Date: Thu, 30 Jan 2025 08:26:54 -0500
From: Bruce Momjian <bruce@momjian.us>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: jbe-mlist@magnetkern.de,
	PostgreSQL-development <pgsql-hackers@lists.postgresql.org>
Subject: Re: Parameter NOT NULL to CREATE DOMAIN not the same as CHECK (VALUE
 IS NOT NULL)
Message-ID: <Z5t-HtfrECeNegGZ@momjian.us>
References: <173591158454.714.7664064332419606037@wrigleys.postgresql.org>
 <Z37p0paENWWUarj-@momjian.us>
 <1694911.1736371474@sss.pgh.pa.us>
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="kWMRNxtW5m9DMfJ2"
Content-Disposition: inline
In-Reply-To: <1694911.1736371474@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/Z5t-HtfrECeNegGZ%40momjian.us>
Precedence: bulk


--kWMRNxtW5m9DMfJ2
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline

On Wed, Jan  8, 2025 at 04:24:34PM -0500, Tom Lane wrote:
> Bruce Momjian <bruce@momjian.us> writes:
> > I think this needs some serious research.
> 
> We've discussed this topic before.  The spec's definition of IS [NOT]
> NULL for composite values is bizarre to say the least.  I think
> there's been an intentional choice to keep most NOT NULL checks
> "simple", that is we look at the overall value's isnull bit and
> don't probe any deeper than that.
> 
> If the optimizations added in v17 changed existing behavior,
> I agree that's bad.  We should probably fix it so that those
> are only applied when argisrow is false.

I have developed the attached patch using your argisrow suggestion which
fixes the test I posted.  Is this something we should backpatch?

-- 
  Bruce Momjian  <bruce@momjian.us>        https://momjian.us
  EDB                                      https://enterprisedb.com

  Do not let urgent matters crowd out time for investment in the future.


--kWMRNxtW5m9DMfJ2
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment; filename="domain.diff"

diff --git a/doc/src/sgml/ref/create_domain.sgml b/doc/src/sgml/ref/create_domain.sgml
index ce555203486..c111285a69c 100644
--- a/doc/src/sgml/ref/create_domain.sgml
+++ b/doc/src/sgml/ref/create_domain.sgml
@@ -283,7 +283,8 @@ CREATE TABLE us_snail_addy (
   <para>
    The syntax <literal>NOT NULL</literal> in this command is a
    <productname>PostgreSQL</productname> extension.  (A standard-conforming
-   way to write the same would be <literal>CHECK (VALUE IS NOT
+   way to write the same for non-composite data types would be
+   <literal>CHECK (VALUE IS NOT
    NULL)</literal>.  However, per <xref linkend="sql-createdomain-notes"/>,
    such constraints are best avoided in practice anyway.)  The
    <literal>NULL</literal> <quote>constraint</quote> is a
diff --git a/src/backend/optimizer/plan/initsplan.c b/src/backend/optimizer/plan/initsplan.c
index 2cb0ae6d659..e291547112d 100644
--- a/src/backend/optimizer/plan/initsplan.c
+++ b/src/backend/optimizer/plan/initsplan.c
@@ -3100,6 +3100,13 @@ restriction_is_always_true(PlannerInfo *root,
 		if (nulltest->nulltesttype != IS_NOT_NULL)
 			return false;
 
+		/*
+		 * Empty rows can appear NULL in some contexts and NOT NULL in others,
+		 * so avoid this optimization for row expressions.
+		 */
+		if (nulltest->argisrow)
+			return false;
+
 		return expr_is_nonnullable(root, nulltest->arg);
 	}
 
@@ -3149,6 +3156,13 @@ restriction_is_always_false(PlannerInfo *root,
 		if (nulltest->nulltesttype != IS_NULL)
 			return false;
 
+		/*
+		 * Empty rows can appear NULL in some contexts and NOT NULL in others,
+		 * so avoid this optimization for row expressions.
+		 */
+		if (nulltest->argisrow)
+			return false;
+
 		return expr_is_nonnullable(root, nulltest->arg);
 	}
 

--kWMRNxtW5m9DMfJ2--