Re: FOR EACH ROW triggers on partitioned tables

Thomas Munro wrote:

> What is this test for?
>
> +create trigger failed after update on parted_trig
> + referencing old table as old_table
> + for each statement execute procedure trigger_nothing();
>
> It doesn't fail as you apparently expected. Perhaps it was supposed
> to be "for each row" so you could hit your new error with
> errdetail("Triggers on partitioned tables cannot have transition
> tables.")?

You're absolutely right. Fixed in the attached version.

I also include two requisite fixes for missing CCI calls in existing
code: one is in StorePartitionBounds which I think is backpatchable to
pg10 (this is the one that was causing me to add the one Peter
complained about in [1]), and the others are in the partition indexing
code. In terms of the current tests, the first one is necessary in
order for things to work after this patch; the ones in the second patch
I only added after code review in order to understand where the first
one was. (In that second patch I also remove one which now seems
unnecessary and in hindsight was probably there because I was lacking
the others.)

Patch 0003 is the feature at hand. Compared to v3, this version adds
some recursing logic during ENABLE/DISABLE TRIGGER, so the test that was
previously failing now works correctly.

I kept the test on "irregular" partitioning from v5, too; it works here
without any further changes.

One thing I'd like to add before claiming this committable (backend-
side) is enabling constraint triggers. AFAICT that requires a bit of
additional logic, but it shouldn't be too terrible. This would allow
for deferrable unique constraints, for example.

[1] https://postgr.es/m/171cb95a-35ec-2ace-3add-a8d16279f0bf@2ndquadrant.com

--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services