Thread

Re: Row pattern recognition

Henson Choi <assam258@gmail.com> — 2026-05-26T22:31:02Z

Hi Tatsuo, jian,

While looking into v47, I noticed that
> raw_expression_tree_walker_impl() lacks tracking RPCommonSyntax and
> its children nodes. Probably this does nothing wrong with RPR
> functionalities but just for completeness I created a patch on top of
> v47.


Thanks for catching this. I applied the patch and the RPR regress passes
cleanly. I also re-ran it with debug_raw_expression_coverage_test turned
on (on an assert-enabled build), and the full regress is green as well.

One observation while testing: the GUC catches missing case handlers
once the walker actually reaches a node, but it cannot flag a missing
WALK on its own -- if no caller drives the walker into a subtree, the
omission stays silent. So your inspection was the part that found the
gap; the GUC just confirms the patch closes it. With the patch in, RPR
raw subtrees are on the safety net for any future node-type additions.

I'll include the patch in v48 as nocfbot-0015. My suggestion would
be to defer the fold until the jian-response patches (numbered from
0016 onward, which I'll be sending shortly) have also gone through a
review round, so the whole bundle can land together in one pass.
Patch attached below for convenience.


jian -- thanks for the thorough review. It covers a lot of ground,
and I'm still working through it. Current expectation is that most
items will be accepted; a few smaller ones may end up with a
different conclusion than your suggestion, and those are still under
analysis on my side.

The plan is to turn the responses into a patch series and send them
out for another round of review. More to follow once the batch is in
shape.


Thanks,
Henson