I don't see any, but before we discuss using date as a primary key, let's make a couple of assumptions about the dates that we're planning to store in our table.
First, let's assume that we're using the Gregorian calendar. As you may know, this calendar was adopted to replace the Julian calendar in most Western countries, but they did not all do it on the same day. The change has taken hundreds of years, between 1582 and the twentieth century, and during that time, what was May 13th for one person was May 24th for someone else. So the second assumption is that we're going to store dates in our database only from the current era. This is because a date in any of those earlier years is not enough to pin down exactly what day it was; more information would be needed, such as whether it is a Gregorian or Julian calendar date.
You see, that is the purpose of a primary key: to give identity to a real-world entity (thing, event, etc.). What we are identifying when we use a date is a particular day.
Let me give you a real example. Suppose we were interested in the day when we will next have a full moon on an equinox. We may not know what date that is, but it definitely is a single, particular day. It has identity. Even if we don't know what its date is, we do know that it exists (has identity in the real world, albeit in the future) because we know enough about celestial mechanics to be 100% positive that there will definitely be another full moon on an equinox, some day in the future.
With me so far?
Okay, now, the only two requirements for a primary key are:
- it is never null
- it uniquely identifies each row in the table
On this basis, a date makes an excellent primary key, provided we know which calendar we're using. And if we assume Gregorian, and we don't need to store historical dates, we're fine. Note that with date as the primary key, we could not enter the day of the next full moon on an equinox into our new table, until we actually compute the date, which cannot be null since it's the primary key.
Okay, you did ask whether there would be any problems in the long term. In the long term, we could theoretically change calendar systems. We could have thirteen months of twenty-eight days—accountants would love that setup—with one or two extra "intra-annual" days thrown in each year to make the total come out to 365 or 366 as necessary. Nothing stopping us. Well, except perhaps the worldwide investment in existing computer systems. You thought getting ready for Y2K was expensive? Imagine switching calendars! My personal bet is that we will never get off the Gregorian calendar, at least not before you and I and everyone we know is long, long gone, and intelligent computers that can reprogram themselves have been invented.
Of course, if you want to be safe, to ensure that your database survives even beyond that far off day, you could use a surrogate key. A surrogate key is an artificial key, like a distinct integer, which takes the place of the primary key. This way, the "real" key, the date in this instance, would be stored in a lookup table, and you would reference all rows to the lookup table using the surrogate key, to find out what the date is. Thus, you would store Fred's birthday not as 2937-12-23 but as something like 4529374.
Then, if the date ever has to change, like in a change from Gregorian to some new calendar system, e.g., from 2937-12-23 to 2937-13-21, then you would simply update the lookup table and none of the related tables would have to be updated because they are using the surrogate key instead! Fred's birthday would still be 4529374!
This (using a surrogate key) is exactly the advice many DBAs will give you. I should emphasize at this point that I am not a DBA, I am a data modeller, and in my opinion, using dates as a primary key is perfectly safe, at least as long as the original assumptions are true, that they are Gregorian dates and do not include ambiguous dates from the past. If those assumptions are true, using a surrogate key would be silly.
This was first published in November 2005