Suppose you're writing a query to find all the invoices that were written on January 6, 2003. You know from the control totals that 122 invoices were written that day. But when you run this query:

SELECT * FROM Invoices
  WHERE InvoiceDate = '2003-01-06'

The result set is empty. What's going on?

How dates and times are stored in SQL Server

Before you can effectively query date/time (or temporal) data, you have to know something about how date/time values are stored. SQL Server supports two date/time data types: datetime and smalldatetime. The difference between the two is the amount of storage used. Datetime uses 8 bytes of storage, while smalldatetime uses only 4 bytes. For this reason, datetime can represent date/time values within a wider range and with more precision than smalldatetime. These differences are summarized in the table below.

Both datetime and smalldatetime represent the date and time as a value that's equal to the number of days in relationship to a base date. In SQL Server, that base date is midnight on January 1, 1900. As you can see in the table, the smalldatetime type can only represent dates from this base date on. In contrast, the datetime type can also represent dates that are before January 1, 1900. To do that, it stores those values as negative numbers.

To visualize how date/time values are stored, you can think of them as consisting of two parts. The integer portion represents the number of whole days since January 1, 1900. The fractional portion represents the fraction of a day that's passed since midnight. For example, the date/time value representing noon on January 4, 1900 is stored as 3.5. In this case, 3 represents three full days since the base date and 0.5 represents one half of a day between midnight and noon. To see this, submit the following query:

SELECT CAST(CAST('1900-01-04 12:00' AS 
  datetime) AS float)

Note: The CAST function explicitly changes the data type of a value as specified. So in this statement, the inner CAST changes the string literal '1900-01-04 12:00' to a value of data type datetime. Then, the outer CAST changes that datetime value to a value of data type float. The final result is a floating-point representation of the datetime value that represents noon on January 4, 1900.

So far, so good. But the problems that crop up in querying date/time data are caused by confusion over two fundamental facts that aren't so obvious. First, date/time data types are approximate numerics, not exact numerics. Second, date/time data types can't store a date without a time or a time without a date.

Date/Time values are approximate numerics

Datetime and smalldatetime are like the floating-point data types, float and real, in that they're approximate numerics. That means the value retrieved from SQL Server may be different from the value that was originally stored. For example, if you store the expression 10/3.0 in a column of data type float, you'll retrieve a value 3.3333330000000001. Although this is a reasonable representation of ten thirds, it's not exact since it's rounded past the 6^th digit. In fact, if you add three such values together, you get 9.9999990000000007, not 10. Of course, most programmers understand this as a rounding error. And it's a persistent problem for all digital computers, not just those running SQL Server. Still, you need to be aware of it as you code search conditions.

In contrast, when working with exact numeric data, the value retrieved from SQL Server is exactly the value that was originally stored. For example, if you store 10/3.0 in a column of data type int, it's stored as 3 and retrieved as 3. In this case, SQL Server implicitly casts the result of the expression as a real value, 3.333333. Then, SQL Server implicitly casts 3.333333 as an integer because it's being stored in a column of type int. Although this is still a rounding error, it occurs before the value is stored, not as a result of the physical limitations of computer storage. In other words, the error was introduced by using the wrong data type, not by the inherent limitation of the data type itself. Since the system always returns the same value as was stored, the data type is exact.

Now, to see how this affects date/time values, consider the date and time value for 8:00AM on January 4, 1900. As you saw above, noon on this day is stored as 3.5, or halfway through the fourth day. In contrast, 8:00AM is one third of the way through the day, so its representation will be approximate. To see this for yourself, submit the following query:

SELECT CAST(CAST('1900-01-04 08:00' AS 
  datetime) AS float)

You'll get the following result:

3.3333333333333335

But if you submit this query:

SELECT CAST(3.3333333 AS datetime), 
  CAST(3.3333334 AS datetime)

you'll get the following results:

1900-01-04 07:59:59.997             
  1900-01-04 08:00:00.003

As you can see, these three values are all quite close. In fact, they're close enough to be considered 8:00AM for most applications. However, in a search condition based on a single value, such as:

WHERE (DTValue = '1900-01-04 08:00')

you'd only match those rows where the stored value exactly matches 3.3333333333333335. You'll see how to get around this later in this article.

