Bug when displaying the value 9.995 using as_tibble()

[gco056]

Hi,

When I print a vetcor of numeric values using 'as_tibble()', the value 9.995 is wronlgy rounded at 9.1.
Other values like 9.996, 9.994 are correclty displayed. It looks that this only occurs for this single value 9.995. When I try with 8.995 or 7.995, it works fine.

Here after is an exemple of the bug.

library(tidyr)

df <- tibble(val = c(9.994, 9.995, 9.996))
df # Incorrect: 9.99, 9.1, 10.0
df$val # Correct: 9.994, 9.995, 9.996

df <- tibble(val = c(8.994, 8.995, 8.996))
df # Correct: 8.99, 8.99, 9.0
df$val # Correct: 8.994 8.995 8.996

df <- tibble(val = c(7.994, 7.995, 7.996))
df # Correct: 7.99, 8.00, 8.00
df$val # Correct: 7.994 7.995 7.996

The version of R I'm using is : R version 4.5.1 (2025-06-13 ucrt)
Library tibble : Version 3.2.1
Library tidyr : version 1.3.1

Have a great day,

[krlmlr]

Thanks, confirmed.

[krlmlr]

One level deeper:

pillar::num(9.995)
#> <pillar_num[1]>
#> [1] 9.1

^{Created on 2025-07-04 with reprex v2.1.1}

[krlmlr]

Also:

format(pillar::num(9.995))
#> [1] "9.1"

^{Created on 2025-07-04 with reprex v2.1.1}

[krlmlr]

Likely underlying cause (with a modified pillar_shaft_number() ):

x <- c(7.995, 8.995, 9.995, 10.995)
writeLines(format(pillar::num(x)))
#>  8.00
#>  8.99
#>  9.1 
#> 11.0
constructive::construct(pillar:::pillar_shaft_number(x)$dec)
#> list(
#>   sigfig = 3L,
#>   num = rep(TRUE, 4L),
#>   neg = logical(4),
#>   lhs = c("8", "8", "9", "11"),
#>   lhs_zero = logical(4),
#>   rhs = c(0, 0.9900000000000002, 0.9900000000000002, 0),
#>   rhs_digits = rep(c(2, 1), each = 2L),
#>   dec = rep(TRUE, 4L),
#>   exp = rep(NA_integer_, 4L),
#>   unit = NULL,
#>   si = FALSE
#> ) |>
#>   structure(width = 5L)

^{Created on 2025-07-04 with reprex v2.1.1}

[krlmlr]

I did a fuzz testing with a million random numbers and, so far, couldn't find another number where this happens.

The test runs a bit too long for my taste, looking for a quicker test:

Fuzz test

set.seed(20250704)
N <- 1000000
x <- (runif(N) + 1) * 10^(runif(N, -3, 3))
y <- as.numeric(cli::ansi_strip(format(pillar::num(x, notation = "dec"))))

summary((x - y) / max(x, y))
#>       Min.    1st Qu.     Median       Mean    3rd Qu.       Max. 
#> -2.509e-04 -3.738e-07  1.000e-11  4.300e-08  3.889e-07  2.509e-04

^{Created on 2025-07-04 with reprex v2.1.1}

Problematic case

x <- 9.995
y <- as.numeric(cli::ansi_strip(format(pillar::num(x, notation = "dec"))))

summary((x - y) / max(x, y))
#>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
#> 0.08954 0.08954 0.08954 0.08954 0.08954 0.08954

^{Created on 2025-07-04 with reprex v2.1.1}

[krlmlr]

This affects various inputs of the form 9.99..995, depending on the sigfig argument:

for (sigfig in 1:16) {
  x <- 10 - 0.5 * 10^(-sigfig + 1)
  x <- c(x - 2^-(20:70), x + 2^-(70:20))
  fmt <- cli::ansi_strip(format(pillar::num(x, sigfig = sigfig, notation = "dec")))

  neq <- which(abs(x - as.numeric(trimws(fmt))) > (10^(-sigfig + 2)))
  if (length(neq) == 0) next

  writeLines(paste0("sigfig: ", sigfig))
  print(neq)
  print(x[neq], digits = 17)
  print(fmt[neq])
}
#> sigfig: 3
#>  [1] 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56
#> [26] 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71
#>  [1] 9.9949999999999992 9.9949999999999992 9.9949999999999992 9.9949999999999992
#>  [5] 9.9949999999999992 9.9949999999999992 9.9949999999999992 9.9949999999999992
#>  [9] 9.9949999999999992 9.9949999999999992 9.9949999999999992 9.9949999999999992
#> [13] 9.9949999999999992 9.9949999999999992 9.9949999999999992 9.9949999999999992
#> [17] 9.9949999999999992 9.9949999999999992 9.9949999999999992 9.9949999999999992
#> [21] 9.9949999999999992 9.9949999999999992 9.9949999999999992 9.9949999999999992
#> [25] 9.9949999999999992 9.9949999999999992 9.9949999999999992 9.9949999999999992
#> [29] 9.9949999999999992 9.9949999999999992 9.9949999999999992 9.9949999999999992
#> [33] 9.9949999999999992 9.9949999999999992 9.9949999999999992 9.9949999999999992
#> [37] 9.9949999999999992 9.9949999999999992 9.9949999999999992 9.9949999999999992
#>  [1] " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 "
#> [10] " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 "
#> [19] " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 "
#> [28] " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 " " 9.1 "
#> [37] " 9.1 " " 9.1 " " 9.1 " " 9.1 "
#> sigfig: 6
#> [1] 30 31
#> [1] 9.9999949999999984 9.9999949999999984
#> [1] " 9.1000 " " 9.1000 "
#> sigfig: 7
#> [1] 30 31
#> [1] 9.9999994999999977 9.9999994999999977
#> [1] " 9.10000 " " 9.10000 "
#> sigfig: 11
#> [1] 30 31
#> [1] 9.9999999999499991 9.9999999999499991
#> [1] " 9.100000000 " " 9.100000000 "
#> sigfig: 14
#> [1] 30
#> [1] 9.9999999999999485
#> [1] " 9.100000000000 "
#> sigfig: 16
#> [1] 29 30
#> [1] 9.9999999999999964 9.9999999999999982
#> [1] " 9.              " " 9.              "

^{Created on 2025-07-04 with reprex v2.1.1}

[krlmlr]

On the upside, the 8.99 and 9.99 is just a consequence of floating-point representations:

format(7.995, digits = 3)
#> [1] "8"
format(8.995, digits = 3)
#> [1] "8.99"
format(9.995, digits = 3)
#> [1] "9.99"
format(10.995, digits = 4)
#> [1] "11"

^{Created on 2025-07-04 with reprex v2.1.1}

[krlmlr]

A fix is underway: r-lib/pillar#785 .

[krlmlr]

Submitted a pillar update to CRAN. Thank you for the report!