hubert depesz lubaczewski <depesz@depesz.com> writes:
> so - as you can see in one query, the value changes in plpgsql. and i need some
> id that will be unchanged within one end-user-supplied query.
AFAIR, the only state that's guaranteed to work like that is
statement_timestamp. Of course you have to worry whether your machine
is fast enough to do more than one client interaction within whatever
the clock resolution is.
I think the real question here is why you want this behavior at all;
to me it smells of not having thought the problem through correctly.
As an example of why this bothers me: what if the user's query is
rewritten into several queries by a RULE? Should you consider each
of those to be a separate user-issued SQL command? Does your answer
change if you know that the user himself prepared the RULE? (Do you
think users will be happy if statement X followed by statement Y
acts differently in a rule than elsewhere?)
regards, tom lane