I have found a problem with pg_dump (9.3 and 9.4) against a specially
crafted 9.3 database.
This is not a critical bug for me.
Here is a capture from an xterm session, showing the full sequence of
commands to reproduce the problem.
marc:skit$ psql
psql (9.4.0, server 9.3.5)
Type "help" for help.
postgres=3D# drop database regressdb;
DROP DATABASE
postgres=3D# create database regressdb;
CREATE DATABASE
postgres=3D# \q
marc:skit$ psql -d regressdb
psql (9.4.0, server 9.3.5)
Type "help" for help.
regressdb=3D# create=20
regressdb-# function ddl_notice() returns event_trigger as
regressdb-# $$
regressdb$# begin
regressdb$# raise notice 'DDL!!!!!!';
regressdb$# end;
regressdb$# $$
regressdb-# language plpgsql;
CREATE FUNCTION
regressdb=3D#=20
regressdb=3D#=20
regressdb=3D# create event trigger ddl_notice on ddl_command_start
regressdb-# when tag in ('DROP FUNCTION')
regressdb-# execute procedure ddl_notice();
CREATE EVENT TRIGGER
regressdb=3D#=20
regressdb=3D# alter event trigger ddl_notice disable;
ALTER EVENT TRIGGER
regressdb=3D#=20
regressdb=3D# comment on event trigger ddl_notice is
regressdb-# 'an event trigger';
COMMENT
regressdb=3D# \q
marc:skit$ pg_dump regressdb >x
cannot duplicate null pointer (internal error)
marc:skit$=20
This is on debian 7.7 with the following packages installed:
dpkg info:
postgresql-client-9.4 9.4.0-1.pgdg70+1=20
postgresql-9.3 9.3.5-2.pgdg70+1
uname -a:
Linux bloodnok 3.6.3-upstream #1 SMP Tue Nov 13 16:18:51 PST 2012 x86_64 G=
NU/Linux
Also reproduced on my i686 laptop with pg9.3 client and server.
The problem seems to be related to the comment on the event trigger. If
I remove the comment, the failure no longer occurs.
__
Marc
