Skip to content

gh-issue-10058-body.md

Latest commit

 

History

History
28 lines (17 loc) · 1.88 KB

gh-issue-10058-body.md

File metadata and controls

28 lines (17 loc) · 1.88 KB

Summary

IntervalSpec next-run calculation uses fixed UTC second alignment (((ts-phase)/interval+1)*interval+phase) and ignores ScheduleSpec.timezone for day-sized intervals. Schedules that mean “every N local days at a fixed local time” therefore do not track DST (e.g. US/Pacific: local midnight stays 07:00 UTC in PDT but should move to 08:00 UTC in PST after the fall transition).

StructuredCalendar / cron paths already use the loaded *time.Location; interval-only schedules do not.

Reproduction (behavior before fix)

  • ScheduleSpec.timezone = America/Los_Angeles (or any IANA zone with DST).
  • Interval = 24h, Phase = 8h (first tick aligned to 1970-01-01 00:00:00 in that zone, i.e. time.Unix(8*3600,0) local representation).
  • Observe rawNextTime / GetNextTime just after 2018-11-04 local midnight and 2018-11-05 local midnight:
    • Before fix: every tick is k * 86400 seconds from the same UTC phase (e.g. always 07:00Z in this setup).
    • After fix: 2018-11-04 00:00 local is still PDT → 07:00Z; 2018-11-05 00:00 local is PST → 08:00Z.

Proposed semantics

When interval is a positive multiple of 86400 and the schedule has a non-UTC location, interpret the step as N whole calendar days in that location at the same local time-of-day as the grid reference time.Unix(phase mod interval, 0).In(loc).

When the location is UTC (or unset/empty in a way that resolves to UTC), keep the existing fixed-UTC-second behavior for compatibility.

Non-goals

  • 43h or other non–whole-day intervals: unchanged.
  • This does not try to turn “24h” into “civil 24h” in zone for non-86400-multiple durations.

Motivation

Common use case: IntervalSpec for “every N days or weeks” with an end-user or workspace-scoped IANA time zone, where local wall time should remain stable when DST changes.