[ sorry for slow response ]
Stephen Rees <srees(at)pandora(dot)com> writes:
> Code that works on Pg8.3 raises an error on Pg9.1, is this a bug?
Yes, it sure looks that way. Somebody decided they could mark userlocks
as transactional, which was probably a bad idea altogether, and
certainly wasn't followed up on adequately.
regards, tom lane