On Tue, Aug 18, 2009 at 10:32:35AM +0200, Daniel Verite wrote:
> Sam Mason wrote:
> > I've just realized another case where it's not consistent; why does the
> > following return true:
> >
> > SELECT row(null) IS NULL;
> >
> > and yet the following false:
> >
> > SELECT row(row(null)) IS NULL;
>
> You're intentionally assuming that row(null) IS NULL evaluating to true
> implies that row(null) can be replaced by NULL. As discussed upthread, this
> is not the case.
But you've still not said how is this useful! I can reformulate maths
so that 1+0 <> 1+(0), but this is not useful behavior. Programmers
need logical abstractions upon which to build and without them you end
up with even more bugs.
> > I think I'm saying that PG should be deliberately breaking specified
> > behavior and go back to pre-8.2 behavior in this regard.
>
> But let's run your example with 8.1:
>
> # SELECT row(null) IS NULL;
> ?column?
> ----------
> t
>
> # SELECT row(row(null)) IS NULL;
> ?column?
> ----------
> f
>
> These are the same results that you say are inconsistant, so pre-8.2 behavior
> doesn't help here...
Doh, that'll learn me--I never actually tried older versions. I was
just repeating what the docs said about the behavior changing in 8.2.
http://www.postgresql.org/docs/current/static/functions-comparison.html#AEN7444
and was mis-interpreting what it was saying.
--
Sam http://samason.me.uk/
