Date filter returning wrong week number

[hbbh]

I'm assuming that the dateFilter with the 'w' param supplied should be returning weeks running from Monday to Sunday.

I've noticed that it's returning the incorrect weeks for the first two Sundays in 2015 (ISO-8601):

2015-01-04 (Sunday) - week 2 (should be week 1)
2015-01-11 (Sunday) - week 3 (should be week 2)

Correct week numbers can be found here:
http://www.epochconverter.com/date-and-time/weeknumbers-by-year.php

I've got an example jasmin spec confirming this here:
http://plnkr.co/edit/tAQ22cxpVbJKOC8uIFlF?p=preview

Is this a bug or am I missing something?

[pkozlowski-opensource]

@hbbh currently the date filter doesn't calculate week numbers according to the ISO-8601. Instead it assumes week 1 to be the first week of the year with a Thur (where week starts on Sun).

I don't think we want to support ISO-8601 way of calculating things as it has all sorts of non-intuitive behaviours but what we should do is to allow people specify what is the fist day of a week in their respective locale. This is going to happen via #10919

Longer discussion on the topic here: #10450

[lgalfaso]

A week runs from Sun to Sat. A day belongs to the week that the Thu in that week belongs to. In this case, week 1 is from Jan 1 (Thu) to Jan 3; and week 2 is from Jan 4 (Sun) to Jan 10.

All the above follows ISO-8601 logic (even when we are not compilant with it)

Closing this issue

[pkozlowski-opensource]

@lgalfaso hmm, I really think this is about first day of a week. Let's take the 2015-01-04 date (Sun). 2015-01-01 (Thur) is w1, no question about it. But 2015-01-04 (Sun) can be either w1 or w2, depending on what you take as the first day of a week (Mon or Sun as today).

So in the end I think it is locale-dependent and we should fix it. I'm not sure we've got a different issue for this so I would like to keep this one open - unless I'm getting this all wrong here.

[lgalfaso]

@pkozlowski-opensource I do not know what the best solution would be. ISO-8601 defines the first day of the week to be Monday and should be represented with the number 1 and the last day Sunday that should be represented with the number 7.
The CLDR defines the day 0 to Sunday and 6 as Saturday, regardless of what first day of the week is a given locale.

Even with the 0 vs 7 variation, this is still all good.

Anyhow, one thing is clear first day of the week should use one of these conventions. Say we use CLDR (as ISO-8601 does not have this information). This is, if the first day of the week is a Saturday, then first day of the week should be 6.

Now, there are two options week number when first day of the week is not Sunday (say it it Saturday):

• The week number is still calculated using ISO-8601 (as we have now)
• The week number is calculated using first day of the week (what I think you are proposing)

Now, if we go with the formal case, then not everybody may be happy but it is (somehow) easy to explain. If we go with the later, then we may have to make other corrections.

• day of the week returns 0 for Sunday and 6 for Saturday. Should this be shifted depending on first day of the week?

If the answer is yes, then first day of the week should also be shifted (and will always be 0)
If the answer is no, then weeks will run as 6, 0, 1, 2, 3, 4, 5 (and I think this can cause more confusion)

This is, the result of this will make people unhappy regardless, and this is why I would like to avoid to have to deal with this. This is, not run into this question in the first place. For good or for bad, this implies that "The week number is calculated using first day of the week" may turn more controversial than what we have now.

I wish there was a standard here :)

[pkozlowski-opensource]

@lgalfaso I hear what you are saying but:

• currently we are no fully-ISO- compliant anyway (see tests)
• having a week starting always on Sun is no-go for many applications I've seen and working with, so as much as I try to avoid hard questions to avoid complexity, this issue is part of the "sad reality"

I would love to skip this problem and declare the current behaviour "good" but it is just not enough.

I can own this item and see what I can do but I wouldn't like to skip this problem (at lest not just yet).

[lgalfaso]

@pkozlowski-opensource is you want to own this issue, then that works for me. Now, as a side-note, if there is any format I think works best is to try to align with http://www.unicode.org/reports/tr35/tr35-dates.html#Date_Field_Symbol_Table

[pkozlowski-opensource]

@lgalfaso coll. I will re-open it then and assign to me.