well, think about this for a minute.


A client fetches from a cursor - they fetch say 100 rows at a time. Suppose further that your table function returns 200 rows.

Now, the client requests the first 100 rows - we run your code - you return the first 100 and when you call pipe row() for the 100th time - we SUSPEND your code and sent the 100 records back to the client.

Now, the call is over - the database call (fetch 100 records) is done. Pragma serially reusable says "when the call is complete, deinstantiate the code - reset it back to square one, reset all variables, get all its memory bits out of the PGA - make it go away"

But - you have 100 more records to return - you will have "forgotten" that if we did that.


It doesn't even begin to make sense to use a serially reusable, get me out of memory after each call, bit of code with a table function.

The reason it is not supported is - it cannot make sense to do it. You *need* to maintain a state between calls to the database for a table function - since the plsql routine does NOT finish in general in a single call - it takes repeated fetches (repeated calls) to get this table function "exhausted"



My alternative would be to not use a serially reusable package in that context - since they do not, cannot, work correctly.