Question

I have a table containing house numbers as varchar2 like 10, 10a etc. I want to detect all the rows that are not numeric like ’10a’.
My solution would be to output every row that causes exception when there is an attempt to convert data like ’10a’ to number.

declare
number_correct number;
number_incorrect varchar2(4000);
begin 
for rec in(select '10' house_nr from dual union select '10a' house_nr from dual)
loop
number_correct:=to_number(rec.house_nr);
number_incorrect:=rec.house_nr;
end loop;
exception
when others then
dbms_output.put_line('correct: '||number_correct);
dbms_output.put_line('incorrect: '||number_incorrect);
end;

It should show that incorrect is 10a, but it doesn’t.

Answer 1

While @Dave is right about the actual question you asked, I think what you’re trying to achieve is different than what you’re doing. As written the PL/SQL block will only evaluate values until it gets to the first non-number. If you want all values evaluated, then you’ll need something like this:

BEGIN
   FOR rec IN (SELECT '10' house_nr FROM DUAL
               UNION ALL
               SELECT '10a' house_nr FROM DUAL
               UNION ALL
               SELECT '11' house_nr FROM DUAL) LOOP
      error_fl         := FALSE;
      BEGIN
         number_correct   := TO_NUMBER(rec.house_nr);
      EXCEPTION
         WHEN VALUE_ERROR THEN
            error_fl   := TRUE;
         WHEN OTHERS THEN
            RAISE;
      END;

      IF error_fl THEN
         DBMS_OUTPUT.put_line('incorrect: ' || rec.house_nr);
      ELSE
         DBMS_OUTPUT.put_line('correct: ' || rec.house_nr);
      END IF;
   END LOOP;
END;

By moving the exception handling to inside the loop, we can continue processing after an error is raised. Also, this version only returns the “incorrect” message if the error that occurs is in fact a conversion error, as opposed to an unexpected error. When you’re writing error handling for specific conditions like this, it’s important to be as precise as possible.

---

As @Stephen points out, regex is generally faster than PL/SQL procedures or functions and a single SQL statement is typically better than a procedural loop. However, regex functions can only be used in where clauses, so if you want to see results for all values, you’d need to query your table twice:

WITH test_num as (SELECT '10' house_nr FROM DUAL
                  UNION ALL
                  SELECT '10a' house_nr FROM DUAL
                  UNION ALL
                  SELECT '11' house_nr FROM DUAL)
SELECT 'correct' AS status, a.*
FROM   test_num a
WHERE  REGEXP_LIKE(t1, '[^[:digit:]]')
UNION ALL
SELECT 'incorrect', a.*
FROM   test_num a
WHERE  NOT REGEXP_LIKE(t1, '[^[:digit:]]');

If you’re not comfortable with regex, or can’t come up with a suitable expression, you can still do this in a single SQL statement by creating your own function:

CREATE OR REPLACE FUNCTION is_num(p_string VARCHAR2)
   RETURN NUMBER
   DETERMINISTIC IS
   v_test     NUMBER;
BEGIN
   v_test   := p_string;
   RETURN 1;
EXCEPTION
   WHEN VALUE_ERROR THEN
      RETURN 0;
   WHEN OTHERS THEN
      RAISE;
END;

WITH test_num as (SELECT '10' house_nr FROM DUAL
                  UNION ALL
                  SELECT '10a' house_nr FROM DUAL
                  UNION ALL
                  SELECT '11' house_nr FROM DUAL)
SELECT CASE is_num(t1) WHEN 0 THEN 'correct' ELSE 'incorrect' END AS status, a.*
FROM   test_num a;

This won’t be quite as fast as using regex in most circumstances, but it does have one advantage: function-based indexes are notoriously finicky about regexp functions, but they shouldn’t have any problem with a function like this. It doesn’t seem like you’ll need an index for this particular query, but it’s something to keep in mind.