Re: insert ... delete ... returning ... ?

Tom Lane wrote:
> "Jonah H. Harris" <jonah(dot)harris(at)gmail(dot)com> writes:
>
>> Not stupid, it doesn't work :) This was a limitation of the original
>> design based on (IIRC) executor-related issues.
>>
>
> There are definitional issues not only implementation ones; in
> particular, in subquery-like cases it's entirely unclear how many times
> the DML operation will or should get evaluated.
>

Interesting. Would it be cheating to only allow it in cases where the
evaluation should definately be only once? For example, insert ...
delete, create table ... delete, or part of a join expression?

In any case - I don't have the know how to fix it, and it's certainly
more of a "would be cute" than "I must have it." I'll settle with my
table locks for now. It's no big deal for my application.

I'm noticing a massive reduction in on disk storage required for my
database that I believe is primarily attributable due to Tom's reduced
overhead for short strings. Some of the tables I am importing have a 10
- 20 short string fields (many 0 length strings!). Unfortunately - I
wasn't looking for this specifically, so I didn't keep my old database
instance around. But I'm thinking by memory that the biggest table is
now 1/3 the number of relpages in 8.3 as it was in 8.2. Good job all
around hackers. Again - *NO* problems. It just works.

Cheers,
mark

--
Mark Mielke <mark(at)mielke(dot)cc>