Order of Update

[Terry Lee Tucker <terry@xxxxxxxx>]

Hello List:

I'm having an "interesting" problem. I know what is happening, but I do not 
know why; therefore, I am hoping that some of those very familiar with the 
internals can/will explain it to me.

I have a function A written in plpgsql which is designed to undo changes to 
several records. This function is called from an X Windows interface when the 
user presses the "undo" button. The functions updates one or records in a 
table within loop and then updates a "parent" record from a different table. 
The updates in the loop depend on certain values being present in the 
"parent" record. At the end of the loop, an update to the "parent" record 
sets to NULL the values that were used in the loop updates. The problem is 
that the update to the parent occurs first even though the code performing 
the update is physically located below the loop. Here is the function:
CREATE OR REPLACE FUNCTION removeCash (INTEGER) RETURNS VOID AS '
DECLARE
    cash_recid      ALIAS FOR $1;           -- recid of receipt record
    cashRec         RECORD;                 -- cash record buffer
    itemRec         RECORD;                 -- item record buffer
    itemTypes       TEXT DEFAULT ''(w|j|d|o|c|b)''; -- possible item types
    funcName        TEXT DEFAULT ''removeCash'';    -- function name
    dbg             BOOLEAN DEFAULT True;   -- debug print flag
BEGIN
    IF dbg THEN
        RAISE NOTICE ''% ()'', funcName;
    END IF;

    SELECT INTO cashRec order_num, ref FROM cash WHERE recid = cash_recid;
    IF NOT FOUND THEN
        RAISE EXCEPTION
            ''%: Cannot find a cash record referenced by %.'', funcName,
            cash_recid;
    END IF;

    /* Loop through the item records linked to this cash record. */
    FOR itemRec IN SELECT recid, item_type FROM item
        WHERE order_num = cashRec.order_num
        AND ref = cashRec.ref
        AND item_type ~* itemTypes ORDER BY ref
    LOOP
        /* For type C, simply set the apply amount to zero and link to the
         * value of olink. */
        IF itemRec.item_type ~* ''c'' THEN
            UPDATE item SET apply_amt = 0 WHERE recid = itemRec.recid;
            IF NOT FOUND THEN
                RAISE EXCEPTION
                ''%: Cannot update apply_amt for item %.'', funcName,
                itemRec.recid;
            END IF;
        ELSE
            /* Delete all other item types. */
            DELETE item WHERE recid = itemRec.recid;
            IF NOT FOUND THEN
                RAISE EXCEPTION
                ''%: Cannot delete item record %.'', funcName,
                itemRec.recid;
            END IF;
        END IF;
    END LOOP;

    /* Now, update the cash record setting batch and chkno to null. */
    UPDATE cash SET chkno = NULL, batch = NULL
        WHERE recid = cash_recid;
    IF NOT FOUND THEN
        RAISE EXCEPTION
        ''%: Unable to update cash record %.'', funcName, cash_recid;
    END IF;

    RETURN;
END;
' LANGUAGE plpgsql EXTERNAL SECURITY DEFINER;

There are trigger updates involved in this process. Setting apply_amt to zero 
on the line item record causes an update to be made to the cash record linked 
to the item record subtracting the value of apply_amt from a total applied 
column in the cash record. Is this why the cash record is updated first? 
Also, cash.batch and cash.chkno are the two components of a UNIQUE index. 
Does this have something to do with it?

Any insight anyone can give would be greatly appreciated.
-- 
Quote: 81
"Government does not solve problems; it subsidizes them."

 --Ronald Reagan