Problems with simple_heap_update and Form_pg_relcheck

Started by Holger Krugabout 24 years ago4 messages
#1Holger Krug
hkrug@rationalizer.com

If have some problem with updating `pg_relcheck' entries. Who can help ?

PostgreSQL version:

CVS of Jan 03 with some additions for my CREATE ERROR TRIGGER project
(announced on this list). Among those is the following change in
pg_relcheck:

 	NameData	rcname;
 	text		rcbin;
 	text		rcsrc;
+	int2		rcerrhandlers;/* # of ON ERROR triggers,
+					 * currently on 0 or 1 is supported */
 } FormData_pg_relcheck;

rcerrhandlers is the number of error handlers for a CHECK constraint,
i.e. the number of procedures to be called if a CHECK constraint fails.
Error handlers are stored in pg_triggers like normal triggers. They
reference CHECK constraints by OID in the same way as normal triggers
reference relations by OID. For this to work I have still to add an OID
to pg_relcheck, but the current question does not depend on this.

Error handlers are created by the following statement:

=# create table mytab (arg text CHECK (length(arg)>10));
CREATE
=# CREATE ERROR TRIGGER ehtest on CONSTRAINT CHECK mytab_arg FOR EACH ROW EXECUTE PROCEDURE __proc('arg');
CREATE

In executing the CREATE ERROR TRIGGER command I have to update the entry
of the CHECK constraint in pg_relcheck and ro add an entry to pg_trigger.
The problem I have is related to updating the pg_relcheck entry.

For this I use the following code:

/* Grab an exclusive lock on the pg_relcheck relation */
rcrel = heap_openr(RelCheckRelationName, RowExclusiveLock);

/*
* Create the scan key. We need to match the name of the CHECK
* constraint.
*/
ScanKeyEntryInitialize(&key, 0, Anum_pg_relcheck_rcname,
F_NAMEEQ, PointerGetDatum(constrName));
/*
* Begin scanning the heap
*/
rcscan = heap_beginscan(rcrel, 0, SnapshotNow, 1, &key);

/*
* We take the first CHECK constraint we can find.
*/

if (! HeapTupleIsValid(rctup = heap_getnext(rcscan, 0)))
{
elog(ERROR,"CreateErrorHandler: CHECK constraint \"%s\" does not exist",
constrName);
}

/**
* Copy the tuple to be able to end the scan.
*/
rctup = heap_copytuple(rctup);

heap_endscan(rcscan);

[snip]

/**
* update the error handler counter of the constraint
*/
((Form_pg_relcheck) GETSTRUCT(rctup))->rcerrhandlers = found + 1;
simple_heap_update(rcrel, &rctup->t_self, rctup);

/**
* update the indices, too
*/
[snip]
heap_freetuple(rctup);

heap_close(rcrel, RowExclusiveLock);

pg_relcheck before the update is done:

# select * from pg_relcheck where rcname='mytab_arg';
rcrelid | rcname | rcbin .. | rcsrc .. | rcerrhandlers
16570 | mytab_arg | ({ EXPR :typeOid 16 :opType op :oper { OPER :opno 521 :opid 147 :opresulttype 16 } :args ({ EXPR :typeOid 23 :opType func :oper { FUNC :funcid 1317 :functype 23 } :args ({ VAR :varno 1 :varattno 1 :vartype 25 :vartypmod -1 :varlevelsup 0 :varnoold 1 :varoattno 1})} { CONST :consttype 23 :constlen 4 :constbyval true :constisnull false :constvalue 4 [ 10 0 0 0 ] })}) | (length(arg) > 10) | 0
(1 row)

pg_relcheck after the update is done:

=# select * from pg_relcheck where rcname='mytab_arg';
rcrelid | rcname | rcbin | rcsrc | rcerrhandlers
---------+-----------+---------------+--------------------+---------------
16564 | mytab_arg | ({ EXPR :typ | (length(arg) > 10) | 0
(1 row)

Furthermore when I abort the transaction (with `elog') after the code
segment shown above is executed, the entry in pg_relcheck even vanishes !

I'm really puzzled.

--
Holger Krug
hkrug@rationalizer.com

#2Holger Krug
hkrug@rationalizer.com
In reply to: Holger Krug (#1)
Re: Problems with simple_heap_update and Form_pg_relcheck

On Fri, Jan 11, 2002 at 11:52:39AM +0100, Holger Krug wrote:

Furthermore when I abort the transaction (with `elog') after the code
segment shown above is executed, the entry in pg_relcheck even vanishes !

This remark applies to a slightly different version of the code, when
I do not heap_copytuple but modify the value in place. OK, that's
understandable, so please forget it. But the problem, that
`rcerrhandlers' is not updated, remains.

--
Holger Krug
hkrug@rationalizer.com

#3Tom Lane
tgl@sss.pgh.pa.us
In reply to: Holger Krug (#1)
Re: Problems with simple_heap_update and Form_pg_relcheck

Holger Krug <hkrug@rationalizer.com> writes:

((Form_pg_relcheck) GETSTRUCT(rctup))->rcerrhandlers = found + 1;

What's an "rcerrhandlers"? It doesn't appear in current sources.

pg_relcheck after the update is done:
[rcbin is messed up]

I have a feeling you added rcerrhandlers after the variable-length
fields. Not good, at least not if you want to access it via a C
struct. See, eg,
http://developer.postgresql.org/cvsweb.cgi/pgsql/src/backend/catalog/README

regards, tom lane

#4Holger Krug
hkrug@rationalizer.com
In reply to: Tom Lane (#3)
Re: Problems with simple_heap_update and Form_pg_relcheck

On Fri, Jan 11, 2002 at 11:57:27AM -0500, Tom Lane wrote:

Holger Krug <hkrug@rationalizer.com> writes:

((Form_pg_relcheck) GETSTRUCT(rctup))->rcerrhandlers = found + 1;

What's an "rcerrhandlers"? It doesn't appear in current sources.

Yes, it's part of my project to add error handlers. It's the number
of error handlers for this constraint, currently my code allows
as values either 0 or 1.

I have a feeling you added rcerrhandlers after the variable-length
fields. Not good, at least not if you want to access it via a C
struct. See, eg,
http://developer.postgresql.org/cvsweb.cgi/pgsql/src/backend/catalog/README

I have that feeling, too. Thanks for the prompt help.

--
Holger Krug
hkrug@rationalizer.com