From: Tom Lane <tgl@sss.pgh.pa.us>
To: Tomas Vondra <tomas@vondra.me>
cc: Peter Eisentraut <peter@eisentraut.org>,
        pgsql-hackers@lists.postgresql.org
Subject: Bug in brin_minmax_multi_distance_numeric()
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <2093711.1753983215.1@sss.pgh.pa.us>
Date: Thu, 31 Jul 2025 13:33:35 -0400
Message-ID: <2093712.1753983215@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/2093712.1753983215%40sss.pgh.pa.us>
Precedence: bulk

Per the discussion in [1], I think we need something like

-	PG_RETURN_FLOAT8(DirectFunctionCall1(numeric_float8, d));
+	PG_RETURN_DATUM(DirectFunctionCall1(numeric_float8, d));

in brin_minmax_multi_distance_numeric().  Peter was proposing
that as cosmetic cleanup, but I think it's actually a functional
bug that needs to be back-patched.  It is certainly broken on
32-bit machines where the Datum result of numeric_float8 will
be a pointer, so that we will convert the numeric pointer value
to a float and return that, yielding a totally-garbage distance
value.  But I think it's broken on 64-bit machines too, because
we'll be interpreting the output of numeric_float8 as a uintptr_t
and applying some unwanted conversion to make that a float.

It's not clear to me exactly what this function is used for,
but I guess the consequences would be broken BRIN indexes
on numeric columns?

			regards, tom lane

[1] https://www.postgresql.org/message-id/2091622.1753982274%40sss.pgh.pa.us