java - SimpleDateFormat Week Calculations

Question

I'm getting some puzzling results with SimpleDateFormat and am hoping that someone can shed some light on the issue. The output:

Time          = Mon Dec 27 00:00:00 PST 2010
2010-01 <--- THIS IS WHAT I DON'T UNDERSTAND
Start of week = Sun Dec 26 00:00:00 PST 2010
2010-01
End of Week   = Sat Jan 01 23:59:59 PST 2011
2011-01

Should I be treating the last "week" of the year that extends to the next year as a special case? Or is this the correct way to interpret this? Obviously when attempting to organize week sequentially, the order is incorrect. Adjusting the initial values, Dec 25, 2005 is considered the 53rd week. I haven't looked at Joda yet to see if Joda produces similar results.

The relevant code:

private static Date getStartOfWeek( Date d ) {
  Calendar calendar = Calendar.getInstance();
  calendar.clear();
  calendar.setTime( d );

  calendar.set( Calendar.DAY_OF_WEEK, calendar.getFirstDayOfWeek() );

  return calendar.getTime();  
}

private static Date getEndOfWeek( Date d ) {
  Calendar calendar = Calendar.getInstance();
  calendar.clear();

  calendar.setTime( d );
  calendar.add( Calendar.WEEK_OF_YEAR, 1 );
  calendar.set( Calendar.DAY_OF_WEEK, calendar.getFirstDayOfWeek() );
  calendar.add( Calendar.MILLISECOND, -1 );

  return calendar.getTime();
}


Calendar calendar = Calendar.getInstance();
calendar.clear();
calendar.set( 2010, Calendar.DECEMBER, 27 );
Date d = calendar.getTime();
Date start = getStartOfWeek( d );
Date end = getEndOfWeek( d );
SimpleDateFormat fmt = new SimpleDateFormat( "yyyy-ww" );

System.out.println( "Time          = " + d );
System.out.println( fmt.format( d ) );
System.out.println( "Start of week = " + start );
System.out.println( fmt.format( start ) );
System.out.println( "End of Week   = " + end );
System.out.println( fmt.format( end ) );

Background: I found this when using the crosstab (date grouped into week) in JasperReports.

EDIT: I am using JDK 1.6.0_25

EDIT: It seems that I will have to use Joda to get the correct result. TO get the week start/end, I ended up using: LocalDate.withDayOfWeek. To retrieve the year and week number, I used DateTime.getWeekyear and DateTime.getWeekOfWeekyear.

Answer 1

The bug is in your formatting code, not Java.

The surprising behavior is due to an esoteric rule in date notation. Note that ISO 8601 (rather confusingly) specifies different rules for year boundaries when using week numbers. In particular, 2010-12-27 is considered part of 2011 when using week numbers.

As a result, you should be using the "week year" YYYY rather than the usual year yyyy. (See http://download.oracle.com/javase/7/docs/api/java/util/GregorianCalendar.html#week_year and the last example in http://download.oracle.com/javase/7/docs/api/java/text/SimpleDateFormat.html.)

Also, the standard notation for dates uses an explicit 'W', so you should use new SimpleDateFormat( "YYYY-'W'ww" ) instead.

Edit: There's another problem. Java seems to default to the non-standard calendar.getMinimalDaysInFirstWeek() == 1, so you have to set

calendar.setMinimalDaysInFirstWeek( 4 );

in order to get the correct year.

Edit: From reading the Calendar javadocs, you might also need to set the starting day to Monday. Further, the YYYY format specifier seems to be new in Java 1.7. In light of this, unless you're willing to upgrade to a pre-release Java version, I recommend just using Joda Time.