This guide is about the hard part of time: future dates, recurring events, and time zones that change. The basics are covered in the cookbook; this guide covers the four things that bite real calendars.

  1. You can't materialise an infinite stream. Bounded scheduling is a design decision.

  2. Wall-clock time is authoritative; UTC is a projection. Tempo stores the user's wall time and derives the UTC instant on demand.

  3. "Floating" events vs zoned events — Tempo distinguishes them through what's on the value, not a flag.

  4. Future dates survive zone-rule changes — because Tempo never caches a UTC value, re-reading a future event automatically uses the current Tzdata.

Each section below gives the principle, the Tempo idiom, and the pitfall to avoid.

Setup — required for every example

Every code example in this guide uses the ~o sigil from Tempo.Sigils. Before running any of them — in iex, a script, or a module — you must bring the sigil into scope:

import Tempo.Sigils

The import adds only sigil_o/2 and sigil_TEMPO/2 to the caller's namespace; no helper functions leak in.

1. Bounded enumeration

An RRULE like FREQ=MONTHLY;BYDAY=2MO ("the second Monday of every month") is infinite — occurrences continue forever. You cannot call Enum.to_list/1 on an infinite sequence.

Tempo splits the two operations: building the recurrence AST is cheap and always bounded-free, but materialising the AST into a concrete %Tempo.IntervalSet{} requires a bound.

rule = %Tempo.RRule.Rule{freq: :month, interval: 1, byday: [{2, 1}]}
{:ok, ast} = Tempo.RRule.Expander.to_ast(rule, ~o"2025-01-01")

# `ast` is a %Tempo.Interval{recurrence: :infinity, ...} — not an error.
# Materialising it requires a bound:

{:ok, set} = Tempo.to_interval(ast, bound: ~o"2025-07-01")
Tempo.IntervalSet.count(set)
#=> 7

The AST is the recurrence rule; the IntervalSet is its occurrences inside a window. :bound is always supplied at materialisation, never at the rule. Tempo's default member-preserving semantics keep each occurrence as a distinct member of the IntervalSet — which is what you want for scheduling. For the covered-instant form (individual occurrences merged into contiguous spans), pipe through Tempo.IntervalSet.coalesce/1 — useful for free/busy questions but not for "list the events."

For ad-hoc use, Stream.take/2 and Enum.take/2 work directly on the AST — it's enumerable, lazily:

ast |> Stream.take(10) |> Enum.to_list()

Pitfall

Forgetting the bound and calling Tempo.to_interval(ast):

{:error,
 "Cannot materialise an unbounded recurrence (recurrence: :infinity, no UNTIL). Supply a :bound option — any Tempo value whose upper endpoint limits the expansion."}

Tempo refuses rather than hanging — forgotten bounds are a design error, not a runtime surprise.

2. Wall-clock time is authoritative

When a user schedules "08:00 on 1 March 2030 in Paris", the real scheduling intent is the wall-clock reading on that morning — not a specific UTC instant. If Paris changes its DST rules between now and 2030, the UTC instant should shift; the wall-clock event should not.

Tempo stores exactly what the user wrote:

t = ~o"2030-03-01T08:00:00[Europe/Paris]"
t.time
#=> [year: 2030, month: 3, day: 1, hour: 8, minute: 0, second: 0]

t.extended.zone_id
#=> "Europe/Paris"

No UTC seconds are cached on the struct. When you ask for the UTC projection, Tempo consults Tzdata at the time of the call, and the result reflects whatever zone rules Tzdata currently knows about 2030. Re-run the same call after Tzdata updates and the number may change — that's the correct behaviour for future dates.

Principle

Store what the user said. Compute UTC when you need it. This is the same discipline Apple Calendar and Google Calendar enforce: the canonical form of a future event is wall-clock + zone, not UTC.

Pitfall

Serialising a Tempo value as "the UTC seconds" and rehydrating from that:

# Do NOT do this for future events:
cached_utc = Tempo.Compare.to_utc_seconds(event)
store_in_database(cached_utc)

When Tzdata ships a DST rule change for 2030, your cached number is now wrong — but you've lost the wall-clock information needed to recompute. Serialise the Tempo value itself (Tempo.to_iso8601/1 round-trips faithfully) and project to UTC only at display or comparison time.

3. Floating vs zoned events

Apple Calendar's "floating" events (they call them "All-day" in the UI, but the concept applies to timed events too in iCal RFC 5545) stay at the same wall-clock reading regardless of the viewer's zone — a "6am workout" on holiday is still 6am local.

