Re: functions marked STABLE not allowed to do INSERT

Am Montag, den 14.11.2005, 15:06 -0500 schrieb Tom Lane:
> Tino Wildenhain <tino@wildenhain.de> writes:
> > Now this is really a bug:
> 
> That's in the eye of the beholder (and one who wasn't paying attention
> to previous discussion of this point, evidently).

Yes I was, but only to the fact it is not useable for 
caching and there are some cases (like random) 
for which STABLE would be bad thing [tm].

> The reason why the no-data-change rule is now enforced, not only
> recommended, is that a stable/immutable function now actually would
> not see any changes it did make.  Consider code like
> 
>     INSERT INTO foo VALUES (42, ...);
>     SELECT * INTO rec FROM foo WHERE key = 42;
>     IF NOT FOUND THEN
>         RAISE EXCEPTION 'where did my row go?';
> 
> If this were allowed in stable/immutable functions, the RAISE would
> in fact be reached in 8.1, because the SELECT will be done with the
> snapshot of the query that called the function.  This is a feature,

Ah this was the missing bit. I though this would only be true
for IMMUTABLE.

Thanks for the explanation. I'm not fine w/ it.

Regards
Tino