AW: [HACKERS] Re: [PATCHES] Changes to sequence.c

Billy G. Allie wrote:
>
> Vadim B. Mikheev wrote:
> >Billy G. Allie wrote:
> >>
> >> I encountered a problem (bug? feature?) where "select currval('sequence')"
> >> will generate an error if "select nextval('sequence')" is not executed
> first.
> >
> >This is feature :)
> >1. This is what Oracle does.
> >2. currval () is described as returning value returned by
> >   last nextval() in _session_.
> >
> >Vadim
> >
> Does this mean we should not modify this behavior because "this is what Oracle
> does"?  I can envision where using currval() before nextval() can be useful.

Actually, what you are proposing was initial behaviour of currval().
This was changed to be more consistent with 1. & 2. (note - not only 1.,
but 2. also).

But personally I haven't objection against changing this again.
Men, vote pls!

Vadim

No, I would not change this again, my question is iff instead of elog(ERROR
the old code could be reinserted. This would mean, that when a session did a
previous nextval it gets it's session currval, but if it did not, it gets a global
currval as in previous implementation. The problem is somebody who
calls currval often without ever calling nextval (like a monitor) will totally kill
performance. (same as a select max(field))

Andreas