Also note that the first Thursday of January 2026 is *January 1st*. -Matt
On Mon, Oct 19, 2026 at 9:38 PM Matt Johnson-Pint <[email protected]> wrote: > Yeah, I just realized I answered Tim's and Brooks' questions, but not > yours. Sorry David! > > > ...Windows appears to calculate the DST end date as Thursday, January 7, > 2027, instead of Friday, January 1, 2027 > > That's definitely not an IANA TZ Data issue. You'll need to continue with > Microsoft Support if you're getting this out of actual Windows UWP > classes. If you're actually pulling the registry data yourself, then it's > just your misinterpretation of the data. > > Windows dynamic time zone rules are year-over-year in the registry data, > and they either have 0 or 2 transitions exactly. So any single-transition > change like this shift to permanent DST in Manitoba has to be modeled as a > year where the "DST period" either snaps to the start of the year or the > end of the year. But rather than cut off at any particular time on say Dec > 31 23:59:59.999, if DST is in effect at the end of a year before the end > data is reached, then it stays in effect going into the following year. > > Manitoba in particular is modeled as: > > 2025: UTC-6 base offset, DST is +1 hour (ie., UTC-5). Start DST on 2nd > Sunday of March at 02:00. End DST on 1st Sunday of November at 02:00 > 2026: UTC-6 base offset, DST is +1 hour (ie., UTC-5). Start DST on 2nd > Sunday of March at 02:00. End DST on 1st Thursday of January at 00:00 > 2027: UTC-5 base offset, DST does not apply > > So when you read the DST end date for 2026 as being the 1st Thursday in > January,* it doesn't mean it floats into January of 2027* - it means that > DST ended already in January 2026. Since it was already not in effect, that > actual date would be a no-op. In other words, the DST delta was already 0, > so ending it keeps it at 0. This is a hack because of the nature of this > data structure - but it's an effective hack. > > 2027 has no DST at all, it's just UTC-5 the whole year. > > This trick has been used in the past without issue. A good example is > Yukon in 2020. > > Again - this is about the Windows registry legacy data. It has nothing to > do with how TZDB has modeled America/Winnipeg. That scheme is much more > dynamic and much more close to reality. Just look up America/Winnipeg in > the source data, and you'll see how it's done. > > Cheers, > Matt > > On Mon, Oct 19, 2026 at 9:33 PM David <[email protected]> wrote: > >> Hi Tim, Brooks, and Matt, >> >> Thank you all for digging into this—and especially Matt for the clear >> breakdown of the dual architecture. That distinction explains why we see >> divergent behavior between modern UWP/.NET APIs (backed by ICU/tzdb) and >> legacy Win32 registry consumers. >> >> Now that the mechanism behind the data split is clear, I’d like to steer >> back to my core question regarding how this cutover is modeled and what >> operational effects we should anticipate around the January 1 vs. January 7 >> boundary: >> >> 1. >> >> *The 2026 vs. 2027 Baseline Shift:* >> >> In 2026, Manitoba operates on a standard baseline of UTC-06:00, with >> DST (+1 hour) active since the second Sunday in March 2026. Because >> Manitoba abolished the autumn fallback, it remains at UTC-05:00 through >> year-end as Daylight Time. In 2027, the baseline permanently becomes >> UTC-05:00 (Manitoba Standard Time) with no seasonal DST. >> 2. >> >> *The First Thursday (Jan 7) vs. First Friday (Jan 1) Boundary:* >> >> Because January 1, 2027 is a Friday, encoding transitions using >> day-of-week occurrence rules (such as the first Thursday of the month) >> pushes the boundary to Thursday, January 7, 2027, leaving the 2026 DST >> rule >> open into the first week of the new year. >> 3. >> >> *Questions for the Maintainers:* >> - >> >> *tzdb Cutover Modeling:* How has tzdb encoded the America/Winnipeg >> transition across late 2026 and 2027 to ensure the new base offset >> takes >> effect cleanly without leaving an active DST state hanging into the new >> year? >> - >> >> *State & Flag Mismatch (Jan 1–7):* During that 7-day window, both >> rules resolve to the same wall-clock offset (UTC-05:00), but under >> opposite >> daylight states: one treats it as Daylight Time (isdst = 1, UTC-6 >> base + 1h bias, CDT), while the other treats it as Standard Time (isdst >> = 0, UTC-5 base, CST/MST). >> - >> >> *Downstream Edge Cases:* Beyond nominal wall-clock time, what >> known failure modes exist when client applications or libraries >> evaluate >> timestamps in that 7-day window—specifically regarding calendar >> recurrence >> rules, database timestamp ordering, or POSIX/glibc conversions in mixed >> environments? >> >> I’d appreciate any perspective on how reference tz implementations handle >> this year-end handoff, and whether extending the 2026 DST rule to January 7 >> introduces risks of parser misinterpretation or historical query anomalies. >> >> Best, >> >> David Abbott >> >> On Mon, Oct 19, 2026 at 10:53 PM Matt Johnson-Pint < >> [email protected]> wrote: >> >>> I'll share what I can remember, but with the caveat that I left >>> Microsoft in 2021, and also I cannot disclose information that would be >>> considered confidential. >>> >>> I'll start with the current source of truth for Windows time zone >>> information, which is https://aka.ms/dstblog. From the blogs, it would >>> appear the current person in charge of TZ updates at Microsoft is Naveen >>> Shankar. I don't know Naveen personally. (Naveen, if you're reading this, >>> keep up the great work!). Microsoft posts official guidance and news about >>> updates it's making to Windows time zone data on that blog. They have >>> recently posted several articles about new Canadian time zones. >>> >>> Note that these posts describe Windows time zones in terms of their >>> "legacy" or "registry" time zone identifiers - for example, instead of >>> America/Vancouver, the new identifier for Windows is "British Columbia >>> Standard Time". Some of these newer identifiers sound meaningful, but >>> historically they have been haphazardly assigned without regard to >>> uniformity. "Romance Standard Time", "Russia Time Zone 3", and "UTC-02" >>> are some examples of the variations you might find. Basically, these >>> started in the earliest days of Windows NT and while several folks tried to >>> make sense of them over the years, others just threw stuff in as the need >>> arose. To mask those inconsistencies, each identifier is given a >>> "display name" that is localized for different languages. Those are in >>> tzres.dll, for any given Windows language pack. For those interested, >>> here's a little program that extracts those names from the resource files: >>> https://github.com/tomkludy/TimeZoneWindowsResourceExtractor >>> >>> Keep in mind that lots of different technologies are layered into >>> Windows. The legacy time zone data will always be there and must continue >>> to be serviced in some manner, because there's just too many old components >>> and applications based on .NET Framework, the Win32 API, and god knows what >>> else out there that all depend on this data. For example there's the Win32 >>> GetDynamicTimeZoneInformation >>> API that returns a DYNAMIC_TIME_ZONE_INFORMATION structure. So it's quite >>> good of Microsoft to maintain this proprietary legacy time zone data. >>> >>> Now, the bigger question is - does Windows use the actual IANA TZDB? The >>> answer is yes - sort of. Newer Windows technologies such as UWP use APIs >>> that work with IANA identifiers rather than the legacy data. For example, >>> Windows.Globalization.Calendar.GetTimeZone returns a string such as >>> America/Los_Angeles. Also, .NET 6 and higher support IANA time zones >>> natively. In both cases, the data is provided by ICU, which modern Windows >>> has integrated as a core component. ICU's normal time zone data >>> distribution process applies, but also Windows distributes it via Windows >>> Updates for most users. >>> >>> Most time-aware components that are part of Windows itself are now using >>> these newer