not only from a date, but also from an interval.

1) What does extract?
2) What I want as a result?

Reply

1) from what I understand, it extracts the month from a date,
so for example, EXTRACT (MONTH FROM to_date (’01-Mar-2010 ‘)) = 3
then:

SELECT EXTRACT(MONTH FROM to_date('01-mar-2010')) AS month1_ext ,
       EXTRACT(MONTH FROM to_date('05-jan-2010')) AS month2_ext
from dual

MONTH1_EXT             MONTH2_EXT              
---------------------- ---------------------- 
3                      1                     

it does is simply, 3 – 1 = 2, then in my opinion should not be
used when we want the difference between two dates.

2)
if I have to do a temporal analysis on the data expressed in month, I have two situations,
I think. Compare fixed dates (eg the first of each month), or to compare between any dates.
In the first case, I should not have problems using Extract. In the second case, should be
established a rule. Example: from 1 to 15 of every month as the first of this month, from
16 onwards consider the first of next month.In the absence of rules, I do not think you can
use Extract and I do not think it makes sense to have a difference between two
dates in months(1.5 months is a month or two months ???).

Regards

That is certainly an improvement for the first day of the month:

D1        D2        DATE_DELTA T
--------- --------- ---------- ------------------
01-JAN-10 02-JAN-10          1 0 years, 0 months.
01-JAN-10 03-JAN-10          2 0 years, 0 months.
01-JAN-10 04-JAN-10          3 0 years, 0 months.
01-JAN-10 05-JAN-10          4 0 years, 0 months.
01-JAN-10 06-JAN-10          5 0 years, 0 months.
01-JAN-10 07-JAN-10          6 0 years, 0 months.
01-JAN-10 08-JAN-10          7 0 years, 0 months.
01-JAN-10 09-JAN-10          8 0 years, 0 months.
01-JAN-10 10-JAN-10          9 0 years, 0 months.
01-JAN-10 11-JAN-10         10 0 years, 0 months.
01-JAN-10 12-JAN-10         11 0 years, 0 months.
01-JAN-10 13-JAN-10         12 0 years, 0 months.
01-JAN-10 14-JAN-10         13 0 years, 0 months.
01-JAN-10 15-JAN-10         14 0 years, 0 months.
01-JAN-10 16-JAN-10         15 0 years, 0 months.
01-JAN-10 17-JAN-10         16 0 years, 0 months.
01-JAN-10 18-JAN-10         17 0 years, 0 months.
01-JAN-10 19-JAN-10         18 0 years, 0 months.
01-JAN-10 20-JAN-10         19 0 years, 0 months.
01-JAN-10 21-JAN-10         20 0 years, 0 months.
01-JAN-10 22-JAN-10         21 0 years, 0 months.
01-JAN-10 23-JAN-10         22 0 years, 0 months.
01-JAN-10 24-JAN-10         23 0 years, 0 months.
01-JAN-10 25-JAN-10         24 0 years, 0 months.
01-JAN-10 26-JAN-10         25 0 years, 0 months.
01-JAN-10 27-JAN-10         26 0 years, 0 months.
01-JAN-10 28-JAN-10         27 0 years, 0 months.
01-JAN-10 29-JAN-10         28 0 years, 0 months.
01-JAN-10 30-JAN-10         29 0 years, 0 months.
01-JAN-10 31-JAN-10         30 0 years, 0 months.
01-JAN-10 01-FEB-10         31 0 years, 1 months.
01-JAN-10 02-FEB-10         32 0 years, 1 months.

It appears that the solution you provided, and possibly any solution using EXACT, still has issues. For example, if you check January 5, you will see that the number of months advances when February 1 appears for the D2 column.

D1        D2        DATE_DELTA T
--------- --------- ---------- ------------------
...
05-JAN-10 30-JAN-10         25 0 years, 0 months.
05-JAN-10 31-JAN-10         26 0 years, 0 months.
05-JAN-10 01-FEB-10         27 0 years, 1 months.
05-JAN-10 02-FEB-10         28 0 years, 1 months.
05-JAN-10 03-FEB-10         29 0 years, 1 months.
05-JAN-10 04-FEB-10         30 0 years, 1 months.
05-JAN-10 05-FEB-10         31 0 years, 1 months.
...
05-JAN-10 27-FEB-10         53 0 years, 1 months.
05-JAN-10 28-FEB-10         54 0 years, 1 months.
05-JAN-10 01-MAR-10         55 0 years, 2 months.
05-JAN-10 02-MAR-10         56 0 years, 2 months.
05-JAN-10 03-MAR-10         57 0 years, 2 months.
05-JAN-10 04-MAR-10         58 0 years, 2 months.
05-JAN-10 05-MAR-10         59 0 years, 2 months.

It might simply be a problem with the precision of the EXACT functionality, as mentioned in the comment by Gary.

(Edit: In the comment I originally described how the number of months advances when January 5 is the start date (column D1) and February 1 appears in the D2 column. Unfortunately, the code output section of the comment showed what happened when March 1 appeared in the D2 column. I added the January 5/February 1 output to the code output section in the comment.)

WITH DATES AS (
SELECT
  TO_DATE('01-JAN-2010','DD-MON-YYYY') + (ROWNUM-1) D
FROM
  DUAL
CONNECT BY
  LEVEL<=365)
SELECT
  D1.D D1,
  D2.D D2,
  D2.D-D1.D DATE_DELTA, 
  to_char(EXTRACT(YEAR FROM D2.D) - EXTRACT(YEAR FROM  d1.d)) || ' years, ' ||
  to_char( EXTRACT(MONTH FROM D2.D) -  EXTRACT(MONTH FROM D1.D)) || ' months. ' as t 
  from
  DATES D1,
  DATES D2
WHERE
  D1.D<D2.D;

but if I extract month and year in this way must also be aware that the difference between two dates is only indicative, in this example I have a difference of 71 days indicated by two months

D1             D2             DATE_DELTA  T                         
-------------- -------------- ----------- ------------------
01-JAN-2010    13-MAR-2010            71  0 years, 2 months.

