Re: Revisited: Transactions, insert unique.

"Ross J. Reedstrom" wrote:
>
> On Mon, Apr 24, 2000 at 11:01:57AM -0700, Joachim Achtzehnter wrote:
>
> > [...] It is a perfectly valid approach when used with an SQL92 compliant
> > database. We just have to live without it until postgresql improves
> > on this point. This is certainly not a show stopper for most of us
> > unless perhaps when somebody has to port a ton of code from another
> > database :-(
>
> I'm going to jump in here, because this is a particular axe I grind:
>
> I've bent my brain around the SQL92 standards docs, and there's _no_
> requirement for this type of behavior on error. Yes, it's a useful thing
> to have, and yes, all the bigname commercial RDBMS's work that way, but that
> doesn't mean postgres isn't SQL92 compliant on that point (it misses on
> other points, though). So, go ahead and complain, I agree it's a pain for
> those porting code. But don't say it's a standards issue, until you can
> point to chapter and verse to defend your position.

Fair enough. My "non-std" comment an unfortunate choice of words. I was
alluding to pgsql "doing it differently" from everyone else (i.e., in a
"non-std manner"), not to a violation of an "official" SQL standard. My
recollection of the recent SQL standards discussion on this point was that
at least one standards document (which one?) was ambiguous on this point.

Regards,
Ed Loehr