Why isn't SQL Server letting me store '21/04/17' as a date? - sql

Question

I've got a table that currently has all columns stored as nvarchar(max), so I'm converting all the datatypes to be what they should be. I have a column of dates, however when I run this:
ALTER TABLE Leavers ALTER COLUMN [Actual_Termination_Date] date;
I get
"Conversion failed when converting date and/or time from character string".
This is relatively normal, so I did the following to investigate:
SELECT DISTINCT TOP 20 [Actual_Termination_Date]
FROM LEAVERS
WHERE ISDATE([Actual_Termination_Date]) = 0
which returned:
NULL
13/04/2017
14/04/2017
17/04/2017
19/04/2017
21/04/2017
23/04/2017
24/04/2017
26/04/2017
28/04/2017
29/03/2017
29/04/2017
30/04/2017
31/03/2017
42795
42797
42813
42817
42820
42825
The null and excel style date formats (e.g. 42795) are no problem, however it's the ones appearing as perfectly normal dates I'm having a problem with. I usually fix issues like this by using one of the following fixes:
SELECT cast([Actual_Termination_Date] - 2 as datetime)
FROM LEAVERS
WHERE ISDATE([Actual_Termination_Date]) = 0
or
SELECT cast(convert(nvarchar,[Actual_Termination_Date], 103) - 2 as datetime)
FROM LEAVERS
WHERE ISDATE([Actual_Termination_Date]) = 0
When these return back the dates as I would expext, I'd then do an UPDATE statement to change them in the table and then convert the column type. However I keep getting an error message telling me that various dates can't be converted such as:
Conversion failed when converting the nvarchar value '21/04/2017' to data type int.
Any thoughts? Thanks!

Answer 1

Probably because of your language setting. For '21/04/2017' to work, you'll need to be using the BRITISH language, or other language that uses dd/MM/yyyy. I suspect you are using ENGLISH which is actually American.
American's use MM/dd/yyyy meaning that '21/04/2017' would mean the 4th day of the 21st month in the year 2017; obviously that doesn't work.
The best method is to use an unambiguous format, regardless of language and data type. For SQL Server that's yyyyMMdd and yyyy-MM-ddThh:mm:ss.nnnnnnn (yyyy-MM-dd and yyyy-MM-dd hh:mm:ss.nnnnnnn are not unambiguous in SQL Server when using the older datetime and smalldatetime data types).
Otherwise you can use CONVERT with a style code:
SELECT CONVERT(date,'21/04/2017', 103)
The problem with your data, however, is that you have values that are in the format dd/MM/yyyy and integer values. The int (not varchar) value 42817 as a datetime in SQL Server is 2017-03-25. On the other hand, if this data came from Excel then the value is 2017-03-23. I am going to assume the data came from Excel, not SQL Server (because the ACE drivers have a habit of reading dates as numbers, because the thing they aren't is "ace").
You'll need to therefore convert the values to an unambiguous format first, so that'll be yyyyMMdd. As we have 2 different types of values, this is a little harder, but still possible:
UPDATE dbo.Leavers
SET Actual_Termination_Date = CONVERT(varchar(8), ISNULL(TRY_CONVERT(date, Actual_Termination_Date, 103), DATEADD(DAY, TRY_CONVERT(int, Actual_Termination_Date),'18991230')), 112);
Then you can alter your table:
ALTER TABLE dbo.Leavers ALTER COLUMN [Actual_Termination_Date] date;
DB<>Fiddle using Michał Turczyn's DML statement.

Answer 2

Put the column into a canonical format first, then convert:
update leavers
set Actual_Termination_Date = try_convert(date, [Actual_Termination_Date], 103);
ALTER TABLE Leavers ALTER COLUMN [Actual_Termination_Date] date;
The update will do an implicit conversion from the date to a string. The alter should be able to "undo" that implicit conversion.
Back up the table before you do this! You are likely to discover that some dates are not valid -- that is pretty much the rule when you store dates as strings although in a small minority of cases, all date strings are actually consistently formatted.

Answer 3

The actual date does not matter. The error happens when you try to subtract 2 from a string:
[Actual_Termination_Date] - 2
The clue comes from the error message:
Conversion failed when converting the nvarchar value '21/04/2017' to data type int.
To fix the problem, use DATEADD after the conversion:
SELECT DATEADD(days, -2, convert(datetime, [Actual_Termination_Date], 103))

Answer 4

You just have inconsistent date format within your column, which is terrible.
Having wrong datatype lead to it, that's why it is so important to have proper data types on columns.
Let's investigate it a little:
-- some test data
declare #tbl table (dt varchar(20));
insert into #tbl values
(NULL),
('13/04/2017'),
('14/04/2017'),
('17/04/2017'),
('19/04/2017'),
('21/04/2017'),
('23/04/2017'),
('24/04/2017'),
('26/04/2017'),
('28/04/2017'),
('29/03/2017'),
('29/04/2017'),
('30/04/2017'),
('31/03/2017'),
('42795'),
('42797'),
('42813'),
('42817'),
('42820'),
('42825');
-- here we handle one format
select convert(date, dt, 103) from #tbl
where len(dt) > 5
or dt is null
-- here we handle excel like format
select dateadd(day, cast(dt as int), '1900-01-01') from #tbl
where len(dt) = 5
So, as you can see you have to apply to different approaches for this task. CASE WHEN statement should fit here nicely, see below SELECT:
select case when len(dt) = 5 then
dateadd(day, cast(dt as int), '1900-01-01')
else convert(date, dt, 103) end
from #tbl