"Kevin Grittner" <Kevin.Grittner@wicourts.gov> writes:
> At this point, SERIALIZABLE transactions appear to have worked, with
> receipt 3 happening before the update of deposit_date; however, there
> was a window of time when the update to deposit_date was visible and
> receipt 3 was not.
> This absolutely can't happen in a standard-compliant implementation.
I think you mean "you'd like to believe that can't happen in a
standard-compliant implementation". It doesn't include any of the
specific behaviors that are forbidden by the spec, though, so I'm less
than convinced.
An appropriate way to prevent the problem is probably for the
transaction that changes the deposit_date to take out a write-excluding
lock on the receipts table before it does so.
regards, tom lane