Given a start and end datetime, generate a row index for each 12-hour shift spanned

There's probably a more concise way, but one version could use lubridate and dplyr from the tidyverse to construct a sequence of shift changes within each long shift.

library(tidyverse)
helper function to find shift start from input time
shiftback 
  mutate(across(start:end, ymdhms)) |>
  reframe(shiftbegan = seq.POSIXt(
    shiftback(start), shiftback(end + hours(11) + minutes(59), "12 hours"),
    .by = c(id, start, end)) |>
  mutate(startadj = pmax(start, shiftbegan),
         endadj = pmin(end, shiftbegan + hours(12))) |>
  select(-shiftbegan) |>
  filter(startadj < endadj) 

Resulting in:

  id               start                 end           startadj             endadj
1  A 2026-02-01 15:30:00 2026-02-02 09:30:00 2026-02-01 15:30:00 2026-02-01 18:00:00
2  A 2026-02-01 15:30:00 2026-02-02 09:30:00 2026-02-01 18:00:00 2026-02-02 06:00:00
3  A 2026-02-01 15:30:00 2026-02-02 09:30:00 2026-02-02 06:00:00 2026-02-02 09:30:00
4  B 2026-02-05 19:00:00 2026-02-06 20:00:00 2026-02-05 19:00:00 2026-02-06 06:00:00
5  B 2026-02-05 19:00:00 2026-02-06 20:00:00 2026-02-06 06:00:00 2026-02-06 18:00:00
6  B 2026-02-05 19:00:00 2026-02-06 20:00:00 2026-02-06 18:00:00 2026-02-06 20:00:00

We can feed this into ggplot2 to get a visual confirmation:

... |>
  ggplot() +
  ggarrow::geomarrowsegment(
    aes(startadj, xend = endadj, y = id, yend = id)) +
  scalexdatetime(breaks = seq.POSIXt(
    ymdh(2026010106), ymdh(2026123118), "12 hours"),
    date_labels = "%m/%d\n%I%p")

I think @Tobo's join approach is more elegant and makes more sense in most cases. (That said, I can imagine bizarre edge cases where it could be unwieldy, like tracking 1min ranges for events spanning 100 years -- then you'd be creating 50 million buckets when it might be simpler to make those "on the fly" as needed.)

A dplyr adaptation of that approach is below. Note, the overlaps() helper for non-equi joins is very useful here, as it pulls in the full range of possible overlapping shifts in one step. I specify bounds = () here so that a shift that starts or ends exactly at a shift change does not include the adjacent shift.

x$start