BUG #6360: with hold cursor, cause function executed twice and wrong results

The following bug has been logged on the website:

Bug reference: 6360
Logged by: ctwang
Email address: wcting163(at)163(dot)com
PostgreSQL version: 9.0.5
Operating system: linux 2.6.18-128.7AXS3
Description:

create table test_execute(id int,name varchar(40));
insert into test_execute values(1,'jack');

create or replace function p_test_execute() returns void
as
$$
begin
raise notice 'hello world';
update test_execute set id=id*2;
end;
$$ LANGUAGE plpgsql;

begin;
declare JDBC_CURS_1 cursor with hold for select p_test_execute() from
test_execute;
fetch 50 from JDBC_CURS_1;
NOTICE: hello world

end;
NOTICE: hello world
COMMIT

select * from test_execute;
id | name
----+------
4 | jack

I expect id = 2, but it is **4** instead,

The reason is that the function p_test_execute is executed twice, when
*fetch*, it is first executed, and when transaction commit, because the
cursor is a *holdable* cursor, it is executed again.

I read the code, for holdable cursor, when commit, following call will
execute:
CommitHoldablePortals-->PersistHoldablePortal-->ExecutorRewind

Is *ExecutorRewind* necessary, is it the root of this bug?
Does *ExecutorRewind* cause plan re-execute?