Re: a SELECT FOR UPDATE question

On Thu, Feb 10, 2005 at 10:40:44AM -0500, Tom Lane wrote:
>
> Remember RC mode takes a new snapshot for each query.  You'd need to
> use serializable mode --- and do something to actually freeze the
> transaction snapshot, which BEGIN does not --- to see the issue in a
> manual test.

It sounds like the best a check could do would be the amazingly
astute "some transaction held a lock on this row at one time and
may or may not still hold that lock, and even if it did when you
checked it might have gone away by now and some other transaction
that you don't know about might hold a lock."

Does that about sum it up?  ;-)

--
Michael Fuhr
http://www.fuhr.org/~mfuhr/