Regards

It helps me to think of it in these terms:

For:
select extract(month from ( dt1 – dt2 ) year to month) from dual;

Oracle calculate dt1 – dt2.
Oracle converts this number of days to a “interval year to month” value.
A “interval year to month” value cannot have a fractional month, so the month value is rounded.

So we do not get the same precision as months_between.

See http://forums.oracle.com/forums/thread.jspa?threadID=354651 for more discussion.

((TO_DATE(’03-MAR-2010′,’DD-MON-YYYY’)-TO_DATE(’18-FEB-2010′,’DD-MON-YYYY’))) year to month
is a valid SQL Expression

see
http://download.oracle.com/docs/cd/E11882_01/server.112/e10592/expressions009.htm#sthref1885

the difference in the brackets is **NOT** “evaluated at first” to a number

SELECT EXTRACT(month FROM order_date) "Month",
  COUNT(order_date) "No. of Orders"
  FROM orders
  GROUP BY EXTRACT(month FROM order_date)
  ORDER BY "No. of Orders" DESC;

Let’s put the SQL statement to the test:

CREATE TABLE T7 (ORDER_DATE DATE);
 
INSERT INTO
  T7
SELECT
  TO_DATE('01-JAN-2010','DD-MON-YYYY')+(ROWNUM-1)
FROM
  DUAL
CONNECT BY
  LEVEL<=300;
 
COMMIT;
 
SELECT EXTRACT(month FROM order_date) "Month",
  COUNT(order_date) "No. of Orders"
  FROM T7
  GROUP BY EXTRACT(month FROM order_date)
  ORDER BY "No. of Orders" DESC;
 
Month No. of Orders
----- -------------
    1            31
    5            31
    7            31
    3            31
    8            31
    9            30
    4            30
    6            30
    2            28
   10            27

That worked exactly as expected, but are we really seeing the 31 days that fall into January, or is Oracle rounding and stealing part of the days from February (the previous tests show that this is not the way the rounding was working – the days would actually be stolen from January and given to Feburary). Now to verify:

SELECT EXTRACT(month FROM order_date) "Month",
  COUNT(order_date) "No. of Orders"
  FROM T7
WHERE
  ORDER_DATE BETWEEN  TO_DATE('01-JAN-2010','DD-MON-YYYY') AND  TO_DATE('31-JAN-2010','DD-MON-YYYY')
  GROUP BY EXTRACT(month FROM order_date)
  ORDER BY "No. of Orders" DESC;
 
Month No. of Orders
----- -------------
    1            31

That worked exactly as I thought that it should.

Now back to the original question, what is wrong with that SQL statement?

Back to the example of the numeric value 13. Something odd is going on with this EXTRACT function. In this case, the result of subtracting a date from a date is not a number:

SELECT
  extract(year from (13) year to month) || ' years, ' ||
  extract(month from (13) year to month) || ' months. ' T
FROM
  DUAL;
  
ERROR at line 2:
ORA-30083: syntax error was found in interval value expression
  
SELECT
  (TO_DATE('03-MAR-2010','DD-MON-YYYY')-TO_DATE('18-FEB-2010','DD-MON-YYYY')) DAYS,
  extract(year from ((TO_DATE('03-MAR-2010','DD-MON-YYYY')-TO_DATE('18-FEB-2010','DD-MON-YYYY'))) year to month) || ' years, ' ||
  extract(month from ((TO_DATE('03-MAR-2010','DD-MON-YYYY')-TO_DATE('18-FEB-2010','DD-MON-YYYY'))) year to month) || ' months. ' T
FROM
  DUAL;
 
      DAYS T
---------- --------------------
        13 0 years, 1 months.
 
SELECT
  (TO_DATE('02-MAR-2008','DD-MON-YYYY')-TO_DATE('18-FEB-2008','DD-MON-YYYY')) DAYS,
  extract(year from ((TO_DATE('02-MAR-2008','DD-MON-YYYY')-TO_DATE('18-FEB-2008','DD-MON-YYYY'))) year to month) || ' years, ' ||
  extract(month from ((TO_DATE('02-MAR-2008','DD-MON-YYYY')-TO_DATE('18-FEB-2008','DD-MON-YYYY'))) year to month) || ' months. ' T
FROM
  DUAL;
 
      DAYS T
---------- --------------------
        13 0 years, 0 months.

The number 13 causes the SQL statement to fail. In the second of the above SQL statements, if Oracle assumed that 13 was in January of any year, months should evaluate to 0, but Oracle returned a value of 1 month. In the third of the above SQL statements, if Oracle assumed that 13 was in February of this year, like the second example, then Oracle should have returned a value of 0 months.

Another interesting result was provided in the code section of my previous response. It appears that Oracle is not strictly rounding the results of “extract(month from (D2.D-D1.D) year to month)” using the same calculation as used by the MONTHS_BETWEEN function. If that was the case, the date range of 01-JAN-2010 through 16-JAN-2010 would be rounded down to 0 months, but 16-JAN-2010 is indicated as the start of 1 month:

D1        D2        DATE_DELTA T                      MONTHS   M   D
--------- --------- ---------- -------------------- -------- --- ---
01-JAN-10 15-JAN-10         14 0 years, 0 months.     0.4516   0  14
01-JAN-10 16-JAN-10         15 0 years, 1 months.     0.4839   0  15
01-JAN-10 17-JAN-10         16 0 years, 1 months.     0.5161   0  16

I probably need to spend some more time working with the EXTRACT function.