OK, I propose the following further minor tweaks. (I modified the docs
following the wording we have for COSTS and BUFFERS).
There are two things that still trouble me a bit. First, we assume that
the planner is using an AllocSet context, which I guess is true, but if
somebody runs the planner in a context of a different memcxt type, it's
going to be a problem. So far we don't have infrastructure for creating
a context of the same type as another context. Maybe it's too fine a
point to worry about, for sure.
The other question is about trying to support the EXPLAIN EXECUTE case.
Do you find that case really useful? In a majority of cases planning is
not going to happen because it was already done by PREPARE (where we
_don't_ report memory, because we don't have EXPLAIN there), so it seems
a bit weird. I suppose you could make it useful if you instructed the
user to set plan_cache_mode to custom, assuming that does actually work
(I didn't try).
--
Álvaro Herrera 48°01'N 7°57'E — https://www.EnterpriseDB.com/
"El hombre nunca sabe de lo que es capaz hasta que lo intenta" (C. Dickens)