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

On 2015-04-26 18:02:06 -0700, Peter Geoghegan wrote:
> Remaining challenges
> =================

So I did the executor changes I'd mentioned downthread, and Peter agreed
that it'd quite workable.

Right now this, besides cleanup, docs and syntax leaves only one real
issue I know of. Which is the question what EXCLUDED actually refers to.

Consider a table
blarg(key int primary key, data text); with a BEFORE INSERT
trigger that modifies 'data'. For the statement

INSERT INTO blarg(key, data) VALUES(1, 'whatever')
ON CONFLICT (key) DO UPDATE SET data = EXCLUDED.data;

would you rather have EXCLUDED.data refer to the tuple version from
VALUES (or a SELECT or ...) or to version from the BEFORE trigger?

To me it seems better to have it refer to version *not* affected by the
trigger (which will be called either way). I think it'd be slightly
easier to understand and more flexible. But I could live with either
decision.

This isn't a technical problem, we just have to decide what we want to
do.

Greetings,

Andres Freund