On Fri, 23 Jan 2003, Hannu Krosing wrote:
> > 1. [OIDs are] not a relational concept.
> so are other system tuples (cid, tid, tableiod, ...).
But there's a key difference here; nobody's advertising these others as
any sort of row identifier: i.e., a candidate key. And besides, I wouldn't
object at all to getting rid of these, except that they store essential
system information and I can't figure out how to get rid of them. :-)
> It is an OO concept.
Well, it's not, because we have an OID wrap-around problem, so an OID is
actually not an OID at all, but simply an arbitrary number tacked on to
a row. Other rows, in the same or other tables can have the same OID.
> > 2. The OID wraparound problem can get you.
> put an unique index on OID column.
That still doesn't make it a real OID, because you can't guarantee that
two rows in different tables won't have the same OID.
> > 3. Other SQL databases don't do this.
> Ask Date, hell tell you that SQL is evil, i.e. not relational ;)
I did, he said that, and I agreed with him. :-) So now we have something
that's evil because it's not relational and also evil because it's not
SQL. Double-yuck!
> > 5. We should default to what gives us better performance, rather than
> > worse.
> Not if it breaks anything ;)
I disagree. We have to weigh the cost of the breakage versus the
benefits in each individual circumstance. We've broken plenty of things
before because we felt it was better to do so than maintain backward
compatability.
Because of its history as a research tool, there's a lot of experimental
stuff in postgres that, in hindsight, we can say didn't work so well.
When we find something that's not working so well, we should at least
consider making some sort of move toward the "right thing," rather than
continuing to do the wrong thing forever just for the sake of backwards
compatability.
Summary: I don't want to hear absolutes like "we should never break
backwards compatibility." I want to hear arguments that the cost of
breaking backwards compatability is X, and the benefit of the new way of
doing things is Y, and here is why you think X > Y.
cjs
--
Curt Sampson <cjs@cynic.net> +81 90 7737 2974 http://www.netbsd.org Don't you know, in this new Dark Age, we're
alllight. --XTC