On Sat, 14 Feb 2026, Tom Lane wrote:
Marcos Magueta <[email protected]> writes:
I would be willing to at least get started with a patch for this, but
before that, I want to assess the interest and thoughts on how to properly
implement it.
Everyone who has looked at that has run away screaming. The
consequences for performance, complexity, deadlock potential,
etc are all horrible. There are also fun questions about
appropriate permissions. As the spec is written, it appears
that anyone with read permission on a table can block updates
on that table (by creating an assertion that attempted updates
will fail).
Oracle have introduced a new type of permission to cater for that oddity. I
can't actually remember how their permission system hangs together with respect
to schema ownership, but they allow a schema-level declaration to say whether or
not the owner of schema X can create an assertion about objects in schema Y
(this was gleaned from watching one of their dev videos on this topic).
It strikes me that that might not be granular enough, and you could perhaps
instead allow assertions to be created for only those expressions whose re-check
circumstances match you own mutation rights (as per my other reply on this
thread), but that's probably academic at this point. An alternative would be a
whole set of other permissions too, I guess.
Cheers,
-Joe
