pgsql gives ExecutePlan error after empty UPDATE?

Started by David C Mudiealmost 26 years ago2 messagesgeneral
Jump to latest
#1David C Mudie
mudie@digitaldeck.com

Hello! I am using PostgreSQL 7.0.2 on FreeBSD 3.4.

I am trying to update a table with the results of a
(possibly empty) selection:

zorro_4330=# create table foo(bar int4) \g
CREATE
zorro_4330=# update foo set bar = count(*) from foo \g
ERROR: ExecutePlan: (junk) `ctid' is NULL!

In this example, there are no rows in table "foo" to update.
Why is postgres returning this cryptic error instead of just UPDATE 0?

If I add a sample row, the same query works fine:

zorro_4330=# insert into foo values (1) \g
INSERT 50688 1
zorro_4330=# update foo set bar = count(*) from foo \g
UPDATE 1

If I change my query to update no rows, I get the error again:

zorro_4330=# update foo set bar = count(*) from foo where bar < 0 \g
ERROR: ExecutePlan: (junk) `ctid' is NULL!

Postgres seems to be reporting an error because the update matched no rows,
but this a perfectly legitimate database operation. I have not
been able to find any documentation on "ExecutePlan" or why it
thinks there is a problem.

Please let me know if you have any suggestions or workarounds for this problem,
or if there is somewhere more appropriate to discuss this. Thanks!

David

-----------------------------------------------------------------------------
David C Mudie DigitalDeck Inc mudie@digitaldeck.com
San Mateo CA 94402 http://www.digitaldeck.com

#2Tom Lane
tgl@sss.pgh.pa.us
In reply to: David C Mudie (#1)
Re: pgsql gives ExecutePlan error after empty UPDATE?

David C Mudie <mudie@digitaldeck.com> writes:

zorro_4330=# update foo set bar = count(*) from foo \g
ERROR: ExecutePlan: (junk) `ctid' is NULL!

Postgres seems to be reporting an error because the update matched no rows,
but this a perfectly legitimate database operation.

Actually, it's flat-out illegal according to SQL92: thou shalt not
use an aggregate in UPDATE, quoth the standard. Try it with the
count() in a sub-select, which is legal SQL:

update foo set bar = (select count(*) from foo);

There is a thread going on in pghackers right now about whether it
makes sense to allow aggregates outside sub-selects in UPDATE, and
if so what it should mean exactly.

regards, tom lane