XmlAgg first n ordered elements only problem

Question

I cannot find a way on Oracle to limit the number of rows aggregated by XmlAgg to only the first n in a specified order. I have produced a simple example demonstrating my problem which should be easy to replicate, as follows:
I have two tables, INCIDENT and INCIDENT_LOG_ENTRY (there may be many entries for a given incident).
I wish to extract details (as XML) of an incident and its most recent two log entries only.
-- Create INCIDENT table and two incidents:
create table INCIDENT (ID NUMBER(10,0) PRIMARY KEY, INCIDENT_SUMMARY VARCHAR2(200));
insert into INCIDENT values (1, 'Hold up');
insert into INCIDENT values (2, 'Car Crash');
-- Create INCIDENT_LOG_ENTRY table and log entries for both incidents:
CREATE table INCIDENT_LOG_ENTRY (ID NUMBER(10,0) PRIMARY KEY, INCIDENT_ID NUMBER(10,0), ENTRY_DATE_TIME DATE, ENTRY_TEXT VARCHAR2(500));
insert into INCIDENT_LOG_ENTRY values (1, 1, TO_DATE('2009-01-01 08:15:11', 'YYYY-MM-DD HH24:MI:SS'), 'Hold up on Main Street');
insert into INCIDENT_LOG_ENTRY values (2, 1, TO_DATE('2009-01-01 08:17:40', 'YYYY-MM-DD HH24:MI:SS'), 'Suspect pursued in high speed chase');
insert into INCIDENT_LOG_ENTRY values (3, 1, TO_DATE('2009-01-01 08:20:29', 'YYYY-MM-DD HH24:MI:SS'), 'Suspect lost in traffic');
insert into INCIDENT_LOG_ENTRY values (4, 1, TO_DATE('2009-01-03 11:55:31', 'YYYY-MM-DD HH24:MI:SS'), 'Suspect apprehended in hospital');
insert into INCIDENT_LOG_ENTRY values (21, 2, TO_DATE('2009-01-01 08:29:15', 'YYYY-MM-DD HH24:MI:SS'), 'Collision between car jumping red light and lorry');
insert into INCIDENT_LOG_ENTRY values (22, 2, TO_DATE('2009-01-01 08:45:53', 'YYYY-MM-DD HH24:MI:SS'), 'Driver taken to hospital');
Here is the query (note order by is within xmlAgg as per Oracle documentation):
SELECT xmlAgg(xmlElement("INCIDENT", xmlForest(i.ID, i.INCIDENT_SUMMARY),
xmlElement("INCIDENT_LOG_ENTRIES",
(SELECT xmlAgg(xmlElement("INCIDENT_LOG_ENTRY", xmlForest(ile.ID, ile.ENTRY_DATE_TIME, ile.ENTRY_TEXT)) order by ile.ENTRY_DATE_TIME desc)
FROM INCIDENT_LOG_ENTRY ile
WHERE ile.INCIDENT_ID = i.ID
AND rownum <= 2
FROM INCIDENT i where i.ID = 1
And here is its output:
<INCIDENT>
<ID>1</ID>
<INCIDENT_SUMMARY>Hold up</INCIDENT_SUMMARY>
<INCIDENT_LOG_ENTRIES>
<INCIDENT_LOG_ENTRY>
<ID>2</ID>
<ENTRY_DATE_TIME>01-JAN-09</ENTRY_DATE_TIME>
<ENTRY_TEXT>Suspect pursued in high speed chase</ENTRY_TEXT>
</INCIDENT_LOG_ENTRY>
<INCIDENT_LOG_ENTRY>
<ID>1</ID>
<ENTRY_DATE_TIME>01-JAN-09</ENTRY_DATE_TIME>
<ENTRY_TEXT>Hold up on Main Street</ENTRY_TEXT>
</INCIDENT_LOG_ENTRY>
</INCIDENT_LOG_ENTRIES>
</INCIDENT>
This is not the desired result - I want the two most recent log entries (4 and 3). Clearly the rownum has taken effect before the ordering is applied by XmlAgg. However, if I try to force the ordering first by using a nested subquery, Oracle complains that the incident (table alias 'i') is not visible within the subquery:
SELECT xmlAgg(xmlElement("INCIDENT", xmlForest(i.ID, i.INCIDENT_SUMMARY),
xmlElement("INCIDENT_LOG_ENTRIES",
(SELECT xmlAgg(xmlElement("INCIDENT_LOG_ENTRY", xmlForest(ile.ID, ile.ENTRY_DATE_TIME, ile.ENTRY_TEXT)) order by ile.ENTRY_DATE_TIME desc)
FROM (select * from (select * from INCIDENT_LOG_ENTRY WHERE INCIDENT_ID = i.ID order by ENTRY_DATE_TIME) where rownum <= 2) ile
FROM INCIDENT i where i.ID = 1
Which results in:
SQL Error: ORA-00904: "I"."ID": invalid identifier
If anyone knows how to resolve this problem I would be extremely grateful.
(BTW, it works without any trouble on SQL Server:
select i.ID, i.INCIDENT_SUMMARY,
(select top 2 ile.ID, ile.ENTRY_TEXT, ile.ENTRY_DATE_TIME
from INCIDENT_LOG_ENTRY ile
where ile.INCIDENT_ID = i.ID
order by ile.ENTRY_DATE_TIME desc for xml path('INCIDENT_LOG_ENTRY'), type) as "INCIDENT_LOG_ENTRIES"
from INCIDENT i
where i.ID = 1
for xml path('INCIDENT'), type
Which yields the desired result:
<INCIDENT>
<ID>1</ID>
<INCIDENT_SUMMARY>Hold up</INCIDENT_SUMMARY>
<INCIDENT_LOG_ENTRIES>
<INCIDENT_LOG_ENTRY>
<ID>4</ID>
<ENTRY_TEXT>Suspect apprehended in hospital</ENTRY_TEXT>
<ENTRY_DATE_TIME>2009-01-03T11:55:31</ENTRY_DATE_TIME>
</INCIDENT_LOG_ENTRY>
<INCIDENT_LOG_ENTRY>
<ID>3</ID>
<ENTRY_TEXT>Suspect lost in traffic</ENTRY_TEXT>
<ENTRY_DATE_TIME>2009-01-01T08:20:29</ENTRY_DATE_TIME>
</INCIDENT_LOG_ENTRY>
</INCIDENT_LOG_ENTRIES>
</INCIDENT>
)

