Michael Fuhr <mike@fuhr.org> writes:
> Would LOAD 'plpgsql' work? Would that cause a fresh compile of the
> function the next time it's called, resulting in a new cached plan?
I think that would cause plpgsql to lose track of its entire function
table, which is a brute force way of doing that ... but it doesn't
really solve Nicola's problem, because the nasty part of this is
plans that are already cached by other backends.
regards, tom lane