Re: Surprising behaviour of \set AUTOCOMMIT ON

I don't think patch is doing correct things:

1)

postgres=# \set AUTOCOMMIT off
postgres=# create table foo (a int );
CREATE TABLE
postgres=# \set AUTOCOMMIT on

Above test not throwing psql error, as you used strcmp(newval,"ON"). You
should use pg_strcasecmp.

2)

postgres=# \set AUTOCOMMIT OFF
postgres=# create table foo ( a int );
CREATE TABLE
postgres=# \set VERBOSITY ON
\set: Cannot set VERBOSITY to ON inside a transaction, either COMMIT or ROLLBACK and retry

Above error coming because in below code block change, you don't have check for
command (you should check opt0 for AUTOCOMMIT command)

-            if (!SetVariable(pset.vars, opt0, newval))
+            if (transaction_status == PQTRANS_INTRANS && !pset.autocommit &&
+                !strcmp(newval,"ON"))
+            {
+                psql_error("\\%s: Cannot set %s to %s inside a transaction, either COMMIT or ROLLBACK and retry\n",
+                            cmd, opt0, newval);
+                success = false;
+            }

3)

postgres=# BEGIN;
BEGIN
postgres=# create table foo ( a int );
CREATE TABLE
postgres=# \set AUTOCOMMIT ON

Don't you think, in above case also we should throw a psql error?

4)

postgres=# \set AUTOCOMMIT off
postgres=# create table foo ( a int );
CREATE TABLE
postgres=# \set XYZ ON
\set: Cannot set XYZ to ON inside a transaction, either COMMIT or ROLLBACK and retry

May be you would like to move the new code block inside SetVariable(). So that
don't need to worry about the validity of the variable names.

Basically if I understand correctly, if we are within transaction and someone
tries the set the AUTOCOMMIT ON, it should throw a psql error. Correct me
if I am missing anything?