On Wed, May 6, 2015 at 1:48 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:
>> TARGET is also very descriptive, because it situationally describes
>> either the existing tuple actually present in the table, or (from a
>> RETURNING clause) the final tuple present in the table post-UPDATE.
>> We use the term "target" for that pervasively (in the docs and in the
>> code).
>
>
> but I find that totally unconvincing. It's clear that TARGET refers to the
> table being upserted, but it's totally unclear on *which* version of the
> tuple it refers to.
Then we're simply talking about 2 different things. My understanding
is that it *is* the relation. And like UPDATE's RETURNING, it will be
the same relation/alias but a different tuple here. Andres said this
was a mutating tuple or something like that, and I suppose it is. But
Vars are variables.
Now, Andres (and now you) want to change it so that the TARGET alias
becomes magical and expression-like, so that it really does refer to a
tuple and not a relation (and so is closer to EXCLUDED.*). And you
seem pretty set on that. That being the case, clearly TARGET is
unsuitable for the reasons you state.
I suppose that it doesn't much matter, but that's how I understood the
situation all along. So I can see why you don't like "TARGET" in light
of that. I would vote for EXISTING as an alternative, given that it's
pretty clear that what is now TARGET.* will become a magic
alias/expression thing. EXISTING is the EXISTING tuple, which goes
well with EXCLUDED.
--
Peter Geoghegan