Robert Haas <robertmhaas@gmail.com> writes:
>> Also, keep in mind we want the ObjectID in all CREATE, ALTER and DROP
>> statements, so my current patch is still some bricks shy of a load… (I
>> added ObjectID only in the commands I added rewrite support for, apart
>> from DROP).
>
> I shall rely on you to provide those bricks which are still missing.
Please find attached a patch to change most functions called from
standard_ProcessUtility() to return an Oid (passes `make
maintainer-clean; configure; make install check`). Most of them only,
because it only make sense for functions touching an object that exists
in the catalogs and have a distinct Oid.
That completes ALTER and CREATE ObjectID support, I did nothing about
the DROP case in the attached. The way I intend to solve that problem is
using get_object_address() and do an extra lookup from within the Event
Trigger code path.
Yes that's an extra lookup, the only alternative that I see would be
another giant refactoring so that all and any DROP support function is
split in a lookup and lock part first, then the actual DROP. I don't see
that as a particular win either, as the advantage of doing a separate
(extra) lookup in the ddl_command_start Event Trigger is that the actual
DROP code then will check that the target still exists.
The attached patch is known to miss support for ExecCreateTableAs
because I wanted to keep the ball rolling and couldn't figure out where
to find the target relation id in time. If you feel like filling that
gap that would be awesome, if not I will take care of that later, either
in a version v1 of that patch, or in a follow-up patch, or embedded into
the next patch to come, the cleaned up Event Trigger Info patch without
Normalized Command String support. As you see fit.
Merry Christmas All,
--
Dimitri Fontaine
http://2ndQuadrant.fr PostgreSQL : Expertise, Formation et Support