Need to split firstDay into firstDay for calendar display vs for week-of-year calculation

Per the CLDR Design WG discussion on 2025-01-27 there is not a strong case to split “firstDay” into “firstDay for calendar display” vs “for week-of-year calculation”, and therefore this ticket will be closed.

Two additional tickets have been filed as follow up:

• https://unicode-org.atlassian.net/browse/CLDR-18275#icft=CLDR-18275 "Use ISO week numbering as the default for formatting"

• https://unicode-org.atlassian.net/browse/CLDR-18276#icft=CLDR-18276 "Research and possibly deprecate week-of-month formatting"

It sounds like different (work) weeks may be highly specific depending on user needs (see comment about DOT week numbering for tires etc) and we would need more information in order to decide how to effectively support them.

There is a need to identify particular (work) weeks. This need is fulfilled by different means in different places, languages, cultures and even situations.

Ordinal week, with different reference frames:

• Ordinal week of the year, “calendar week”. 1–52/53(/54?)

• Ordinal week of the quarter-year (≈ season). 1–(12/)13/14

• Ordinal week of the month. 1–4/5(/6?)

Week containing a particular anchor date, with special cases:

• Week starting at a particular date.

• Week ending at a particular date.

• Week including the first/last day of a month/quarter/year ⇒ devolves into one type of ordinal week.

Globally, Monday through Thursday are always part of the same full week and work week. One may include Friday as well because where it is considered to begin the week, it only does so in the evening (after work hours) as far as I am aware. Those five days of the week (DOW) are therefore unambiguous as anchor dates.

This issue asserts that the DOW that is used locally for the first column in printed and on-screen calendars may differ from the one used in the algorithm that determines the week number. There still isn’t any proof or even just indication that this was actually the case anywhere.

However, numbering weeks is not the only way to designate and identify them, as explained above. I’m pretty sure that, in some locales, therefore there is no agreed algorithm to unambiguously number weeks. Where there is though, there are usually three options:

• The first DOW must be in the month/quarter/year, i.e. the full week at the beginning, any part at the end. minDays=7

• The last DOW must be in the month/quarter/year, i.e. any part of the week at the beginning, all parts at the end. minDays=1

• The majority of DOWs must be in the month/quarter/year. minDays=4

Where the minimum is below the length of the week, i.e. 7 days, variations may occur for counting partial weeks, i.e. whether the other days count as part of an ordinal week of the previous/next month/quarter/year. If they do, a month (except February) may have a maximum of 6 weeks and a (leap) year 54 weeks. (No definition of quarter-year, except for astronomic seasons, makes any have 93 or more days, so their maximum remains at 14 weeks.)

Those variants differ of course by what actually is the first (and therefore last) DOW. The assumption underlying this issue seems to be that there may be locales where Sunday (or perhaps Saturday) is considered the first DOW, but for the ordinal week number of the year, Monday or rather the work days as a whole appear more relevant. I believe that even if such cases exist, most or even all of them can already be modeled by changing minDays accordingly! For instance, for weeks starting Sunday:

• If the first day of the work week (= after the weekend) instead of the first DOW (i.e. Monday instead of Sunday) must be included, use 6. instead of 7.

• If the last day of the work week (= before the weekend) instead of the last DOW (i.e. Friday instead of Saturday) must be included, use 2. instead of 1.

• If the majority of the work week (starting Monday) must be included, use 4. For a six-day work week, 3 may suffice.

• If the last day of the two-day weekend instead of the last DOW (i.e. Sunday instead of Saturday) must be included, use 8 – or 1. For a one-day weekend, it remains 7.

I opened https://unicode-org.atlassian.net/browse/CLDR-18276 to discuss the week-of-month numbering.

1. I’m not saying that we do support than in CLDR data.

2. I agree that Week 3 (and for that matter, “the first week” or “the last week”) are ambiguous if you don’t know the “first week” conditions.

"Week of February 18" is a common expression. That's not what is in CLDR data, though. It has data to express "Week 3 of February". Which in February is fairly unambiguous since the month started on a Saturday, but could be confusing if the month started on a Wednesday.