FA analysis time is larger than expected after applying `to_integer`

[LittleBeannie]

# Enrollment
enroll_rate <- define_enroll_rate(duration = diff(c(0, 2, 4, 6, 16)), rate = c(1, 2, 3, 5), stratum = "All")

# Failure rates
duration <- diff(c(0, c(4, 12, 20)))
control_rate <- s2pwe(times = c(4, 12, 20), survival = c(0.8, 0.6, 0.5))$rate
fail_rate <- define_fail_rate(duration, fail_rate = control_rate, dropout_rate = c(0.001, 0.001, 0.001), hr = c(1, 0.6, 0.6), stratum = "All")

# Design
x <- gs_design_ahr(
  enroll_rate = enroll_rate,
  fail_rate = fail_rate,
  alpha = 0.025,
  beta = 0.1,
  analysis_time = c(12, 18, 24, 34),
  ratio = 1,
  binding = FALSE,
  upper = "gs_spending_bound",
  upar = list(sf = "sfLDOF", total_spend = 0.025, param = NULL),
  lower = "gs_b",
  lpar = c(-Inf, -Inf, -Inf, -Inf),
  h1_spending = TRUE,
  test_lower = FALSE,
  info_scale = "h0_info"
) 

The original design x with sample size of 1043.46 has a planned final analysis at 34 months.

The integer-rounded design from to_integer(x) with sample size of 1044 gives a final analysis time of about 39 months.

This is not reasonable and we conduct the following investigation to understand the root cause of this issue, and to explore potential solutions.

Short answer: The issue is caused by the original fail_rate having all finite durations.