Is it possible to conditionally pivot columns of different data types in Oracle? - sql

Question

Using Oracle 12c. I have a query that returns all the data I need, however it is split across multiple rows. I'm trying to use pivot to consolidate the data from n rows into 1 row, transposing n unique value sets from 2 columns into n columns, conditionally choosing which column from a set of columns to use the value from for each of the n transposed columns.
I'm struggling to find out if this is possible, or if it can't be done.
Following is a simplified example of what the query returns:
| ID | Inst_Created | Inst_Modified | Inst_Group | Inst_Prop | VAL_INT | VAL_REAL | VAL_STR |
+----+--------------+---------------+------------+-----------+---------+----------+---------+
| 01 | 1571954537 | 1571954537 | GenGroup1 | IntProp1 | 0 | (null) | (null) |
| 01 | 1571954537 | 1571954537 | GenGroup1 | RealProp2 | (null) | 12.34567 | (null) |
| 01 | 1571954537 | 1571954537 | GenGroup1 | RealProp3 | (null) | 123.4567 | (null) |
| 01 | 1571954537 | 1571954537 | GenGroup1 | StrProp4 | (null) | (null) | dirpath |
| 01 | 1571954537 | 1577754537 | GenGroup2 | IntProp5 | 1337 | (null) | (null) |
| 01 | 1571954537 | 1577754537 | GenGroup2 | RealProp6 | (null) | 13.37 | (null) |
| 01 | 1570054537 | 1570854537 | GenGroup3 | StrProp7 | (null) | (null) | testing |
| 01 | 1570054537 | 1570854537 | GenGroup3 | StrProp8 | (null) | (null) | valid |
| 02 | 1571954540 | 1571954540 | GenGroup1 | IntProp1 | 1 | (null) | (null) |
| 02 | 1571954540 | 1571954540 | GenGroup1 | RealProp2 | (null) | 12.34568 | (null) |
| 02 | 1571954540 | 1571954540 | GenGroup1 | RealProp3 | (null) | 123.4568 | (null) |
| 02 | 1571954540 | 1571954540 | GenGroup1 | StrProp4 | (null) | (null) | dirpat2 |
| 02 | 1571954540 | 1577754540 | GenGroup2 | IntProp5 | 1338 | (null) | (null) |
| 02 | 1571954540 | 1577754540 | GenGroup2 | RealProp6 | (null) | 13.38 | (null) |
| 02 | 1570054540 | 1570854540 | GenGroup3 | StrProp7 | (null) | (null) | testin2 |
| 02 | 1570054540 | 1570854540 | GenGroup3 | StrProp8 | (null) | (null) | valid2 |
+----+--------------+---------------+------------+-----------+---------+----------+---------+
The Inst_Created and Inst_Modified columns are mock epoch timestamps. The VAL_INT, VAL_REAL, and VAL_STR columns are NUMBER(38,0), FLOAT, and VARCHAR2(1024) data types respectively. Only one of the three VAL_* columns will be non-null for any given row; You can know which one of the three it'll be by the Inst_Group, Inst_Prop columns.
Following is a simplified example of what I'm trying to produce by using pivot:
| ID | Earliest_Created | Latest_Modified | G1P1_Int | G1P2_Real | G1P3_Real | G1P4_Str | G2P5_Int | G2P6_Real | G3P7_Str | G3P8_Str |
+----+------------------+-----------------+----------+-----------+-----------+----------+----------+-----------+----------+----------+
| 01 | 1570054537 | 1577754537 | 0 | 12.34567 | 123.4567 | dirpath | 1337 | 13.37 | testing | valid |
| 02 | 1570054540 | 1579954540 | 1 | 12.34568 | 123.4568 | dirpat2 | 1338 | 13.38 | testin2 | valid2 |
+----+------------------+-----------------+----------+-----------+-----------+----------+----------+-----------+----------+----------+
I've figured out how to use the pivot_for_clause and pivot_in_clause clauses to turn the (Inst_Group, Inst_Prop) values into columns while renaming the column. What I'm struggling to figure out is how to get the values into the transposed columns, with them being the right data type.
I had previously attempted to use the LISTAGG function in the query, casting all the VAL_* values to varchar2, so that there was only one column for VAL values. I gave up on that method due to the problems in trying to get each transposed column to be the correct original data type, and due to the performance hit all of that seemed to be causing.

Answer 1

You can achieve it using conditional aggreagation:
Select ID, EARLIEST_CREATED, LATEST_MODIFIED,
MAX(CASE WHEN INST_GROUP = 'GenGroup1' and INST_PROP1 = 'IntProp1' THEN VAL_INT END) AS G1P1_INT,
MAX(CASE WHEN INST_GROUP = 'GenGroup1' and INST_PROP1 = 'RealProp2' THEN VAL_REAL END) AS G1P2_REAL,
...
..
.
From your_table
Group by ID, EARLIEST_CREATED, LATEST_MODIFIED;
Cheers!!