dhis2-devs team mailing list archive

[Bob Jolliffe <bobjolliffe@xxxxxxxxx>]

Hi

2009/5/12 Saptarshi Purkayastha <sunbiz@xxxxxxxxx>:
>
> 2009/5/12 Bob Jolliffe <bobjolliffe@xxxxxxxxx>
>>
>> Hi Saptarshi
>>
>> 2009/5/12 Saptarshi Purkayastha <sunbiz@xxxxxxxxx>:
>> > Its bad to re-invent the wheel, but I think we just keep doing that all
>> > the
>> > time. The Java API has all these things built-in, but we seem to have
>> > our
>> > own conventions for everything...
>> > The Calender class does all the good work with enough locale specific
>> > things...
>> >
>> >
>> > http://java.sun.com/javase/6/docs/api/java/util/Calendar.html#getFirstDayOfWeek()
>>
>> Sorry, but its not quite as simple as that.  The Calendar class lets
>> you specify important things like the setFirstDayOfWeek and
>> setMinimalDaysInFirstWeek, but it doesn't decide on those values for
>> you.
>
> Thankfully it decided for me based on my current locale.
> Calendar.getInstance() gives me the calendar for the locale and the
> getFirstDayOfWeek() returned 1 (SUNDAY) as my locale was en_US... I changed
> my locale to English (United Kingdom) (en_GB) and it returns 2 (MONDAY). So
> it should work fine for system with the same locale and we really shouldn't
> decide.

Thats interesting.  Hadn't realized locale affects FirstDayofWeek.
That's something.  Letting calendar decide on basis of locale is one
option.  Does it also impact on MinimumDaysInFirstWeek?  This one is
probably more significant as it determines how many weeks in the year
and which weeks are in which year.

So either we can generate weekly periods based on ISO8601 or based on
locale.  Neither is ad-hoc or arbitrarily "invented" so that is fine.
Perhaps better still might be to provide a user check box or something
to allow the user to decide.  Though I am still not sure what best
practice is regarding weekly periods - SDMX for example seems to hint
at the ISO8601 approach (I would also lean this way).  Makes sense
when a large part of its application area is financial data and
probably international in nature.  Comparing statistical data across
series where the weekly periods are determined by locale could be
painful.

The other question which occurs to me and which maybe the DHIS2 gurus
might answer is whether it really makes any difference, other than at
the point where we must generate periods - obviously then we must
decide what periods to create, using what scheme, standard, locale or
what have you.  But other than that is there any assumption built in
anywhere regarding, for example, first day of a week or first week of
the year?

> There are also REAL ways of converting between locale Calendars and
> Timezones and that should be able to solve any import/export from different
> locales.

I haven't seen anything regarding converting between week numbers :-(
Maybe its not a big deal.  I'd still like to know what the actual
problem with weekly data experienced by Orvalho was.  Better to be
focussing on real problems than potential ones ...