Dates without times and times without dates

SQL Server doesn't provide data types for storing just the date or just the time. So if you store a date/time value without an explicit time, the fractional portion of the value is set to zero. This represents midnight as 00:00:00. Similarly, if you store a date/time value without an explicit date, the integer portion of the value is set to zero. This represents the date January 1, 1900. To see this, submit the following query:

SELECT CAST('1900-01-04' AS datetime), 
  CAST('10:00' AS datetime)

which returns the following result:

1900-01-04 00:00:00.000              
  1900-01-01 10:00:00.000

Whether you can ignore the date or the time component when you query a date/time column depends on how the column has been designed and used.

The effect of database design on querying

Database designers don't always use date/time columns appropriately. At the time the database is designed, each date/time column should be identified as to whether it will store both dates and times, dates only, or times only. The designer, by using defaults, constraints, and triggers, can enforce these rules to prevent the accidental storage of data that are either unnecessary or not applicable.

For example, a column in an accounts payable system for the date an invoice is received is unlikely to need the time. In that case, the designer should plan to use the column solely for dates and never store the time component. A trigger could be assigned to prevent the non-integer portion of the date value from being stored when updating or inserting.

Generally, however, the programmer is forced to work with an existing database. In this case, you should examine the way in which the date/time values are being used before you assume the designer did his or her job correctly.

The simplest way to do that is to submit a query using a search condition similar to the following, where DT is the date/time column in question:

WHERE CAST(FLOOR(CAST(DT AS float))AS 
  datetime) = 0 OR
        DT - CAST(FLOOR(CAST(DT AS float))AS datetime) 
  = 0

Note: The FLOOR function returns the largest integer that is less than or equal to the specified value. In this expression, FLOOR is applied to the floating-point representation of the DT column. This simply strips off the fractional portion of the number.

The first expression returns the date (integer) portion of the value, while the second returns the time portion. If this query returns no rows, it's likely that the column has been used consistently to store both dates and times, since the date is never 0 and the time is never 0.

Keep in mind, of course, that if the above query returns rows, it doesn't necessarily imply that the column has been used inconsistently. If the time happens to be exactly midnight or the date happens to be January 1, 1900, then it'll show up in the result set. In that case, you can test for columns with time-only or date-only data by using either of these two queries:

WHERE TOnly <> Tonly - 
  (CAST(FLOOR(CAST(TOnly AS float))AS datetime))
  
  WHERE DOnly <> CAST(FLOOR(CAST(DOnly AS 
  float))AS datetime)

Here, TOnly and DOnly are date/time columns that you expect contain only times or dates, respectively. If the query returns rows, then those rows don't contain the type of data you expected.

Determining what kind of data are stored in the date/time columns of each table is important for intelligent querying. If the columns are used consistently, then your job is easier. However, even if the columns are used inconsistently, you'll at least know which query pitfalls to watch out for as you code your queries.

Performance considerations in querying

A search based on an indexed column completes faster than a search based on a non-indexed column. So date/time columns that are searched frequently should be indexed. Be aware, though, that if you then use a function in the search condition, the index can't be used in the same way, which slows performance. For searches that are executed thousands of times a day on a production database, this can cause significant performance problems. For this reason, you should avoid using functions in such search conditions whenever possible. As you'll see in the examples that follow, this sometimes results in solutions that are less flexible than those that use functions.

In addition, keep in mind that some applications require that you search for portions of a date/time column. The portion could be date only, time only, or even a smaller portion, such as a year or hour. In that case, it may improve performance to split a single date/time column into two or more separate columns, and then index those that are searched most often.

How to search by date

Frequently. you'll need to search a date/time column for a specific date, regardless of time. If the data in the column have been used consistently with the time component set to zero, that's no problem. You just search for the date you're looking for.

But consider the following table, called DateSample:

ID  DateVal
  --  -----------------------
  1   2001-02-28 10:00:00.000
  2   2002-02-28 13:58:32.823
  3   2002-02-29 00:00:00.000
  4   2002-02-28 00:00:00.000

As you can see, the DateVal column is used inconsistently. The third and fourth values indicate that the column might have been intended to store dates only, but the first two values indicate that this wasn't enforced.

As a result, if you use the following query to retrieve rows with the date February 28, 2002:

SELECT * FROM DateSample
  WHERE DateVal = '2002-02-28'

the result set includes only row 4 instead of both rows 2 and 4. That's because the date literal is implicitly cast as a datetime value which, in this case, has a zero time component. Since this doesn't exactly match the value in row 2, that row isn't returned.