technologies rather than Win32, so it's fair to say - yes. >>> Windows does use TZDB and IANA time zones. But that doesn't mean it >>> doesn't _also_ have the legacy component. Keeping these two in sync is >>> part of the responsibility of the team Naveen now belongs to (presumably). >>> Part of that is informing CLDR so they can keep the windowsZones.xml file >>> current, or simply by making new legacy time zones as needed so there's >>> something to map to. >>> >>> With regard to tzsync.exe - my understanding is that it is for limited >>> "in the moment" or "at the transition" backfills of the registry data - >>> mostly as a workaround for edge cases that the registry format can't >>> support - such as more than two transitions in a single year (e.g. Ramadan >>> suspension). I'm not sure if it uses tzdata directly for that or not - if >>> it does, it's probably via ICU. >>> >>> Again, my memory may be fuzzy, but I hope I answered your question. >>> >>> TLDR: Yes, Windows uses IANA TZ data (via ICU), and it ALSO maintains >>> its legacy registry format because older APIs require it. >>> >>> -Matt >>> >>> On Mon, Oct 19, 2026 at 5:27 PM Brooks Harris <[email protected]> wrote: >>> >>>> On 2026-10-19 03:44 PM, Tim Parenti wrote: >>>> >>>> On Mon, 19 Oct 2026 at 15:13, Brooks Harris <[email protected]> >>>> <[email protected]> wrote: >>>> >>>> >>>> The short answer is that Microsoft does not use tzdb and internally >>>> represents timezone data in the Windows registry in its own way. So this >>>> very likely has nothing to do with how tzdb has encoded Manitoba's change >>>> and sounds like a bug in how Microsoft have chosen to encode the change. >>>> >>>> >>>> I'm not sure that's true. As I understand it they map TzDb through CLRD: >>>> https://raw.githubusercontent.com/unicode-org/cldr/master/common/supplemental/windowsZones.xml >>>> >>>> I think you may have the directionality backwards on the purpose of that >>>> file. That file is maintained by CLDR and appears to be used for mapping >>>> Windows zone definitions to TZDB identifiers, not the other way around. >>>> See its edit >>>> history:https://github.com/unicode-org/cldr/commits/main/common/supplemental/windowsZones.xml >>>> >>>> >>>> Well, it can be used in either direction. I use it both ways. >>>> >>>> I think our question is how is Windows using it? >>>> >>>> But this is only mapping TzDb identities to Windows display names. >>>> The Registry has many intricate values concerning time zones, including >>>> "Dynamic DST", binary blobs, and references to tzres.dll. There are also >>>> .dlls and .exes in C:\Windows\System32\ that sound like they are involved: >>>> tzautoupdate.dll, tzsyncres.dll, tzsync.exe. So it's not looking simple. >>>> >>>> I think our current question is "Is Windows using TzDb source data to >>>> populate the Windows time zone information?" >>>> >>>> Some months ago Matt Johnson-Pint mentioned he'd worked on this when at >>>> Microsoft: >>>> Email Re: [tz] Re: Sunshine Protection Act passes US House - 2026-07-16 >>>> I've cc'd him. Matt, any comments? >>>> >>>> While not directly relevant to the current question I'd like to better >>>> understand the methods Microsoft is using and how the information is >>>> contained in Windows. (I've done much research on this but have not >>>> completely connected the dots.) >>>> >>>> -Brooks >>>> >>>> As I read it, this would allow Windows applications to identify the >>>> system's selected Windows timezone and territory, map them to a >>>> corresponding TZDB identifier, and then use that identifier to pull up any >>>> of the various localization strings that CLDR defines for each TZDB >>>> identifier. Nothing about this file says anything about the underlying >>>> timezone data itself, nor that Microsoft is somehow using this file in >>>> reverse to do any data transformations. >>>> >>>> Robert Bastian (Cc'd) is involved with CLDR and the maintenance of that >>>> file and may be able to say more if I got any of that wrong. >>>> >>>> -- >>>> Tim Parenti >>>> >>>> >>>> >>>>
