Re: [PATCHES] Work-in-progress referential action trigger

On Fri, 9 Sep 2005, Tom Lane wrote:

> Stephan Szabo <sszabo@megazone.bigpanda.com> writes:
> > Is there a case other than a before trigger updating a row we will want to
> > act upon later in the statement where we'll get a row with xmax of our
> > transaction and cmax greater than the current command?
>
> The greater-cmax case could occur via any kind of function, not only a
> trigger, ie
>
> 	update tab set x = foo(x) where ...
>
> where foo() is a volatile function that internally updates the tab
> table.

I *thought* I was missing a case, I just couldn't figure out what.

> I suppose you could say that this is horrible programming practice and
> anyone who tries it deserves whatever weird behavior ensues ... but
> it's not the case that every such situation involves a trigger.

Well, the change I was thinking of would have made it an error if foo(x)
updated a row that was then later selected by the update rather than the
current behavior which I think would have ignored the already updated row,
so that's probably not going to work.