I have made a clearer example of the bug I reported to -hackers
yesterday:

http://archives.postgresql.org/pgsql...0/msg01252.php

The following example shows a simple case that fails on 8.0+ (including
CVS HEAD), but works fine on 7.4. There are two almost identical
situations, and one causes an ERROR and the other a PANIC. The only
difference is the column type: INT versus TEXT, respectively.

I am on FreeBSD. An OOM condition must be caused to see this bug. In
7.4, an OOM condition is not even caused for the query, so perhaps it
has a the same bug, but handles foreign keys differently. Incidently,
foreign keys are all AFTER triggers, even in 7.4, but I don't understand
why 7.4 doesn't exhaust itself of memory collecting the trigger events,
as is described in the following mailing list post:

http://archives.postgresql.org/pgsql...5/msg00036.php

Also, and this is pure conjecture, this bug may be related to the
following change in the 8.0 release notes:
"Nondeferred AFTER triggers are now fired immediately after completion
of the triggering query, rather than upon finishing the current
interactive command. This makes a difference when the triggering query
occurred within a function: the trigger is invoked before the function
proceeds to its next operation. For example, if a function inserts a new
row into a table, any nondeferred foreign key checks occur before
proceeding with the function."

Regards,
Jeff Davis


Step 1: Create 4 tables
-----------------------------------------
CREATE TABLE r1( i INT PRIMARY KEY );
INSERT INTO r1 VALUES(1);
CREATE TABLE r2( i INT PRIMARY KEY );
INSERT INTO r2 VALUES(1);
CREATE TABLE r3( i INT PRIMARY KEY );
INSERT INTO r3 VALUES(1);
CREATE TABLE r4( i INT PRIMARY KEY );
INSERT INTO r4 VALUES(1);


Step 2: Cause an out of memory condition. The result is an ERROR, as
expected.
-----------------------------------------

BEGIN;

CREATE TABLE crashme (
attr1 INT REFERENCES r1(i),
attr2 INT REFERENCES r2(i),
attr3 INT REFERENCES r3(i),
attr4 INT REFERENCES r4(i),
attr5 TEXT
);

INSERT INTO crashme(attr1,attr2,attr3,attr4,attr5) SELECT 1,1,1,1,'t'
FROM generate_series(1,5000000);


Step 3: Do almost exacly the same thing, except attr5 is INT and not
TEXT type. This causes a PANIC instead of an ERROR. The bug is that this
should be only an ERROR, since everything is the same except the column
type for attr5.
---------------------------------------------------
BEGIN;

CREATE TABLE crashme (
attr1 INT REFERENCES r1(i),
attr2 INT REFERENCES r2(i),
attr3 INT REFERENCES r3(i),
attr4 INT REFERENCES r4(i),
attr5 INT
);

INSERT INTO crashme(attr1,attr2,attr3,attr4,attr5) SELECT 1,1,1,1,1 FROM
generate_series(1,5000000);





---------------------------(end of broadcast)---------------------------
TIP 2: Don't 'kill -9' the postmaster