Oracle Database 12.1.0.2.0 running on Oracle Linux 6.4:
While studying MERGE with invisible columns, I discovered that invisible columns in the target table cannot be read. The workaround appears to be
MERGE INTO (SELECT <column list> FROM <target table>) AS <alias>
However, the documentation does not seem to allow this. Here are the details.
Test data
> CREATE TABLE t_target(
k1 NUMBER PRIMARY KEY,
c1 NUMBER,
i1 NUMBER invisible
)
table T_TARGET created.
> INSERT INTO t_target (k1,c1,i1)
SELECT 2, 2, 2 FROM dual
UNION ALL
SELECT 3, 3, 3 FROM dual
UNION ALL
SELECT 4, 4, 4 FROM dual
3 rows inserted.
> CREATE TABLE t_source(
k1 NUMBER PRIMARY KEY,
c1 NUMBER,
i1 NUMBER invisible
)
table T_SOURCE created.
> INSERT INTO t_source (k1,c1,i1)
SELECT 1, 1, 1 FROM dual
UNION ALL
SELECT 2, 2, 9999 FROM dual
UNION ALL
SELECT 3, 3, 3 FROM dual
3 rows inserted.
First try
Please note that I have a WHERE clause within the WHEN MATCHED clause. Its purpose is to avoid updating a row when the data is already correct. The WHERE clause tries to read the invisible column from the target table.
> MERGE INTO t_target o
USING (
SELECT k1, c1, i1 FROM t_source
) n
ON (o.k1 = n.k1)
WHEN MATCHED THEN UPDATE SET
c1=n.c1, i1=n.i1
WHERE 1 IN (
decode(o.c1,n.c1,0,1),
decode(o.i1,n.i1,0,1)
)
WHEN NOT MATCHED THEN INSERT
(k1, c1, i1)
VALUES(n.k1, n.c1, n.i1)
...
Error at Command Line : 10 Column : 12
Error report -
SQL Error: ORA-00904: "O"."I1": invalid identifier
As you can see, I put a subquery after the USING clause so that "n.i1" would be "visible", but that's not enough since the "I1" column in the target table is still invisible.
Second try
> MERGE INTO (
SELECT k1, c1, i1 FROM t_target
) o
USING (
SELECT k1, c1, i1 FROM t_source
) n
ON (o.k1 = n.k1)
WHEN MATCHED THEN UPDATE SET
c1=n.c1, i1=n.i1
WHERE 1 IN (
decode(o.c1,n.c1,0,1),
decode(o.i1,n.i1,0,1)
)
WHEN NOT MATCHED THEN INSERT
(k1, c1, i1)
VALUES(n.k1, n.c1, n.i1)
2 rows merged.
Here I used a subquery in the INTO clause as well, and it worked.
Unfortunately, this doesn't seem to be allowed in the documentation: INTO refers to a table or view as schema objects.
My question is:
How can I refer to invisible columns in the target table without creating a new object? My workaround using a subquery seems to work fine, but can I recommend it if it is not documented?
Can I substitute an "inline view" for a view and still be supported?