How can you get around the time component? If the query is run often, you should base the search on a range of values, as in:

SELECT * FROM DateSample
  WHERE DateVal BETWEEN '2002-02-28' AND '2002-02-28 23:59:59.997'

Remember that the BETWEEN clause retrieves values that are equal to the upper and lower limits, so you can't code the upper limit as just '2002-02-29'. If you do, then you'll incorrectly retrieve row 3. Another way to get the same result is to use comparison operators:

SELECT * FROM DateSample
  WHERE DateVal >= '2002-02-28' AND DateVal < '2002-02-29'

If the query is run infrequently (to produce a report only once a month, for instance), you can code an expression in the WHERE clause that strips the date/time value of its fractional component. For example, this query:

SELECT * FROM DateSample
  WHERE CAST(FLOOR(CAST(DateVal AS float)) AS datetime) = '2002-02-28'

returns both rows 2 and 4. In addition, there are many other expressions that you can use to accomplish this same result (my SQL book, Murach's SQL for SQL Server, covers a couple of others).

By the way, if you wished to retrieve rows with the day February 28, regardless of year, you could code the following query:

SELECT * FROM DateSample
  WHERE MONTH(DateVal) = 2 AND DAY(DateVal) = 28

which retrieves rows 1, 2, and 4. Since there isn't a way to accomplish this without using one or more functions, however, this query shouldn't be run frequently against a production database. If you need to perform this kind of search on a query that runs often, you should change the design of the database, if possible. Then, you can create a separate, indexed column to store the portion of the date/time value that you need to search.

How to search by time

Searching a column for a specific time, regardless of date, is similar to searching for date-only values. If the column consistently stores just the time portion, then searching the data is simplified. However, unlike date values, the time value is represented by an approximate numeric. So even when the date portion can be ignored, you must still consider rounding errors.

To illustrate time-only searches, consider following table, called TimeSample:

ID  TimeVal
  --  -----------------------
  1   2002-02-28 10:00:00.000
  2   1900-01-01 13:58:32.823
  3   1900-01-01 09:59:59.997
  4   1900-01-01 10:00:00.000

Here, the TimeVal column is used inconsistently, sometimes storing time only and sometimes storing both date and time. So if you use the following query to retrieve rows with the time 10:00AM:

SELECT * FROM TimeSample
  WHERE TimeVal = '10:00:00'

you only get row 4. Row 1 isn't retrieved because the date literal is implicitly cast as a datetime value with a zero date component, which doesn't match the date component of row 1. In addition, row 3 isn't retrieved because this value is close to, but not equal to, 10:00AM.

To ignore the date component of a column, you can code an expression that strips the date/time value of its integer component, such as:

SELECT * FROM TimeSample
  WHERE TimeVal - CAST(FLOOR(CAST(TimeVal AS float)) AS datetime) = '10:00'

which returns rows 1 and 4. Unfortunately, there's no way to accomplish this without using one or more functions. For this reason, it's critical to store time-only data correctly in the first place. If you need to do this kind of search often, you should, if possible, change the database design.

To search for time values that are approximately equal is simply a matter of coding a range of values. If the time-only data are stored consistently without the date component, you can use a query like this:

SELECT * FROM TimeSample
  WHERE TimeVal BETWEEN '09:59' AND '10:01'

or

SELECT * FROM TimeSample
  WHERE TimeVal > '09:59' AND TimeVal < '10:01'

Both of these queries return rows 3 and 4. Of course, you have to decide what literal values to use for the range of approximation that you prefer.

If the time-only data are stored inconsistently, then you need to accommodate both non-zero date components and a range of time values. For instance, you can use a query like this:

SELECT * FROM TimeSample
  WHERE TimeVal - CAST(FLOOR(CAST(TimeVal AS float)) AS datetime) > '09:59'
    AND TimeVal - CAST(FLOOR(CAST(TimeVal AS float)) AS datetime) < '10:01'

which returns rows 1, 3, and 4. Again, though, there's no way to accomplish this without the use of a function. So you may need to change the database design, if you can.

One other way to approximate time values is to use the smalldatetime data type rather than the datetime data type in the original table. Since smalldatetime always rounds the time portion to the nearest minute, times in the range from 09:59:29.999 to 10:00:29.998 are stored as 10:00. If approximation to the nearest whole minute is sufficient for your application, then using smalldatetime will prevent the need to search for a range of values.