How SQL's convert function work when converting datetime to float? - sql

Question

why the out put of this query:
declare #currentDate as datetime
set #currentDate ='01/07/2010'
select convert(float, #currentdate)
...is 40183 ?
So for those who are getting confuse with my question, my question is How to know the result of above query without executing it ?

Answer 1

DateTime is often represented as a day count from a pre-determined date (generally know as the epoch) on the integer part and the percentage of the day elapsed since mid-night on the fractional part.
SQL Server is not the exception to this, thus the conversion to Float makes a lot of sense. Day 0 is Jan 01 1900 00:00:00 (AFAIK, in no particular time-zone, so you shall consider it "local time").
So, you can try this:
declare #ADate DateTime;
set #ADate = '19000101 00:00:00';
select CONVERT(float, #ADate); --should print 0
set #ADate = '19000101 12:00:00';
select CONVERT(float, #ADate); --should print 0.5
set #ADate = '19001231 06:00:00';
select CONVERT(float, #ADate); --should print 364.25
So, for your results, 40183 days has been passed since 01/01/1900 00:00:00 and 01/07/2010 00:00:00
Clarification: Unix like systems use a different approach to store datetimes: Seconds since Unix epoch (Jan 1 1970 00:00:00 UTC), which is more known as epoch time.
[Edit]
Date format on this response was changed to YYYYMMDD format on 20140416, after some new years of experience with SQL Server (and as #Damien said in his comment) this is the only safe format.

Answer 2

DateTime values are actually stored as two four-byte integers under the hood. The first four-byte integer represents the number of days since 1900-01-01. The second four-byte integer stores the number of milliseconds since midnight. When you convert a datetime into a float, the decimal portion represents the percentage of the 24-day that has past. Thus, 0.5 represents noon.

Answer 3

It's basically converting the datetime to an OLE Date. There's a decent description of the process in the documentation for System.DateTime.ToOADate():
http://msdn.microsoft.com/en-us/library/system.datetime.tooadate.aspx
The quick explanation is that the integer part is the number of days since 12/30/1899. The fractional part (zero in this case) is the time divided by 24.

Answer 4

This should help you understand the TSQL implementation (or implement your own):
DECLARE
#date DATETIME = '20180125 09:15:30.549',
#date_dec DECIMAL (26,10) = 43123.3857702546
SELECT
CAST(#date_dec AS DATETIME) AS [TSQL cast to DATETIME],
CAST(#date AS DECIMAL (26,10)) AS [TSQL cast to DECIMAL]
SELECT
DATEADD(DAY, FLOOR(#date_dec),
DATEADD(HOUR, FLOOR(#date_dec % 1 * 24),
DATEADD(MINUTE, FLOOR((#date_dec % 1 * 24) % 1 * 60),
DATEADD(SECOND, FLOOR(((#date_dec % 1 * 24) % 1 * 60) % 1 * 60),
DATEADD(MILLISECOND, FLOOR((((#date_dec % 1 * 24) % 1 * 60) % 1 * 60) % 1 * 1000), '19000101')
)
)
)
) AS [Manual cast to DATETIME],
DATEDIFF(DAY, '19000101', #date)
+ (
DATEPART(HOUR, #date)
+ (
DATEPART(MINUTE, #date)
+ (
DATEPART(SECOND, #date)
+ DATEPART(MILLISECOND, #date) / CAST(1000 AS FLOAT)
) / CAST(60 AS FLOAT)
) / CAST(60 AS FLOAT)
) / CAST(24 AS DECIMAL (26,10)) AS [Manual cast to DECIMAL]
Note that the result is not always the same as TSQL loses precision on last millisecond digit.