Cheers
Bob

>There are also various subclasses of the Calendar class available
> online for other than GregorianCalendar.
>
> I would suggest not to make the starting week to something constant. Instead
> letting the locale decide I think is the right way forward. As for hacks to
> get data from Access/Excel/Flat files... it has to be the way the person
> importing has to understand and take care.
>
>
>>
>>  And the WeeklyPeriodType is of course using the Calendar class.
>> When generating weekly periods for a year it has setFirstDayOfWeek to
>> Monday.  We don't specify MinimalDaysInFirstWeek so I checked - we get
>> 4 as a default, which is fine.  In other words DHIS2 generates ISO8601
>> compliant week numbers (using the calendar class) which is excellent.
>>
>> There are only three problems:
>> 1.  there is a small bug in the generator which I would like to fix
>>
>> 2.  we don't validate when we construct weekly periods.  For example
>> we construct a weekly period with three parameters - type (weekly),
>> start date and end date.  When we construct we should ensure that
>> startdate is a Monday and that enddate is indeed 6 days apart from
>> startdate.  Otherwise it is possible, through the API, to construct
>> invalid weekly periods.  This probably shouldn't be. (in fact having
>> both start and enddate is strictly redundant - but I can see how it is
>> there in support of generality).
>>
>> 3.  its not clear how to deal with importing weekly data from a system
>> which does not use ISO8601 week numbering.  This would probably
>> include a number of VB/Excel/Access type applications.  Either (i) we
>> must be completely agnostic towards the week numbering scheme and
>> accept anything with a start and end date 6 days apart as a week (or
>> even short weeks as per US convention) or (ii) we need to have an
>> approved process of grafting periods into (and perhaps out of) the
>> ISO8601 scheme.  The first option is always a good interoperability
>> principle but I'm not sure we have the algorithms elsewhere, in the
>> reporting modules for example, to support it.  The second option is a
>> bit gruff, and I'm not sure how it squares with Vietnamese and
>> Ethiopian dates for example.  (What's a Vietnamese week?)  But its
>> straightforward to implement.
>>
>> Agreed we should use the Calendar class for all this.  But how
>> exactly?  Does anybody else have some experience of dealing with the
>> vagaries of exchanging weekly data?  How does DHIS1.4 understand it?
>> Do weeks run from Monday to Sunday or Sunday to Saturday?  Or
>> Wednesday to Tuesday with the first week in the year being the first
>> Wednesday after Easter :-)
>>
>> Regards
>> Bob
>>
>> > ---
>> > Regards,
>> > Saptarshi PURKAYASTHA
>> > Director R & D, HISP India
>> > Health Information Systems Programme
>> >
>> > My Tech Blog:  http://sunnytalkstech.blogspot.com
>> > You Live by CHOICE, Not by CHANCE
>> >
>> >
>> > 2009/5/12 Murodullo Latifov <murodlatifov@xxxxxxxxx>
>> >>
>> >> Hi Caveman,
>> >>
>> >>
>> >>
>> >> ----- Original Message ----
>> >> From: Orvalho Joaquim Augusto <orvaquim@xxxxxxxxx>
>> >> To: Bob Jolliffe <bobjolliffe@xxxxxxxxx>
>> >> Cc: dhis2-devs <dhis2-devs@xxxxxxxxxxxxxxxxxxx>
>> >> Sent: Tuesday, May 12, 2009 2:08:43 PM
>> >> Subject: Re: [Dhis2-devs] weeks calculation
>> >>
>> >> Well
>> >>
>> >> Yes this a problem for us. Because we expect to use correctly DHIS
>> >> [reporting tools for now] and we need to map exactly the periods from
>> >> the
>> >> actual system to DHIS.
>> >>
>> >> But meanwhile there is a another problem with a lot of empty data on
>> >> that
>> >> system. Mine foucus has been on that and I forgot for a while these
>> >>  dateweeks troubles.
>> >>
>> >> Anyway, can you corect the DHIS2 code please to calculate weekdates? If
>> >> it
>> >> is buggy.
>> >>
>> >> Regards
>> >> Caveman
>> >>
>> >> Are current periods in Mozambican database generated by DHIS or they
>> >> are
>> >> also imported from elsewhere?
>> >>
>> >> murod
>> >> _______________________________________________
>> >> Mailing list: https://launchpad.net/~dhis2-devs
>> >> Post to     : dhis2-devs@xxxxxxxxxxxxxxxxxxx
>> >> Unsubscribe : https://launchpad.net/~dhis2-devs
>> >> More help   : https://help.launchpad.net/ListHelp
>> >>
>> >>
>> >>
>> >>
>> >>
>> >> _______________________________________________
>> >> Mailing list: https://launchpad.net/~dhis2-devs
>> >> Post to     : dhis2-devs@xxxxxxxxxxxxxxxxxxx
>> >> Unsubscribe : https://launchpad.net/~dhis2-devs
>> >> More help   : https://help.launchpad.net/ListHelp
>> >
>> >
>> > _______________________________________________
>> > Mailing list: https://launchpad.net/~dhis2-devs
>> > Post to     : dhis2-devs@xxxxxxxxxxxxxxxxxxx
>> > Unsubscribe : https://launchpad.net/~dhis2-devs
>> > More help   : https://help.launchpad.net/ListHelp
>> >
>> >
>
>
> _______________________________________________
> Mailing list: https://launchpad.net/~dhis2-devs
> Post to     : dhis2-devs@xxxxxxxxxxxxxxxxxxx
> Unsubscribe : https://launchpad.net/~dhis2-devs
> More help   : https://help.launchpad.net/ListHelp
>
>