How to call an indirect table in FROM - sql

Question

Please tell me what I am doing wrong. I have tried the following:
SELECT *
FROM 'A'||(TO_CHAR(SYSDATE, 'YYMMDD'))||'B'
SELECT *
FROM CAST('A'||(TO_CHAR(SYSDATE, 'YYMMDD'))||'B' AS varchar2(8))
SELECT *
FROM table( CAST('A'||(TO_CHAR(SYSDATE, 'YYMMDD'))||'B' AS varchar2(8)))
DEFINE tName = CAST('A'||(TO_CHAR(SYSDATE, 'YYMMDD'))||'B' AS varchar2(8))
SELECT *
FROM &tName
SET tName AS CAST('A'||(TO_CHAR(SYSDATE, 'YYMMDD'))||'B' AS varchar2(8))
SELECT *
FROM &tName
SET #tName AS CAST('A'||(TO_CHAR(SYSDATE, 'YYMMDD'))||'B' AS varchar2(8))
SELECT *
FROM #tName
SELECT *
FROM
(
SELECT CAST('A'||(TO_CHAR(SYSDATE, 'YYMMDD'))||'B' AS varchar2(8))
FROM DUAL
)
(Note: I have also tried adding the owner before the table in all of the examples above, as shown in the second example below)
I have verified that
SELECT CAST('A'||(TO_CHAR(SYSDATE, 'YYMMDD'))||'B' AS varchar2(8))
FROM DUAL
displays A170429B
SELECT CAST('owner.A'||(TO_CHAR(SYSDATE, 'YYMMDD'))||'B' AS varchar2(8))
FROM DUAL
displays owner.A170429B
but I can't get the FROM statement to see it as a Table
The database creates a new table every day; which, is why I am trying to do this.

Answer 1

As others have observed, partitioning is the elegant way to arrange this. But Partitioning remains a licensed extra on the Enterprise Edition,and that's expensive. So here's a couple of cheaper options.
The first option is to do what you are trying to do now, and query each table by name. For this approach to work you need to use dynamic SQL.
Here are some tables - t170428, t170429, t170430 - which all look like this
create table t170428 (
id number not null primary key
, type varchar2(10)
, col1 varchar2(10)
, col2 number
, col3 date not null
)
/
To query them we need a SQL type with a signature that matches the tables' projection:
create or replace type tyymmdd_t as object (
id number
, type varchar2(10)
, col1 varchar2(10)
, col2 number
, col3 date
);
/
create or replace type tyymmdd_nt as table of tyymmdd_t
/
Here is a function which dynamic builds a table name from a passed date and returns a nested table of rows from that table:
create or replace function get_date_table
( p_target_date in date)
return tyymmdd_nt
is
return_value tyymmdd_nt;
begin
execute immediate
' select tyymmdd_t(id, type, col1, col2, col3) from t'
||to_char(p_target_date, 'yymmdd')
bulk collect into return_value;
return return_value;
end;
/
To query a table we use the table() function like this:
SQL> select * from table(get_date_table(sysdate));
ID TYPE COL1 COL2 COL3
---------- ---------- ---------- ---------- ---------
9 D2 SUN 1 30-APR-17
10 D2 SUN 2 30-APR-17
SQL> select * from table(get_date_table(date'2017-04-28'));
ID TYPE COL1 COL2 COL3
---------- ---------- ---------- ---------- ---------
1 D1 FRI 1 28-APR-17
2 D1 FRI 2 28-APR-17
3 D1 FRI 3 28-APR-17
4 D1 FRI 4 28-APR-17
5 D1 FRI 5 28-APR-17
SQL> select * from table(get_date_table(sysdate+1));
select * from table(get_date_table(sysdate+1))
*
ERROR at line 1:
ORA-00942: table or view does not exist
ORA-06512: at "FOX.GET_DATE_TABLE", line 7
SQL>
The second option is Partition Views. This is an old technique (from the previous millennium!) which allows us to build a view over multiple tables using the UNION ALL operator and get many benefits of Partitioning such as partition pruning - provided you're using a really old version of Oracle. Partition Views were deprecated in 8.0 and Oracle stopped supporting them in 9iR2. The documentation dates back to Oracle7 Find out more.
Anyway, the principle of partitioned views is this:
Enforce check constraints on the "partition key"
Build indexes on the partition key columns
Gather stats
Build the view
Like the table projection, the constraints and indexes must be the same for all tables.
alter table t170428 add constraint t170428_ptn_key_ck check (col3 = date '2017-04-28');
alter table t170429 add constraint t170429_ptn_key_ck check (col3 = date '2017-04-29');
alter table t170430 add constraint t170430_ptn_key_ck check (col3 = date '2017-04-30');
create unique index t170428_ptn_idx on t170428(col3, id) compress 1;
create unique index t170429_ptn_idx on t170429(col3, id) compress 1;
create unique index t170430_ptn_idx on t170430(col3, id) compress 1;
exec dbms_stats.gather_table_stats('FOX', 'T170428', cascade=>true)
exec dbms_stats.gather_table_stats('FOX', 'T170429', cascade=>true)
exec dbms_stats.gather_table_stats('FOX', 'T170430', cascade=>true)
create or replace view v_all_the_dates as
select * from t170428
union all
select * from t170429
union all
select * from t170430
/
Because Oracle don't support Partition Views in later versions of the database this approach won't give you partition pruning. But it could still be quite efficient providing you are rigorous about the indexing and check constraints.
A third option which might fit is an external table. The creation of dated tables suggests a daily load. If these just act as staging tables for data which arrive as files you could use an external table to access the data. The table would be a stable structure; all you would need to change is the location of the daily feed file. Find out more.

Answer 2

DBMS_XMLGEN can create a completely dynamic SQL statement without any additional privileges.
(But before you use the difficult SQL below, I recommend you try again to get more access to the database and use one of APC's solutions. Write access to the vendor's application schema is not required. The objects can be created in a separate user schema that will not interfere with the application.)
First, create a test table with today's date in the name
begin
execute immediate '
create table a'||to_char(sysdate, 'YYMMDD')||'b
(
column1 number,
column2 number
)';
end;
/
Add a test row. One potential problem with this solution is that when the inner query returns no rows it throw an error ORA-06502: PL/SQL: numeric or value error/nORA-06512: at "SYS.XMLTYPE", line 272/nORA-06512: at line 1.
begin
execute immediate 'insert into a'||to_char(sysdate, 'YYMMDD')||'b values (1,2)';
commit;
end;
/
This query reads from the table without knowing the table's name. It does however require knowing the table's columns. (If the vendor creates different columns in each table that's a more difficult problem that will definitely require custom objects.)
--Step 3: Convert XML back into a row.
select column1, column2
from
(
--Step 2: Create an XML result set of a dynamic query.
select
xmltype(dbms_xmlgen.getxml(
(
--Step 1: SELECT statement that generates a SELECT statement.
select
'select column1, column2
from a'||to_char(sysdate, 'YYMMDD')||'b'
from dual
)
)) xml_results
from dual
)
cross join
xmltable
(
'/ROWSET/ROW'
passing xml_results
columns
column1 number path 'COLUMN1',
column2 number path 'COLUMN2'
);
Results:
COLUMN1 COLUMN2
------- -------
1 2