Re: Set query_id for query contained in utility statement

On Tue, Oct 01, 2024 at 09:26:40AM +0200, Anthonin Bonnefoy wrote:
> This seems to be a known issue. The test_prepare_pgss1's parameters
> are A_Const nodes. Those nodes have a custom query jumble which
> doesn't record location[1] and thus can't be normalised by pgss.
>
> That could be fixed by replacing the switch on nodeTag by
> JUMBLE_LOCATION(location) but this will impact a lot of DDL queries
> and the result doesn't look great (for example, "BEGIN TRANSACTION NOT
> DEFERRABLE, READ ONLY, READ WRITE, DEFERRABLE" would be normalised as
> "BEGIN TRANSACTION $1 DEFERRABLE, $2 ONLY, $3 WRITE, $4")

Yeah, I've made peace with myself regarding the fact that adding a
location tracker to A_Const is just a very bad idea and that we should
never do that, ever. So what I proposed on this thread is a no-go. I
am not saying that there is no solution, just that this solution is a
bad one.

You can see the extent of the damages this would cause with the diffs
generated in pg_stat_statements. If there is a gap in the tests for
commands using A_Const nodes compared to what's on HEAD, we should
expand that to properly track how normalization would apply to each
one of them (see the recent one for SET queries, for example). This
is really useful when manipulating the parse nodes and query jumbling
with some pg_node_attr().
--
Michael