Re: [HACKERS] RE: POSTGRES BUG - FIX IT PLEASE

Mikheev, Vadim writes:

> > You cannot change data twice within a transaction if there's a RI
> > constraint on the table. This is per SQL, nothing we can do about it.
> ^^^^^^^^^^^^^^^
> Is it true?!

Actually, it's not. Jan Wieck first explained this on July 23 in "Re:
[GENERAL] failed Delete after Insert in a transaction", and we've been
believing it ever since, but I just found out that it's wrong.

The standard reads

11.8 <referential constraint definition>

9) If any attempt is made within an SQL-statement to update some
site to a value that is distinct from the value to which that
site was previously updated within the same SQL-statement,
then an exception condition is raised: triggered data change
violation.

10) If a site in an object row is an <object column> of an <update
statement: positioned> or <update statement: searched>, and
there is any attempt within the same SQL-statement to delete the
row containing that site, then an exception condition is raised:
triggered data change violation.

(also 11.8 GR 8 b) i) 2), but it's too boring to quote...)

Note that it talks about "statements", not "transactions".

--
Peter Eisentraut peter_e(at)gmx(dot)net http://yi.org/peter-e/