Answer 1

Hi,
Sorry, I don't know how to get the first 2 rows in the XML functions, which, no doubt, is the best way to do it.
At the table level, you can assign numbers to every log entry in a sub-query, and use that number (instead of ROWNUM) to get the top 2:
WITH   numbered_log_entry
AS
     SELECT     id, incident_id, entry_date_time, entry_text
     ,     RANK () OVER ( PARTITION BY  incident_id
                           ORDER BY          entry_date_time     DESC
                    ) AS rnk
     FROM     incident_log_entry
SELECT  xmlAgg ( xmlElement ( "INCIDENT"
                           , xmlForest (i.ID, i.INCIDENT_SUMMARY)
                   , xmlElement ( "INCIDENT_LOG_ENTRIES"
                                , ( SELECT  xmlAgg ( xmlElement ( "INCIDENT_LOG_ENTRY"
                                                              , xmlForest ( ile.ID
                                                          , ile.ENTRY_DATE_TIME
                                                       , ile.ENTRY_TEXT
                                              ) order by ile.ENTRY_DATE_TIME desc
                              FROM    numbered_LOG_ENTRY ile     -- Changed
                              WHERE   ile.INCIDENT_ID      = i.ID
                              AND     rnk          <= 2     -- Changed
FROM    INCIDENT   i
where      i.ID        = 1;Notice that the main query is exatcly what you posted, except for the 2 line marked "Changed".
Thanks for posting the CREATE TABLE and INSERT statments! That made it really easy to test.
You may have noticed that this site compresses whitespace.
When you have to post formated code or output on this site, type these 6 characters:
&#123;code&#125;
(small letters only, inside curly brackets) before and after sections of formatted text, to preserve spacing.