quosure before looking for colname
    }
  }

NB: we note that using external vectors as the weights also runs into this issue, but we decided that this is correct behaviour as the vectors are not a data column.