int64 summation fails

[ghost]

Problem: 4 + 3 + 4 = 5.43-e323

Reprex:

tibble::tribble(
              ~ group, ~ `1`, ~ `2`, ~ `3`,
              "A", bit64::as.integer64(4), bit64::as.integer64(3), bit64::as.integer64(4)
            ) %>%
    dplyr::group_by(group) %>%
    tidyr::nest() %>% # creates 'data' column
    dplyr::mutate(
             x = purrr::map(data, function(x) {list("sum"=sum(unlist(x)))})
           ) %>%
    tidyr::unnest_wider(c(data, x))

[munoztd0]

The issue you're experiencing is due to the use of the base R sum function with integer64 objects from the bit64 package. The base R sum function does not handle integer64 objects correctly, which is causing the incorrect result.

You should use the bit64::sum.integer64 function instead, which is designed to correctly handle integer64 objects. Here's the corrected code:

tibble::tribble(
  ~ group, ~ `1`, ~ `2`, ~ `3`,
  "A", bit64::as.integer64(4), bit64::as.integer64(3), bit64::as.integer64(4)
) |>
  dplyr::group_by(group) |>
  tidyr::nest() |> # creates 'data' column
  dplyr::mutate(
    x = purrr::map(data, function(x) {list("sum"=bit64::sum.integer64(unlist(x)))})
  ) |>
  tidyr::unnest_wider(c(data, x))
#> # A tibble: 1 × 5
#> # Groups:   group [1]
#>   group     `1`     `2`     `3`     sum
#>   <chr> <int64> <int64> <int64> <int64>
#> 1 A           4       3       4      11

This should resolve the issue and give you the correct sum for your integer64 values.

[hadley]

Technically it's the unlist causing the problem, not sum.

Fortunately, there's an easier way to tackle this with dplyr: use c_across():

library(dplyr, warn.conflicts = FALSE)

df <- tribble(
  ~ group, ~ `1`, ~ `2`, ~ `3`,
  "A", bit64::as.integer64(4), bit64::as.integer64(3), bit64::as.integer64(4)
)

df |> mutate(sum = sum(c_across("1":"3")))
#> # A tibble: 1 × 5
#>   group     `1`     `2`     `3`     sum
#>   <chr> <int64> <int64> <int64> <int64>
#> 1 A           4       3       4      11

^{Created on 2024-03-14 with reprex v2.1.0}

I had actually forgotten how to do this, and found what I needed in https://dplyr.tidyverse.org/articles/rowwise.html#per-row-summary-statistics.

[ghost]

@munoztd0 @hadley Thank you both for explaining the problem and solution!