Signed Zero • discretes

library(discretes)

In R, zero has a latent sign: +0 and -0 print the same but behave differently under reciprocation:

identical(0, -0)
#> [1] TRUE
1 / 0
#> [1] Inf
1 / -0
#> [1] -Inf

When a numeric series contains zero, that sign can matter for derived series (for example, after inversion).

Support for signed zero is included in the discretes package so that the proper signed infinity results by inversion:

x <- arithmetic(0, 3, n_left = 1, n_right = 1)
x
#> Arithmetic series of length 3:
#> -3, 0, 3
1 / x
#> Reciprocal series of length 3:
#> Loading required namespace: testthat
#> -0.3333333, 0.3333333, Inf
1 / -x
#> Reciprocal series of length 3:
#> -Inf, -0.3333333, 0.3333333

The functions has_negative_zero() and has_positive_zero() report whether -0 or +0 is a discrete value in the series. For example, the integers contain a single positive zero:

has_negative_zero(integers())
#> [1] FALSE
has_positive_zero(integers())
#> [1] TRUE

Rules around signed zero follow those of base R. For example, negation can induce negative zeroes (although note that, in R, -0 is only possible as type “double” and not of type “integer”).

has_negative_zero(-1.5 * integers())
#> [1] TRUE

Like numeric vectors, a series can contain both signs of zero:

x <- dsct_union(-0, 0)
x
#> Union series of length 1:
#> 0
has_positive_zero(x)
#> [1] TRUE
has_negative_zero(x)
#> [1] TRUE

However, since they are both identical values, only one zero is enumerated by next_discrete(), prev_discrete(), or get_discretes_*():

num_discretes(x)
#> [1] 1

But, both signs remain latent and appear when the series is inverted:

1 / x
#> Reciprocal series of length 2:
#> -Inf, Inf
num_discretes(1 / x)
#> [1] 2

When choosing to pull positive or negative zero from the series:

For numeric vectors, behaviour is delegated to base::unique().
For combined series through dsct_union(), the zero from the first series that has a zero is taken.
Otherwise, positive zero is selected.