On Mon, Jun 09, 2025 at 12:44:59PM +0200, Alvaro Herrera wrote:
> I also added a recursive call in IsSquashableExpression to itself. The
> check for stack depth can be done without throwing an error. I tested
> this by adding stack bloat in that function. I also renamed it to
> IsSquashableConstant. This changes one of the tests, because a cast
> sequence like 42::int::bigint::int is considered squashable.
>
> Other than that, the changes are cosmetic.
>
> Barring objections, I'll push this soon, then look at rebasing 0003 on
> top, which I expect to be an easy job.
v9-0002 is failing in the CI for the freebsd task:
https://github.com/michaelpq/postgres/runs/43784034162
Here is the link to the diffs, also attached to this message:
https://api.cirrus-ci.com/v1/artifact/task/5378459897167872/testrun/build/testrun/pg_stat_statements/regress/regression.diffs
I am also able to reproduce these failures locally, FWIW. For
example, with a IN clause made of integer constants gets converted to
an ArrayExpr, but in _jumbleElements() we fail to call
RecordConstLocation() and the list is not squashed.
I think that this is can be reproduced by
-DWRITE_READ_PARSE_PLAN_TREES -DCOPY_PARSE_PLAN_TREES
-DRAW_EXPRESSION_COVERAGE_TEST that I always include in my builds.
The freebsd task uses the same with debug_copy_parse_plan_trees=on,
debug_write_read_parse_plan_trees=on and
debug_raw_expression_coverage_test=on.
--
Michael
