[xulkrnan]

Schedules with many exception times get slow fast. A typical case is blocking out every slot of a recurring schedule for a few weeks, e.g. one add_exception_time per 15-minute slot, which comes to a few thousand exception times.

Two things make it quadratic:

• add_exception_rule de-duplicates with Array#include?, and Rule#== builds to_hash for every comparison, so adding n exception times (directly, or through from_yaml / from_hash) is O(n²).
• exception_time? checks every candidate occurrence against every exception rule.

This PR keeps single exception times in two hashes:

• exact time values (time.to_r) for de-duplication, which is the same equality Rule#== gives SingleOccurrenceRule
• whole seconds (time.to_i) for lookup, which is the same equality SingleOccurrenceRule#on? uses

Adding and checking an exception time is now O(1). Recurring exception rules keep the existing include? and on? checks. The index is created with the schedule and rebuilt in place on remove_exception_rule / remove_exception_time. Updating it in place means dup/clone copies share it the same way they already share the rule arrays. Schedules loaded from Marshal dumps made by older versions build it on first use. Nothing in the public API changes.

Benchmark

A 15-minute schedule over 27 days with 2,400 exception times (Ruby 2.7):

	master	this PR
build	5.49s	0.03s
from_yaml	5.51s	0.04s
all_occurrences	0.97s	0.04s

Script

require "benchmark"
require "ice_cube"

start = Time.utc(2026, 1, 1)
finish = start + 27 * 86_400
schedule = nil
build = Benchmark.realtime do
  schedule = IceCube::Schedule.new(start) do |s|
    s.add_recurrence_rule(IceCube::Rule.minutely(15).until(finish))
    t = start + 86_400
    while t < finish - 86_400
      s.add_exception_time(t)
      t += 900
    end
  end
end
yaml = schedule.to_yaml
load = Benchmark.realtime { IceCube::Schedule.from_yaml(yaml) }
enum = Benchmark.realtime { schedule.all_occurrences }
puts format("build=%.2fs from_yaml=%.2fs all_occurrences=%.2fs", build, load, enum)

Behaviour

I compared master and this branch over a set of schedules: UTC, local, ActiveSupport::TimeWithZone and subsecond start times; until, count and hour_of_day rules; and duplicate exception times given in other zones or with fractional seconds, mixed with recurring exception rules. For each I checked occurrences, occurs_at?, occurring_between?, next_occurrences and previous_occurrence, to_ical, YAML, hash and Marshal round trips, and removing and re-adding exceptions. The output was identical.

Specs

The new specs in schedule_spec.rb cover:

• de-duplication, including times that differ by a fraction of a second
• exception times given in another zone
• exception times mixed with exception rules
• removing an exception time or rule, and re-adding one
• a dup copy and its original staying consistent with their exception times
• a schedule with 2,880 exception times, which exceeds the suite's 1s per-example timeout on master