tz strftime %V week numbers are incompatible with Solaris, GNU
While testing some code I discovered that the tz strftime week numbers are incompatible with the %V week numbers generated by Solaris 2.5.1 strftime and with the GNU C Library strftime. I don't know what the correct definition of week numbers is (and I don't particularly want to know :-) but I thought I should report the problem, since it's two against one. #defining XPG4_1994_04_09 fixes some but not all of the discrepancies. Here are all examples of discrepancies from 1994 through 1998. Each column contains the output of strftime applied to "%U %V %W" for the given date. In the cases marked `*', #defining XPG4_1994_04_09 didn't remove the discrepancy. Solaris date & glibc tz tz with XPG4_1994_04_09 1994-01-01 00 53 00 00 52 00 00 53 00 1994-01-02 01 53 00 01 52 00 01 53 00 1995-01-01 01 53 00 01 52 00 01 53 00 1996-12-30 52 53 53 52 01 53 52 01 53* 1996-12-31 52 53 53 52 01 53 52 01 53* 1997-12-29 52 53 52 52 01 52 52 01 52* 1997-12-30 52 53 52 52 01 52 52 01 52* 1997-12-31 52 53 52 52 01 52 52 01 52* While we're on the subject, perhaps XPG4 behavior should be the default, with non-standard behavior being an option.
Paul Eggert wrote:
While testing some code I discovered that the tz strftime week numbers are incompatible with the %V week numbers generated by Solaris 2.5.1 strftime and with the GNU C Library strftime. I don't know what the correct definition of week numbers is (and I don't particularly want to know :-)
Whether you want to know it or not: ;-) %V is used by the date command of POSIX.2 in order to get the ISO 8601 week number (which is very widely used for industrial planning in many European countries). I understand that %V strftime will be defined in the revised ISO C standard as well to be ISO 8601. This is therefore what tz strftime should do. The ISO 8601 week definition is explained on <URL:http://www.ft.uni-erlangen.de/~mskuhn/iso-time.html>. In brief: - Week numbers can range from 01 to 53. Some years have only weeks 01 to 52. - Week 1 is the week that includes January 4, which is identical to the first week that includes the first Thursday of the new year, which is identical to the first week that has the majority of its days in the new year. - Day one of the week is Monday, day seven is Sunday. Today (1996-07-17) is day 3 of week 29 of the year 1996 (1996-W29-3). You can use in the Emacs calendar function the command p c in order to find the ISO 8601 week number that %V should print if you do not have a European style calendar on your desk (where the ISO week numbers serve as a sort of page numbers). /* * Calculate week number and day of week according to ISO 8601. * Implementation derived from code provided by L. Kirby. */ void get_week_number(const struct tm *t, int *week, int *dayofweek) { int tmp1, tmp2, fourthdaynum, year, isleap; tmp1 = 1 - t->tm_wday; tmp2 = t->tm_yday + ((tmp1 > 0) ? 3 : 10) + tmp1; fourthdaynum = tmp2 % 7; *week = tmp2 / 7; if (*week == 0) { year = t->tm_year + 1900 - 1; isleap = !(year % 4) && ((year % 100) || !(year % 400)); *week = (fourthdaynum + isleap >= 6) ? 53 : 52; } else if (*week == 53) { year = t->tm_year + 1900; isleap = !(year % 4) && ((year % 100) || !(year % 400)); if (fourthdaynum > isleap) *week = 1; } if (t->tm_wday == 0) *dayofweek = 7; else *dayofweek = t->tm_wday; return; } Markus -- Markus Kuhn, Computer Science student -- University of Erlangen, Internet Mail: <mskuhn@cip.informatik.uni-erlangen.de> - Germany WWW Home: <http://wwwcip.informatik.uni-erlangen.de/user/mskuhn>
participants (2)
-
Markus Kuhn -
Paul Eggert