mktime invalid normalization

[anglov]

mktime incorrectly normalizes struct tm. When year is leap year and you normalize month interval, one day is lost.

eg.:

for (int i = 0; i < 60; ++i) {
    tm.tm_mon += 1;
    time = mktime(&tm);
    printf("%s", ctime(&time));
}

results with:
Tue Feb 1 06:00:00 2000
Wed Mar 1 06:00:00 2000
Sat Apr 1 06:00:00 2000
Mon May 1 06:00:00 2000
Thu Jun 1 06:00:00 2000
Sat Jul 1 06:00:00 2000
Tue Aug 1 06:00:00 2000
Fri Sep 1 06:00:00 2000
Sun Oct 1 06:00:00 2000
Wed Nov 1 06:00:00 2000
Fri Dec 1 06:00:00 2000
Sun Dec 31 06:00:00 2000
Tue Jan 30 06:00:00 2001
Fri Mar 2 06:00:00 2001
Mon Apr 2 06:00:00 2001
Wed May 2 06:00:00 2001
Sat Jun 2 06:00:00 2001
Mon Jul 2 06:00:00 2001
Thu Aug 2 06:00:00 2001
Sun Sep 2 06:00:00 2001
Tue Oct 2 06:00:00 2001
Fri Nov 2 06:00:00 2001
Sun Dec 2 06:00:00 2001
Wed Jan 2 06:00:00 2002
Sat Feb 2 06:00:00 2002
Sat Mar 2 06:00:00 2002
Tue Apr 2 06:00:00 2002
Thu May 2 06:00:00 2002
Sun Jun 2 06:00:00 2002
Tue Jul 2 06:00:00 2002
Fri Aug 2 06:00:00 2002
Mon Sep 2 06:00:00 2002
Wed Oct 2 06:00:00 2002
Sat Nov 2 06:00:00 2002
Mon Dec 2 06:00:00 2002
Thu Jan 2 06:00:00 2003
Sun Feb 2 06:00:00 2003
Sun Mar 2 06:00:00 2003
Wed Apr 2 06:00:00 2003
Fri May 2 06:00:00 2003
Mon Jun 2 06:00:00 2003
Wed Jul 2 06:00:00 2003
Sat Aug 2 06:00:00 2003
Tue Sep 2 06:00:00 2003
Thu Oct 2 06:00:00 2003
Sun Nov 2 06:00:00 2003
Tue Dec 2 06:00:00 2003
Fri Jan 2 06:00:00 2004
Mon Feb 2 06:00:00 2004
Tue Mar 2 06:00:00 2004
Fri Apr 2 06:00:00 2004
Sun May 2 06:00:00 2004
Wed Jun 2 06:00:00 2004
Fri Jul 2 06:00:00 2004
Mon Aug 2 06:00:00 2004
Thu Sep 2 06:00:00 2004
Sat Oct 2 06:00:00 2004
Tue Nov 2 06:00:00 2004
Thu Dec 2 06:00:00 2004
Sat Jan 1 06:00:00 2005

[agkaminski]

Fixed in 0cde62e. Still possible problem if days or hours cause a year change, so I am leaving this one open.

[badochov]

is it still a problem after recent time related changes?

[anglov]

If so, it would be great to have automatic test for validation.

[jmaksymowicz]

Yes, it seems that the problem is fixed now. Tested libphoenix at 023e717
Code:

struct tm tm = {0};
tm.tm_year = 100;
tm.tm_mday = 1;
tm.tm_hour = 6;
for (int i = 0; i < 60; ++i) {
    tm.tm_mon += 1;
    time_t time = mktime(&tm);
    printf("%s", ctime(&time));
}

Result:

Tue Feb 1 06:00:00 2000
Wed Mar 1 06:00:00 2000
Sat Apr 1 06:00:00 2000
Mon May 1 06:00:00 2000
Thu Jun 1 06:00:00 2000
Sat Jul 1 06:00:00 2000
Tue Aug 1 06:00:00 2000
Fri Sep 1 06:00:00 2000
Sun Oct 1 06:00:00 2000
Wed Nov 1 06:00:00 2000
Fri Dec 1 06:00:00 2000
Mon Jan 1 06:00:00 2001
Thu Feb 1 06:00:00 2001
Thu Mar 1 06:00:00 2001
Sun Apr 1 06:00:00 2001
Tue May 1 06:00:00 2001
Fri Jun 1 06:00:00 2001
Sun Jul 1 06:00:00 2001
Wed Aug 1 06:00:00 2001
Sat Sep 1 06:00:00 2001
Mon Oct 1 06:00:00 2001
Thu Nov 1 06:00:00 2001
Sat Dec 1 06:00:00 2001
Tue Jan 1 06:00:00 2002
Fri Feb 1 06:00:00 2002
Fri Mar 1 06:00:00 2002
Mon Apr 1 06:00:00 2002
Wed May 1 06:00:00 2002
Sat Jun 1 06:00:00 2002
Mon Jul 1 06:00:00 2002
Thu Aug 1 06:00:00 2002
Sun Sep 1 06:00:00 2002
Tue Oct 1 06:00:00 2002
Fri Nov 1 06:00:00 2002
Sun Dec 1 06:00:00 2002
Wed Jan 1 06:00:00 2003
Sat Feb 1 06:00:00 2003
Sat Mar 1 06:00:00 2003
Tue Apr 1 06:00:00 2003
Thu May 1 06:00:00 2003
Sun Jun 1 06:00:00 2003
Tue Jul 1 06:00:00 2003
Fri Aug 1 06:00:00 2003
Mon Sep 1 06:00:00 2003
Wed Oct 1 06:00:00 2003
Sat Nov 1 06:00:00 2003
Mon Dec 1 06:00:00 2003
Thu Jan 1 06:00:00 2004
Sun Feb 1 06:00:00 2004
Mon Mar 1 06:00:00 2004
Thu Apr 1 06:00:00 2004
Sat May 1 06:00:00 2004
Tue Jun 1 06:00:00 2004
Thu Jul 1 06:00:00 2004
Sun Aug 1 06:00:00 2004
Wed Sep 1 06:00:00 2004
Fri Oct 1 06:00:00 2004
Mon Nov 1 06:00:00 2004
Wed Dec 1 06:00:00 2004
Sat Jan 1 06:00:00 2005

I'm working on automatic tests for libphoenix time functions: phoenix-rtos/phoenix-rtos-tests#255 so I'll try to add this as a test case.

[anglov]

I believe this task reamains open because of other possibilities to year changes than month. Original normalisation issue was fixed here:
#29

I'm really grateful that you add this as test (as this kind of things are crucial and problems can be really painful) but I think proper test should also validate other cases. Please check #28 (comment)