Tempo distinguishes floating and zoned through presence of the zone tag:

floating = ~o"2030-03-01T08:00:00"
floating.extended
#=> nil

paris = ~o"2030-03-01T08:00:00[Europe/Paris]"
paris.extended.zone_id
#=> "Europe/Paris"

utc = ~o"2030-03-01T08:00:00Z"
utc.shift
#=> [hour: 0]

fixed = ~o"2030-03-01T08:00:00+05:30"
fixed.shift
#=> [hour: 5, minute: 30]
FormMeaning
~o"2030-03-01T08:00:00"Floating — 8am in whatever zone the reader is in
~o"2030-03-01T08:00:00[Europe/Paris]"Zoned — 8am Paris wall time, UTC derived on demand
~o"2030-03-01T08:00:00Z"UTC-anchored — a specific UTC instant, wall times vary by zone
~o"2030-03-01T08:00:00+05:30"Fixed-offset — UTC+05:30 regardless of zone-rule changes

The right choice depends on what the user said:

  • "Morning workout at 6am" when travelling — floating.

  • "Meeting at 2pm Paris" — zoned ([Europe/Paris]).

  • "Server job at 03:00 UTC" — UTC-anchored (Z).

  • "Event at UTC+05:30" (fixed-offset calendar system) — fixed-offset.

Principle

The interpretation of a wall-clock reading is metadata, not an afterthought. Tempo forces you to make that choice at parse time and preserves it through every operation.

4. Future dates survive zone-rule changes

When Paris last moved its DST rules in 1996, any calendar entry stored as a frozen UTC instant for a post-1996 date had to be rewritten. Modern calendars (Google, Apple, Outlook) avoid this by storing wall-clock + zone and deriving UTC on demand — which is exactly what Tempo does.

Concretely: a Paris event stored today for 2030 will be re-evaluated with whatever Tzdata knows about Paris's rules in 2030 at the time of the computation:

event = ~o"2030-03-01T08:00:00[Europe/Paris]"

# Today, Tzdata thinks this is UTC+1 (standard time in March).
Tempo.Compare.to_utc_seconds(event)
#=> 64052726400

# If Tzdata 2028a ships saying France abolished DST in 2027,
# the next call to to_utc_seconds returns a different number
# — the event is "still 8am Paris wall time, but the UTC shifts."

Tempo.Interval.duration/2 likewise re-evaluates each endpoint's UTC projection on every call. Comparisons and set operations do the same. Nothing is frozen.

When to freeze

There are cases where you want the UTC instant frozen — a CI build that must run "at exactly 03:00 UTC on 15 March 2030" regardless of how zones shift. Store those as UTC (~o"2030-03-15T03:00:00Z"). The Z suffix is the explicit promise: "this is a UTC instant, don't recompute."

Principle

Don't cache UTC. Don't cache DST. Tempo doesn't, and the reason you trust this library over rolling your own is exactly this discipline.

Putting it together

A practical scheduling layer built on Tempo looks like:

defmodule Schedule do
  def weekly_meeting(name, %Date{} = date, %Time{} = time, zone) do
    dtstart =
      Tempo.new!(
        year: date.year,
        month: date.month,
        day: date.day,
        hour: time.hour,
        minute: time.minute,
        zone: zone
      )

    rule = %Tempo.RRule.Rule{freq: :week, interval: 1}
    {:ok, ast} = Tempo.RRule.Expander.to_ast(rule, dtstart)
    %{name: name, rule: ast}
  end

  def occurrences_in(%{rule: ast}, from, to) do
    bound = Tempo.Interval.new!(from: from, to: to)
    {:ok, set} = Tempo.to_interval(ast, bound: bound)
    Tempo.IntervalSet.to_list(set)
  end
end

retrospective =
  Schedule.weekly_meeting("Retro", ~D[2025-06-01], ~T[14:00:00], "Europe/London")

# Occurrences for Q3:
Schedule.occurrences_in(retrospective, ~o"2025-07-01", ~o"2025-10-01")
# 18 weekly occurrences, each with wall time 14:00 in Europe/London

Tempo.new/1 is the runtime companion to the ~o sigil. The sigil is for literal values in source; new/1 takes keyword components that can come from anywhere — function arguments, database rows, API payloads, form inputs. String interpolation to assemble an ISO 8601 value and then parse it back is always the wrong move: it round-trips through formatting twice and bypasses the type-level validation that component construction gives you.

Store the rule as an AST (with zoned wall time). Materialise into an IntervalSet only when you need concrete occurrences, bounded to the query window. Display by projecting each endpoint's wall time through the viewer's preferred zone. Nothing about the stored rule changes when Tzdata does.