Switching to SQL - changing the way we've always done things (was:SQLfor performance-etc.)

hey all,

All this talk about SQL vs native I/O has got me to thinking.

just a superficial switch to SQL processing (imbedded selects/fetches, etc)
doesn't appear to me to have a big advantage over traditional io.  where
you get to the real savings, it seems from reading most of these posts, is
in RI, cascading deletes, triggers, view and index tuning and such.

which begs the question:  how do you implement such a monster, on an
existing application, without a major re-design/re-write of the logic in
nearly all of your programs?

for instance:  what's the good in having RI at the database level when all
your programs still do the 'chain   if not %found, error' routine?

how do you let RI handle the relationship verification, and how do you
inform the user of the errors of his ways?  test for error codes on a
write/update?  most of the code i've seen in the past 20 years is very
specific in the 'enter/edit/if wrong redisplay/if correct write/update'
schema, that it just wouldn't fit if you changed it to enter/write or
update/check for errors, redisplay schema.  I WILL require you to
completely rethink how each and every program is written, and WILL require
major revisions.

this may sound simplistic, but isn't that what we're saying when you
inforce RI at the DB level?

setting up RI for a database does you no good, and probably harms, unless
you make the corresponding application changes.  a daunting task just to
ponder, let alone implement.

Rick