Lost numeric precision with promote_num_to_string on read

[mstackhouse]

It looks like some numeric precision gets lost when using opts_read_json(promote_num_to_string=TRUE). This only appears to happen when a multi-type row is read in, triggering the numbers to convert to character so the object imports as a character matrix.

Here's a reprex:

x <- unname(as.data.frame(list(
  col1=c(143.66666666666699825, 2/3, 1/3, 165/37, 6/7), 
  col2 = c('a', 'b', 'c', 'd', 'e')
)))

y <- yyjsonr::write_json_str(x)

print(y)
#> [1] "[[143.666666666667,\"a\"],[0.6666666666666666,\"b\"],[0.3333333333333333,\"c\"],[4.45945945945946,\"d\"],[0.8571428571428571,\"e\"]]"

with_opt <- yyjsonr::read_json_str(y, opts=yyjsonr::opts_read_json(
  promote_num_to_string = TRUE
))

print(with_opt)
#>      [,1]         [,2]
#> [1,] "143.666667" "a" 
#> [2,] "0.666667"   "b" 
#> [3,] "0.333333"   "c" 
#> [4,] "4.459459"   "d" 
#> [5,] "0.857143"   "e"

without_opt <- yyjsonr::read_json_str(y, opts=yyjsonr::opts_read_json(
  promote_num_to_string = FALSE
))

for (i in seq(1, 5)) {
  print(as.character(without_opt[[i]][[1]]))
  print(with_opt[i,1])
}
#> [1] "143.666666666667"
#> [1] "143.666667"
#> [1] "0.666666666666667"
#> [1] "0.666667"
#> [1] "0.333333333333333"
#> [1] "0.333333"
#> [1] "4.45945945945946"
#> [1] "4.459459"
#> [1] "0.857142857142857"
#> [1] "0.857143"

^{Created on 2025-01-28 with reprex v2.1.1}

[mstackhouse]

For some added context, we're using yyjsonr in our package here for that's reading in data from a specific schema. Our package is here.

The standard calls for the data frame writing out as rows, which can have mixed types. On read, we promote numbers to strings because the data reads in as a character matrix which can easily be converted to a data frame, which we then handle all type conversions using the appropriate metadata.