Ed Reingold noticed that SunOS 4.1.2 mishandles the time zone transitions in Brazil. Typically, it claims that the transitions occur on Friday, whereas the rule says they should occur on Saturday. The bug is still present in tz93c (the 1993-06-06 version of the time zone code). I looked into the problem, and discovered that zic botches any rule whose ON field is of the form `DAY<=N', where DAY is not Sunday, and where the Nth day of the current year is not DAY. In this case, it acts as if you specified the previous day to DAY. As it happens, the only rules of this form in the database are those of Brazil. For example if you run `zdump -v Brazil/East' you'll see that most of the transitions occur on Friday, but they're all supposed to occur on Saturday. Here is a patch to tz93c. I'm amazed that this bug has never been reported in all these years. Shall we chalk it up to Brazil's strict import restrictions on American software? =================================================================== RCS file: zic.c,v retrieving revision 7.8 retrieving revision 7.8.1.1 diff -c -r7.8 -r7.8.1.1 *** zic.c 1993/06/06 23:31:38 7.8 --- zic.c 1993/06/18 02:58:46 7.8.1.1 *************** *** 1819,1825 **** } else { dayoff = oadd(dayoff, (long) -1); if (--wday < 0) ! wday = LDAYSPERWEEK; --i; } if (i < 0 || i >= len_months[isleap(y)][m]) { --- 1819,1825 ---- } else { dayoff = oadd(dayoff, (long) -1); if (--wday < 0) ! wday = LDAYSPERWEEK - 1; --i; } if (i < 0 || i >= len_months[isleap(y)][m]) {