This is whole story how date problem was and how Big DBMSs handled these problems.
During the period between 1 A.D. and today, the Western world has actually used two main calendars: the Julian calendar of Julius Caesar and the Gregorian calendar of Pope Gregory XIII. The two calendars differ with respect to only one rule: the rule for deciding what a leap year is. In the Julian calendar, all years divisible by four are leap years. In the Gregorian calendar, all years divisible by four are leap years, except that years divisible by 100 (but not divisible by 400) are not leap years. Thus, the years 1700, 1800, and 1900 are leap years in the Julian calendar but not in the Gregorian calendar, while the years 1600 and 2000 are leap years in both calendars.
When Pope Gregory XIII introduced his calendar in 1582, he also directed that the days between October 4, 1582, and October 15, 1582, should be skipped—that is, he said that the day after October 4 should be October 15. Many countries delayed changing over, though. England and her colonies didn't switch from Julian to Gregorian reckoning until 1752, so for them, the skipped dates were between September 4 and September 14, 1752. Other countries switched at other times, but 1582 and 1752 are the relevant dates for the DBMSs that we're discussing.
Thus, two problems arise with date arithmetic when one goes back many years. The first is, should leap years before the switch be calculated according to the Julian or the Gregorian rules? The second problem is, when and how should the skipped days be handled?
This is how the Big DBMSs handle these questions:
- Pretend there was no switch. This is what the SQL Standard seems to require, although the standard document is unclear: It just says that dates are "constrained by the natural rules for dates using the Gregorian calendar"—whatever "natural rules" are. This is the option that DB2 chose. When there is a pretence that a single calendar's rules have always applied even to times when nobody heard of the calendar, the technical term is that a "proleptic" calendar is in force. So, for example, we could say that DB2 follows a proleptic Gregorian calendar.
- Avoid the problem entirely. Microsoft and Sybase set their minimum date values at January 1, 1753, safely past the time that America switched calendars. This is defendable, but from time to time complaints surface that these two DBMSs lack a useful functionality that the other DBMSs have and that the SQL Standard requires.
- Pick 1582. This is what Oracle did. An Oracle user would find that the date-arithmetic expression October 15 1582 minus October 4 1582 yields a value of 1 day (because October 5–14 don't exist) and that the date February 29 1300 is valid (because the Julian leap-year rule applies). Why did Oracle go to extra trouble when the SQL Standard doesn't seem to require it? The answer is that users might require it. Historians and astronomers use this hybrid system instead of a proleptic Gregorian calendar. (This is also the default option that Sun picked when implementing the GregorianCalendar class for Java—despite the name, GregorianCalendar is a hybrid calendar.)