Re: Minor ON CONFLICT related fixes

On Mon, May 11, 2015 at 7:11 PM, Andres Freund <andres(at)anarazel(dot)de> wrote:
> Do I understand correctly that you cut this out because there
> essentially was a ruleutils bug with with EXCLUDED? If so, I don't find
> that acceptable. I'm pretty strictly convincded that independent of rule
> support, we shouldn't add things to insert queries that get_query_def
> can't deparse.

No, that wasn't the reason -- deparsing itself works fine. That code
remains within ruleutils.c because I agree with this principle of
yours.

I tested the deparsing logic before making the case added fail as
unsupported. If you remove the new ereport() call that relates to
non-suppport of ON CONFLICT DO UPDATE as a rule action, then the
CREATE RULE still succeeds, and you can deparse the query just fine
(by quering pg_rules, typically). You just get an error within the
optimizer following rewriting of a query involving the application of
a rule that specifies an ON CONFLICT DO UPDATE alternative/DO INSTEAD
action. I found it would say: "variable not found in subplan target
lists".

I can't say that I explored the question very thoroughly, but it seems
bothersome to have more than one relation involved in a Query involved
in rewriting.

--
Peter Geoghegan