Re: INSERT ... ON CONFLICT UPDATE/IGNORE 4.0

On 8 May 2015 at 16:51, Andres Freund <andres(at)anarazel(dot)de> wrote:

> On 2015-05-08 16:36:07 +0100, Geoff Winkless wrote:
> > I thought the previous version suggested multiple possible targets and
> > actions, this suggests that while there can be multiple targets the
> > action is always the same.
>
> I don't think any version of the patch included that functionality. I
> can see it being useful, but it'd make a bunch of things more
> complicated, so I doubt we'll get there for 9.5.
>

​I'm not particularly bothered by it - I only see it ever being useful on
the extreme edge case, and
I certainly wouldn't want
​it
to hold up the release - but it was the only reason I could see for
requiring a target_clause in the first place (I
​expect that's why I
misread the docs previously!). If you can't specify different actions based
on the target_clause, what's the point in forcing the user to enumerate it?

> example (or similar); do we think people will be smart enough to realise
> > that's possible without one?​
>
> Hm. I'm tempted to say that the synopis makes that clear enough.

​Unless I'm missing it, it's really only in "This happens on a row-by-row
basis" and in the "deterministic" paragraph that you even mention
multi-line inserts in the ON CONFLICT section.​
​​

> I
> ​ ​
> personally never check such examples though, so maybe I'm the wrong
> person to judge.
>

​I'm afraid I'm the sort of person who goes straight to the examples :)

Maybe I'll just suggest then that there's a _potential for confusion_ if
you only give examples of the first kind - people might place some
inference on the fact that the examples only show single-row INSERTs.

Geoff