Content-Type: text/plain;
	charset=utf-8
Mime-Version: 1.0 (Mac OS X Mail 16.0 \(3864.400.21\))
Subject: Re: table AM option passing
From: Chao Li <li.evan.chao@gmail.com>
In-Reply-To: <202603311555.w42t5agupy3z@alvherre.pgsql>
Date: Wed, 1 Apr 2026 14:33:28 +0800
Cc: Andres Freund <andres@anarazel.de>,
 Pg Hackers <pgsql-hackers@lists.postgresql.org>,
 Zsolt Parragi <zsolt.parragi@percona.com>,
 Nathan Bossart <nathandbossart@gmail.com>
Content-Transfer-Encoding: quoted-printable
Message-Id: <40E570EE-5A60-49D8-B8F7-2F8F2B7C8DFA@gmail.com>
References: <202603311555.w42t5agupy3z@alvherre.pgsql>
To: =?utf-8?Q?=C3=81lvaro_Herrera?= <alvherre@kurilemu.de>
Archived-At: 
 <https://www.postgresql.org/message-id/40E570EE-5A60-49D8-B8F7-2F8F2B7C8DFA%40gmail.com>
Precedence: bulk


> On Apr 1, 2026, at 00:10, =C3=81lvaro Herrera <alvherre@kurilemu.de> =
wrote:
>=20
> Hi,
>=20
> On 2026-Mar-30, =C3=81lvaro Herrera wrote:
>=20
>> I just pushed the part of 0001 that modifies the API of table_insert =
and
>> its sibling functions.  So here 0001 adds the options bitmask to
>> table_update and table_delete, which is important for REPACK; and =
0002
>> is identical as before and makes the interface (IMO) a bit more
>> future-proof.
>=20
> FWIW I just posted 0001 as part of the repack series here [1]; 0002 is
> inessential, so I didn't; but here's both of them.  I have drafted
> commit messages also.
>=20
> [1] https://postgr.es/m/202603311523.iqhng5ljkzpq@alvherre.pgsql
>=20
> --=20
> =C3=81lvaro Herrera               48=C2=B001'N 7=C2=B057'E  =E2=80=94  =
https://www.EnterpriseDB.com/
> "Pensar que el espectro que vemos es ilusorio no lo despoja de =
espanto,
> s=C3=B3lo le suma el nuevo terror de la locura" (Perelandra, C.S. =
Lewis)
> =
<v4-0001-Give-options-parameter-to-table_delete-table_upda.patch><v4-0002-=
Change-heapam.c-to-obey-tableam.h-option-bits-dir.patch>

Looks like v4-0001/0002 are pure refactoring. A few small comments:

1 - 0001
For table_tuple_delete(), options is added and changingPart is removed, =
but the header comment should be updated to reflect the change.

2 - 0002
For table_tuple_update(), options is added, the header comment should be =
updated as well.

3 - 0002
Now TABLE_INSERT_SKIP_FSM, TABLE_INSERT_FROZEN, TABLE_INSERT_NO_LOGICAL =
belong to the same options word as HEAP_INSERT_SPECULATIVE, but they are =
still defined as:
```
#define TABLE_INSERT_SKIP_FSM 0x0002
#define TABLE_INSERT_FROZEN 0x0004
#define TABLE_INSERT_NO_LOGICAL 0x0008
```

Could it make sense to change them to the left-shift form?

4 - 0002=20
In heap_multi_insert(), heap_prepare_insert(), and heap_insert(), =
options is defined as uint32, but in RelationGetBufferForTuple() and =
raw_heap_insert() it is still defined as int. Would it make sense to =
take this opportunity to change all =E2=80=9Coptions" to uint32 for =
consistency? Otherwise, if this is left for later, a trivial follow-up =
patch just to change int to uint32 may be harder to get processed.

Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/