[PATCHES] Bug in psql (on_error_rollback)

[Michael Paesold]

There is a bug in psql for the new ON_ERROR_ROLLBACK feature. In AUTOCOMMIT off mode it does not work correctly for the first statement.

This is how it works usually:
====================================
postgres=# \set AUTOCOMMIT off
postgres=# \set ON_ERROR_ROLLBACK interactive
postgres=# SELECT 1;
?column?
----------
       1
(1 row)

postgres=# SELECT a;
ERROR:  column "a" does not exist
postgres=# SELECT 1;
?column?
----------
       1
(1 row)

postgres=# BEGIN;
WARNING:  there is already a transaction in progress
BEGIN
postgres=# ROLLBACK;
ROLLBACK
====================================


====================================
postgres=# ROLLBACK;
ROLLBACK
postgres=#
postgres=# SELECT a;
ERROR:  column "a" does not exist
postgres=# SELECT 1;

====================================

With the attaced patch it works correctly even for the first statement.
====================================
postgres=# \set AUTOCOMMIT off
postgres=# \set ON_ERROR_ROLLBACK interactive
postgres=# SELECT a;
ERROR:  column "a" does not exist
postgres=# SELECT 1;
?column?
----------
       1
(1 row)

postgres=# BEGIN;
WARNING:  there is already a transaction in progress
BEGIN
postgres=# ABORT;
ROLLBACK
====================================

Please check the patch and apply to CVS tip.


Best Regards,

For the first statement in a transaction after the implicit BEGIN it does not work: ERROR: current transaction is aborted, commands ignored until end of transaction block I think it would be good to add regression tests for AUTOCOMMIT and ON_ERROR_ROLLBACK and possibly others. There are currently no regression tests specifically for psql features, but since the regression tests are executed via psql, there would be no problem in creating a set of such tests, right?. I could write some. Michael Paesold

psql.patch
Description: Binary data

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