@jose, @Qqwy, I am prototyping some of these implementation details in the next release of my CLDR work. The nature of a week is indeed a complex beast. But since José was smart enough to pick ISO calendar as the Elixir implementation the definition of "first week" seems clear. The first week is that in which the first Thursday falls. (or is the that week with at least 4 days of the new year).
Before I get too far may I ask if this is in agreement? The reference I am using is https://en.wikipedia.org/wiki/ISO_week_date#First_week with the suggest calculations being https://en.wikipedia.org/wiki/ISO_week_date#Calculation For other calendars the definition of "first week" is/may be territory specific. As is "first day" and "weekday". Which is why localisation is such fun ...... --Kip On Tuesday, July 11, 2017 at 3:53:36 PM UTC+2, Wiebe-Marten Wijnja wrote: > > Yes, I think those arguments are arguments for allowing weeks to be > specified in their own calendar implementations, treating weeks like months. > I'd think Elixir itself should then implement the ISO week date calendar, > and locale-specific week logic can be put in locale-specific calendar > implementation modules. This would also keep the door open for "4-5-4" and > friends-based weeks, 10-day based weeks, etc. > Counting in weeks definitely has the same amount of complexity as counting > in months in a normal calendar (I am talking about odd leaping rules here), > so it definitely makes sense to use the same or a similar interface for > this. > > The question then remains to how to make this more user-friendly. Maybe we > can add functions on the `Date`-module that take an optional > `week_calendar: Calendar.Implementation` as last parameter, that underwater > does: > > 1. Take the Date-struct, and convert it to the given Calendar. (failing as > usual if the conversion is not possible because of incompatible calendars) > 2. Perform calculations on the resulting struct using the existing > Calendar implementation functions, where the 'months' are weeks. > 3. Return the desired answer. > > > On Monday, July 10, 2017 at 11:48:26 PM UTC+2, Kip wrote: >> >> Definitely worth thinking about as an approach for weeks. Some of >> calendars may make that tricky - I'll need to ponder some more. >> >> 1. "4-5-4" calendar (and its related friends 445 and 544). This is >> used extensively by the retail industry ( >> https://nrf.com/resources/4-5-4-calendar). >> >> 2. Another class is the financial period calendars of corporations in >> the US that get to define their fiscal year. For example Cisco Inc has a >> financial year defined as "ends of the last Saturday of July". Its fiscal >> year 2017 for example, starts on July 31st 2016. (also to note the >> effective year is different from the gregorian year of the day in >> question). This is a class of calendar that requires configuration >> information - I haven't quite worked out the right strategy for these >> calendars yet (Date.new/3 won't fit these cases as is) >> >> Then there is the locale-specific understanding of a week: >> >> 1. Starts on a Monday, Saturday or Sunday >> >> 2. 1st week starts as the first calendar week which contains four days >> of the new year >> >> 3. 1st week starts as the week in which January 1st occurs >> >> I think these situations are likely to make basing weeks along the ISO >> Week strategy a bit tricky. >> >> On Tuesday, July 11, 2017 at 5:15:37 AM UTC+8, Wiebe-Marten Wijnja wrote: >>> >>> I am all for this! >>> >>> However, I do have a note about the week-related stuff: >>> On one hand, people might expect 'weeks' to be readily available >>> (`:calendar` offers them, and many other libraries in other language >>> environments do). >>> On the other, many calendars do not themselves have a notion of weeks. >>> (Weeks are only secondary units of date measurements (years, months, days >>> are the primary ones) and nearly all calendars, all across the globe, use a >>> seven-day week. >>> Weeks are basically the months of the ISO week date calendar: >>> https://en.wikipedia.org/wiki/ISO_week_date >>> I wonder how much of the week-related functionality could be defined by >>> just taking a date (in e.g. Calendar.ISO) and converting it into the ISO >>> week date calendar. >>> >>> ~Qqwy/Wiebe-Marten >>> >>> >>> On Monday, July 10, 2017 at 9:03:53 AM UTC+2, Kip wrote: >>>> >>>> With the new capabilities of Calendar in 1.5 I am now adding date and >>>> time localisation to a package I maintain ( >>>> https://hex.pm/packages/ex_cldr). >>>> >>>> CLDR ( >>>> http://unicode.org/reports/tr35/tr35-dates.html#Date_Format_Patterns) >>>> defines a number of symbols to aid formatting and localisation. Many of >>>> these can be supported by the Calendar module (or easily derived). Some >>>> would be better defined in a calendar module - which would imply adding >>>> them to the behaviour and to Calendar.ISO. >>>> >>>> These functions are: >>>> >>>> 1. months_in_year(year) which together with the existing >>>> days_in_month/2 would allow the calculation of quarters which is relevant >>>> in a lot of business contexts >>>> 2. week_of_year(date) which returns the week number within which the >>>> date occurs. >>>> 3. week_of_month(date) which returns the week of the month within which >>>> the date occurs >>>> 5. day_of_week_in_month(date) returns the ordinal day (i.e. 2nd in 2nd >>>> Wednesday in July) >>>> 6. year(date) which returns the effective year in which the date lies. >>>> For example in an ISO Week implementation, the "year" may be different to >>>> the "year in the date" >>>> >>>> Although my immediate focus is date formatting and localisation, these >>>> functions are of themselves relevant for other "calendarists" (is that a >>>> word?) Each of these may have quite specific implementation details >>>> related to a given calendar and therefore would seem to be appropriate to >>>> be part of the behaviour. >>>> >>>> I've tried to think of only those functions which cannot be readily >>>> derived without knowledge of the specific calendar but I know this is a >>>> slippery slope - especially with anything to do with time and dates. >>>> >>>> Thoughts? >>>> >>>> --Kip >>>> >>>> -- You received this message because you are subscribed to the Google Groups "elixir-lang-core" group. To unsubscribe from this group and stop receiving emails from it, send an email to elixir-lang-core+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/elixir-lang-core/56690dd1-c89b-4e28-963a-fcd54c72494b%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.
