Re: Row pattern recognition

Hi Henson,

Attached is the v45 patches. These are basically your patches
(0001-0013) on top of v44, plus rebase. Major changes/fixes since v44
include:

- Rebase. Due to commits (SQL/PG) rebase was required.

I had to resolve shift/reduce conflicts while rebasing gram.y. In
the end, I removed follwing in the v44 patch.
---------------------------------------------------------------------------
qual_Op: Op
{ $$ = list_make1(makeString($1)); }
| OPERATOR '(' any_operator ')'
{ $$ = $3; }
+ | '|'
+ { $$ = list_make1(makeString("|")); }
;
---------------------------------------------------------------------------

Also, we don't need the patch to scan.l anymore since the necessary
changes are already there, thanks to SQL/PGQ commit.

- Fix elog message to use lowercase per PostgreSQL convention

Minor style fix: lowercase the elog error message in
nodeWindowAgg.c to follow PostgreSQL coding convention.

- Fix RPR reluctant quantifier flag lost during VIEW serialization

The reluctant flag on quantifiers was lost when a VIEW containing
RPR was serialized and restored via ruleutils.c. This patch fixes
gram.y, rpr.c, ruleutils.c, and parsenodes.h so the flag survives
the round-trip.

- Expand RPR test coverage and improve test comments

Reorganizes test cases across rpr.sql, rpr_base.sql, rpr_explain.sql,
and rpr_nfa.sql. Moves base functionality tests from rpr.sql to
rpr_base.sql and NFA-specific tests to rpr_nfa.sql. Adds better
section comments throughout. Duplicate tests in rpr.sql were
removed. No reduction in test coverage.

- Keep RPR test objects for pg_upgrade/pg_dump testing

Adjusts rpr_base.sql and rpr_explain.sql so that test tables and
views are not dropped at the end of the test. This allows
pg_upgrade and pg_dump regression testing to cover RPR objects.

- Disable run condition pushdown for RPR windows
Revised version of Tatsuo's run condition fix [1]. Existing test
expected output updated accordingly.

- Disable frame optimization for RPR windows

Prevents the planner from optimizing the window frame for RPR
windows. The frame clause in RPR has different semantics (it
bounds the pattern search space), so standard frame optimizations
must be disabled. Adds EXPLAIN tests to verify. Please review
this one -- it's a newly discovered issue.

- Add stock scenario tests for RPR pattern matching

Adds stock trading scenario tests using realistic synthetic data
(stock.data with 1632 rows). Tests V-shape recovery, W-shape,
consecutive rises, and other common pattern matching use cases.

- Fix zero-min reluctant quantifier to produce zero-length match

Fixes the bug reported by SungJun [2]: reluctant quantifiers with
min=0 (A*?, A??) were incorrectly consuming at least one row instead
of producing a zero-length match. The NFA can now internally
distinguish between a zero-length match and no match, so it will
be ready when additional infrastructure (e.g. MATCH_NUMBER) is
added. Now matches Oracle behavior.

- Fix coerce_to_boolean result not applied in DEFINE clause

[1]
https://www.postgresql.org/message-id/20260304.153822.445473532741409674.ishii@postgresql.org
[2]
https://www.postgresql.org/message-id/CAE+cgNiUbKeH1A0PoxV2QjpsoxJLe+pJcGz_gdxwOwu_9zqchw@mail.gmail.com

Best regards,
--
Tatsuo Ishii
SRA OSS K.K.
English: http://www.sraoss.co.jp/index_en/
Japanese:http://www.sraoss.co.jp