Problem

The pre-dev version (0.3.1 and earlier) of split_by() can lead to bugs when there is a split by a variable stored as a double, if the upper bound of one level is the lower bound of another.

Minimal example: this fails to mark sections between 2.5 and 3.2 as not belongning to any level, instead adding a new level between levels 2 and 3. This results in a total of 4 levels, rather than the desired 3.

# v0.3.1
dat <- data.frame(
  a = c(1.0, 1.1, 1.3, 1.9, 2, 2.1, 2.2, 2.9, 3, 3.1, 3.3)
)

dat |>
  split_by(a, 1:2 ~ 2:2.5 ~ 3.2:4, filter=FALSE) |>
  with(df)

# A tibble: 11 × 3
       a LexOPS_splitCond_A tmp  
   <dbl> <fct>              <chr>
 1   1   A1                 A1   
 2   1.1 A1                 A1   
 3   1.3 A1                 A1   
 4   1.9 A1                 A1   
 5   2   A2                 A2   
 6   2.1 A2                 A2   
 7   2.2 A2                 A2   
 8   2.9 A3                 A3   
 9   3   A3                 A3   
10   3.1 A3                 A3   
11   3.3 A4                 A4   

This was due to the way that the breaks were put through unique() before being given to the cut() function:

This could result in the gap between 2.5 and 3.2 in the example being ignored:

breaks <- unique(c(1, 2, 2, 2.5, 3.2, 4))
cut(dat$a, breaks, labels=1:4)

 [1] <NA> 1    1    1    1    2    2    3    3    3    4   
Levels: 1 2 3 4

Fix

The issue was addressed by ff8b8ab. Instead of cut(), a more transparent method of subsetting is used:

In addition, the dev version now correctly detects that a value of 2 could belong to either group, and gives an error to avoid any ambiguity in pipelines.

# v0.3.1.9000 (Commit 372df3b)
dat <- data.frame(
  a = c(1.0, 1.1, 1.3, 1.9, 2, 2.1, 2.2, 2.9, 3, 3.1, 3.3)
)

dat |>
  split_by(a, 1:2 ~ 2:2.5 ~ 3.2:4, filter=FALSE) |>
  with(df)

Error in check_overlapping(breaks) : 
  overlapping levels - ensure that no value could fall into multiple levels