House numbers

Question

Hi,
I am working on a query that needs to group addresses and should return the first and last house number in that group. However the house numbers are stored as varchar2, but I want a first and last with a best effort to sort them numerically.
Housenumber can have values like '1', '1A', '1_2', '10' etc.
My approach was to sort first on the numerical part, using to_number and regexp_substr and a second sort by the varchar2.
Something like:
with t as ( select 1 id, '10' hn from dual union all
            select 1,    '1'     from dual union all
            select 1,    '1A'    from dual union all
            select 1,    '11'    from dual union all
            select 1,    '11B'   from dual union all
            select 1,    '1_2'   from dual union all
            select 1,    '1 B'   from dual union all
            select 1,    '2A'    from dual union all
            select 1,    '1-2'   from dual )
select hn from t
order by to_number(regexp_substr(hn,'[0-9]*')),hnWhich gives the sorted output:
HN
1
1 B
1-2
1A
1_2
2A
10
11
11BPerfect! But I am stuck at the second step. I want the id, the first (ie '1') and last ('11B') grouped:
ID FIRST LAST
1  1     11BIt feels like I've done 99% and only need a tiny step... oh, and we have 3+ million addresses (without an index on the house number), so performance is a concern :)
I am on 10.2.0.4.0
Thanks!

Answer 1

Hi,
So you want only one row per id? That sounds like a job for GROUP BY:
SELECT       id
,       MIN (hn) KEEP ( DENSE_RANK      FIRST
                 ORDER BY TO_NUMBER (REGEXP_SUBSTR (hn, '[0-9]*'))
                  )     AS first
,       MAX (hn) KEEP ( DENSE_RANK      LAST
                 ORDER BY TO_NUMBER (REGEXP_SUBSTR (hn, '[0-9]*'))
                  )     AS last
FROM       t
GROUP BY  id
;Depending on what you know about the numbers, you might be able to do something more efficient.
For example, if you know that the number (or at least the number that you use for sorting) comes at the very beginning of hn, and that it is never more than 6 digits long, you could try:
SELECT       id
,       MIN ( REGEXP_REPLACE ( '     ' || hn
                      , '[0-9 ]*([0-9 ]{6})'
                      , '\1'
           )                 )     AS first
,       MAX ( REGEXP_REPLACE ( '     ' || hn
                      , '[0-9 ]*([0-9 ]{6})'
                      , '\1'
           )                 )     AS last
FROM       t
GROUP BY  id
;Edited by: Frank Kulash on Jun 17, 2011 6:59 AM