numbers near underflow limit display as `Inf.e-324` in tibble?-CodePudding

Per the docs, "on a typical R platform the smallest positive double is about 5e-324." Given a double vector with values above, near, and below this limit:

library(tibble)

small_vec <- c(4e-300, 4e-324, 4e-350)
small_df <- data.frame(x = small_vec)
small_tibble <- tibble(x = small_vec)

Printing small_vec and small_df give about what I'd expect:

small_vec
#> [1] 4.000000e-300 4.940656e-324  0.000000e 00

small_df
#>               x
#> 1 4.000000e-300
#> 2 4.940656e-324
#> 3  0.000000e 00

The second value isn't quite right, which I vaguely understand is due to floating point weirdness. The third number underflows to 0. Fine. But printing as a tibble,

small_tibble
#> # A tibble: 3 × 1
#>           x
#>       <dbl>
#> 1   4 e-300
#> 2 Inf.e-324
#> 3   0

^{Created on 2022-10-19 with reprex v2.0.2}

I'm thrown by Inf.e-324 -- both the idea of Inf with an exponent, and the decimal point. What does this signify? Or is it possibly a bug in the tibble package?

CodePudding user response：

(A little long for a comment.)

Sure seems like a bug, but I think it's not in tibble per se, but in the pillar package underneath. library(pillar); pillar_shaft(small_vec) shows the same pathology. Digging down: looking at pillar:::format.pillar_shaft_decimal(), it uses the $sci component of this object. That component looks fishy:

> pillar_shaft(small_vec)$sci

[harmless-looking stuff]

$lhs
[1] "4"   "Inf" "0"  

$lhs_zero
[1] FALSE FALSE  TRUE

$rhs
[1]   0 NaN   0

[harmless-looking stuff]

So we probably have to see what pillar_shaft is doing during the conversion.

Specifically in

pillar:::pillar_shaft_number_attr(small_vec, NULL, NULL)

which goes to pillar_shaft_number which seems to lead to

pillar:::split_decimal(small_vec, sigfig = 3, digits = NULL)

but, weirdly enough, that one looks OK, so I must have lost the plot somewhere.