Re: Row pattern recognition

From: Henson Choi <assam258(at)gmail(dot)com>
To: Zsolt Parragi <zsolt(dot)parragi(at)percona(dot)com>, Tatsuo Ishii <ishii(at)postgresql(dot)org>
Cc: vik(at)postgresfriends(dot)org, er(at)xs4all(dot)nl, jacob(dot)champion(at)enterprisedb(dot)com, david(dot)g(dot)johnston(at)gmail(dot)com, peter(at)eisentraut(dot)org, pgsql-hackers(at)postgresql(dot)org
Subject: Re: Row pattern recognition
Date: 2026-03-11 23:52:08
Message-ID: CAAAe_zAFv0E_EhFxZueJ1n9pZpPaTVChMhDmndzugzovgz0o2Q@mail.gmail.com
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Hi Zsolt, Tatsuo,

+ foreach_ptr(TargetEntry, te, defineClause)
> + (void) coerce_to_boolean(pstate, (Node *) te->expr, "DEFINE");
>
> Isn't this incorrect? I think it should update te->expr, as currently
> it is possible to construct queries where this produces unexpected
> results.

Good catch, Zsolt. You're right — the return value of
coerce_to_boolean() must be assigned back to te->expr, otherwise
any implicit cast (e.g., a user-defined type with an assignment
cast to boolean) is silently discarded.

The fix is straightforward:

foreach_ptr(TargetEntry, te, defineClause)
te->expr = (Expr *) coerce_to_boolean(pstate, (Node *) te->expr,
"DEFINE");

I've confirmed that your example query produces incorrect results
without the fix (the truthyint value is evaluated as-is without
the cast) and correct results with it.

Patch 13 is attached with the fix and a regression test case based
on your example.

By the way, thank you for joining the TDE hooks discussion when
I was just getting started as a contributor — that meant a lot.
Great work on pg_tde at Percona, and rooting for the extensible
SMGR effort as well. Hope it's all going well!

Tatsuo, please review when you get a chance.

Regards,
Henson

Attachment Content-Type Size
0013-fix-define-boolean-coercion.txt text/plain 3.5 KB

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Daniel Gustafsson 2026-03-11 23:56:07 Re: Changing the state of data checksums in a running cluster
Previous Message Andres Freund 2026-03-11 23:09:26 Re: Buffer locking is special (hints, checksums, AIO writes)