DB2 SQL：函数在SELECT语句中起作用，但在WHERE子句中不起作用

Question

I have a table set up like this:

**dNum | dType | dCode | dChar**
    1  |  blah |  foo  | xxxx
    1  |  blah |  foo2 | 10/11/2016
    2  |  blah |  foo  | xxxx
    2  |  blah |  foo2 | 10/19/2016

... and so on.

For each unique dNum I need to find when dCode equals "foo2" and then when dChar is exactly 15 days ago. My problem is in getting dates from exactly 15 days ago. My current code is as follows:

SELECT dNum
,dChar
FROM thing.table boo
WHERE dCode = 'foo2' AND
      DAYS(CURRENT_DATE) - DAYS(DATE(dChar)) = 15

When I try to run this query I get the following error: "Value in date, time, or timestamp string not valid." What I'm even more confused about, however, is that when I put DAYS(CURRENT_DATE) - DAYS(DATE(dChar)) as a field in my SELECT statement but not in the WHERE clause, that it works just fine. The following code DOES work for me:

SELECT dNum
,dChar
,DAYS(CURRENT_DATE) - DAYS(DATE(dChar)) AS TEST1
FROM thing.table boo
WHERE dCode = 'foo2'

dChar is a CHAR(25) variable, and not all values are dates, but again, that doesn't seem to affect anything when my code is not in the WHERE clause.

Can anybody help me to figure out what is going on here?

Answer 1

What happens if you change the where code to:

WHERE dCode = 'foo2' AND
      (case when dcode = 'foo2'
            then DAYS(CURRENT_DATE) - DAYS(DATE(dChar))
       end) = 15

I realize this is very arcane, but it guarantees that the function is only evaluated when dcode = 'foo2' . What might be happening is that the values are all correct when this condition is true, but problems lie elsewhere.

Answer 2

Apparently, dChar values in rows where dCode equals "foo2" are valid representations of dates. Since the select list is processed after all WHERE predicates have been applied, this does not cause any issues.

If, on the other hand, you put DATE(dChar) in the WHERE clause it may encounter dChar values that can't be converted to dates.

The solution suggested by Gordon Linoff may or may not work (likely the latter), depending on how the optimizer decides to handle the CASE expression.

Edit:

Your question was "what's going on?", so now you know. If you want to go one step further and find a reliable workaround, one option would be to create a user-defined function that would gracefully handle invalid dates and use it in place of DATE() . The function might look something like

create function to_date_safe (str varchar(20)) 
returns date 
deterministic 
no external action contains sql 
begin 
  declare continue handler for sqlstate '22007' -- on conversion error
    return null; 
  return date(str); 
end

Your WHERE clause would then look like

WHERE dCode = 'foo2' AND
  DAYS(CURRENT_DATE) - DAYS(to_date_safe(dChar)) = 15

You may need to replace '22007' with the actual SQLSTATE value that is returned to you on conversion error.

Answer 3

If you filter twice, using nested queries, you'll avoid the problem (trying to parse non date strings before they're filtered out).

SELECT
    *
FROM
(
    SELECT
        dNum,
        dChar
    FROM
        thing.table boo
    WHERE
        dCode = 'foo2'
        AND dChar <> ''
)
    foo2
WHERE
    DAYS(CURRENT_DATE) - DAYS(DATE(dChar)) = 15

EDIT

Or, even better, reorder your query. Convert your date into a string, not the other way around. Then indexes can be used, and non-date format values don't cause issues.

WHERE
        dCode = 'foo2'
    AND dChar = CHAR(CURRENT_DATE - 15 DAYS)