Question about using objects in SQL query.

Question

I had posted this question in the SQL/PLSQL forum but I guess nobody took the time to understand exactly what I am asking so I decided to try here hoping to get the answer. So here is the thing:
I have created generic object type "tree" - the constructor takes as a parameter sql query which returns "node_id" and "parent_node_id" - this is all we need to have a tree. The object has all related to a tree structure member functions and one of them is "oldest_relative" (the tree may not be fully connected - it may be more like a set of many trees, so it's not necessary all nodes to have the same root).
I also have departments table with the following fields: department_id, parent_department_id, department_name,...
all records in the table w/out parent_departments (parent_department_id is null) are considered divisions.
Now if I run the following query:
SELECT "DEPARTMENT_ID", "PARENT_DEPARTMENT_ID", "DEPARTMENT_NAME", tree('select department_id "node_id", parent_department_id "parent_node_id" from departments').oldest_relative("DEPARTMENT_ID") "DIVISION_ID" FROM departments
my question is: Is the tree object created for every row or does Oracle somehow caches the object since the object itself is not changing but only the parameter for the oldest_relative member function.
The table only has a few hunderd records and I can't see much of a difference in the execution time btw the query above and query like this:
SELECT "DEPARTMENT_ID", "PARENT_DEPARTMENT_ID", "DEPARTMENT_NAME", b.t.oldest_relative("DEPARTMENT_ID") "DIVISION_ID"
FROM departments left join (select tree('select department_id "node_id", parent_department_id "parent_node_id" from departments') t from dual) b on 1 = 1
where the object is clearly created just ones. (there is probably a better way to do it instead of this join)
Pls elaborate
George

Answer 1

Not exactly sure what the question is...
As I understand, you are comparing the following two constructor calls:
+select..  tree('select department_id "node_id", parent_department_id "parent_node_id" from departments').oldest_relative("DEPARTMENT_ID") ... FROM ...+
+select tree('select department_id "node_id", parent_department_id "parent_node_id" from departments') ... FROM dual+
These calls are the same (besides the 1st one doing an immediate implicit call to a method of the object being constructed). The number of times these are being called depends on the number of times this SQL projection is applied - and that is determined by the number of rows being projected by the SELECT.
The latter one is against DUAL which only has a single row. So that constructor is only called once. The former can be against multiple rows. Obviously a single pass through a data set is desirable - which means that the sub-select (use by the constructor) should ideally only be executed once and makes the 2nd method more desirable.
However, I'm having a hard time understanding why the class and constructor are at all needed. Why pull data from a SQL table into PL memory? As that is where the class will need to cache and store the results of that construction parameter SQL SELECT. And once in PL memory, how does the object effectively access, search and use this cached data?
PL memory is expensive. It is not sharable.
PL data structures are primitive - these cannot be compared to SQL structures in the form of tables and columns that can be stored in a number of physical ways (index tables, hash tables, partitioned tables, clustered tables, etc). Cannot be indexed like SQL structures using B+tree, bitmap, function and other indexing methods. Cannot be sorted, grouped, analysed, filtered, etc like SQL structured data.
It makes very little sense to read SQL data into a class and then deal with that data, cached in expensive PL memory, using primitive PL structures.
And the same would be true if Java or C# was used. The best place for data is inside the SQL engine. That is the most superior environment for data. It can processes more data, scale better, perform better and offer more flexibility, than pulling data from it and then crunch that data using PL or Java or C#.