Re: track generic and custom plans in pg_stat_statements

On Fri, Jul 25, 2025 at 02:34:07PM -0500, Sami Imseih wrote:
> Sami Imseih <samimseih(at)gmail(dot)com> writes:
>> I think Michael's got a point. As of HEAD there are seven different
>> places that are setting this to PLAN_CACHE_NONE; who's to say that
>> pg_stat_statements or some other extension might not wish to
>> distinguish some of those sources? At the very least, user-submitted
>> versus internally-generated queries might be an interesting
>> distinction. I don't have a concrete proposal for a different
>> categorization than what we've got, but it seems worth considering
>> while we still have the flexibility to change it easily.
>
> Sure, I get it now, I think. An example is the cached plan from a sql
> in a utility statement of a prepared statement, as an example. right?

Attached is my counter-proposal, where I have settled down to four
categories of PlannedStmt:
- "standard" PlannedStmt, when going through the planner.
- internally-generated "fake" PlannedStmt.
- custom cache
- generic cache

We could decide if a few more of the internal "fake" ones are worth
subdividing, but I am feeling that this is kind of OK to start with.
If we want more granularity, I was wondering about some bits to be
able to mix one or more of these categories, but they are all
exclusive as far as I know.
--
Michael