Comparing two rows field by field through a loop

Question

Hi all,
I've got a tricky issue:
There are two tables (almost) identically designed;
In the first one there are lets say some new data and in the 2nd one there is data to be hold & adjusted;
I built up a third table, which contains the meta informations about the two other tables (i.e. it describes the tables, so for each field of these tables, there is the field name, field type, primary key marker, etc.);
The challenge now is, how can I compare two rows (of the same primary key) field by field and get out the differences (and adjusting the target table row) using PL/SQL procedures/functions.
I thought a create a cursor over the 3rd table and call functions to get the fields out (given by the field name of the cursor), compare the two fields and return if it's equal or different...
As I've seen, there is now direct way to replace a field name in a statement - but how to resolve this with a cool workaround?
In addition, there are about 120 pairs of such tables, to be compared in the same way - so I decided to prevent creating hard coded design for the software.
Even the selection of a pair of rows from the two tables should be parameterized, 'cause the primary keys are of course different in the 120 table pairs.
Any hints or suggestions?
Thanks in advance & regards,
Peter

Answer 1

Hi Peter,
I am very far from being an expert in Oracle but, everything I've read so far strongly suggests that PL/SQL is often the fastest among the customized solutions (beating even compiled C code). The reason - from what I've understood - is that PL/SQL is part of the Oracle kernel/engine, other languages simply aren't.
I suggest you do your own research to confirm what I've just said above.
On a completely different note, when I first read your post, I started writing a reply and decided to discard it because, I figured you probably would find it too far off your original thoughts. However, then you wrote:
>
There are some reasons to prevent hard coding, but it may be possible to generate the PL/SQL-code for each pair of tables by an external program.
>
This is exactly what I was mentioning in the post I ended up discarding. The basic step by step process I was going to mention is as follows:
STEP 1: If possible, write one PL/SQL program that does the job properly for one set of tables. Test it thoroughly and identify potential differences in the process if it had to be applied against other tables. Make it as generic as reasonably possible, that is, keeping the code as simple as possible.
STEP 2: turn the working code of step 1 into a template (usually replacing table names, columns, etc by some token that would not be a valid name).
STEP 3: generate a list of the tables and their corresponding columns you need to process. This list may need to be massaged a little to become the generating seed of 120 programs (or so based on what you've said).
STEP 4: if you don't know Perl or AWK then (don't be alarmed...) download AWK (not Perl) and learn it... you can learn the thing in a couple days (maybe even less). It is a very simple and straightforward text processing language, I know non progammers that learned enough in one day to do useful things with it. Using AWK you can easily generate the 120 programs you need from a working template in very little time (a few hours after you've become proficient). By the way, there is AWK, GAWK, NAWK, all pretty much the same thing and all free. The standard manual for AWK is the one written by the authors Aho, Weinberger and Kernigan, about 100 small pages, it's an easy read and no programmer should be without it :) (better than American Express!)
What you've described sounds just like the thing to use AWK for. I once used AWK to generate 27,000 (that's 27 thousand) lines of COBOL code, debugged and fully tested in 2 days! (testing was automated using AWK too!) It would have taken much longer to just type all that code.
Anyway, I hope that gives you something to think about and be of some help.
John.