Feedback-ID: i675e48f3:Fastmail
MIME-Version: 1.0
Date: Sun, 19 Apr 2026 15:21:41 -0400
From: "Greg Burd" <greg@burd.me>
To: "David Rowley" <dgrowleyml@gmail.com>, "Tom Lane" <tgl@sss.pgh.pa.us>
Cc: "Chao Li" <li.evan.chao@gmail.com>,
 "PostgreSQL Hackers" <pgsql-hackers@lists.postgresql.org>
Message-Id: <77f5109f-6d39-4cf1-8d0d-f826435d6cb6@app.fastmail.com>
In-Reply-To: 
 <CAApHDvrp6JwOVH3VwMB=YL3p97F5_0hKJc7J8zHRKuyhKru1dA@mail.gmail.com>
References: 
 <CAApHDvq=eEdw2Qp+aSzSOtTSe+h0fnVQ55CcTNqBkLDYiRZmxw@mail.gmail.com>
 <278B9FE3-F349-4494-99C5-483105C1C999@gmail.com>
 <CAApHDvoALVPgVSCuGdAfNUbxx_NacnAd0nuj8Hi=_-61AEqzaQ@mail.gmail.com>
 <1900289.1776212948@sss.pgh.pa.us>
 <CAApHDvq4_WDfTaWX6nHU8TV+no+aD1zjLphza685Ek7=xgGzew@mail.gmail.com>
 <1913188.1776220247@sss.pgh.pa.us>
 <CAApHDvp=fhr8ksuGyXkwYPXQpYvJ38YMhHvx6JmdD39cAvNRhw@mail.gmail.com>
 <CAApHDvrp6JwOVH3VwMB=YL3p97F5_0hKJc7J8zHRKuyhKru1dA@mail.gmail.com>
Subject: Re: Add bms_offset_members() function for bitshifting Bitmapsets
Content-Type: text/plain
Content-Transfer-Encoding: 7bit
Archived-At: 
 <https://www.postgresql.org/message-id/77f5109f-6d39-4cf1-8d0d-f826435d6cb6%40app.fastmail.com>
Precedence: bulk


On Sat, Apr 18, 2026, at 3:49 AM, David Rowley wrote:
> On Wed, 15 Apr 2026 at 14:33, David Rowley <dgrowleyml@gmail.com> wrote:
>>
>> On Wed, 15 Apr 2026 at 14:30, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> >
>> > David Rowley <dgrowleyml@gmail.com> writes:
>> > > I'd not considered surprise-prone as an aspect. I understand we have
>> > > bms_join and bms_union, which do the same thing if you only care about
>> > > the value of the result and not what happens to the inputs.
>> >
>> > Sure, but bms_join is an optional optimization of the far safer
>> > bms_union operation.  It bothers me to create the optimized case
>> > but not the base case.
>>
>> Hmm, yeah. That seems like a good argument for making a new set. I'll
>> go make it so.
>
> Patch attached for the version that creates a new set rather than
> modifying the input set in-place.
>
> David

Hey David,

> Attachments:
> * v2-0001-Introduce-bms_offset_members-function.patch

I applied, tested, and reviewed these changes.  Thanks for doing this, only a few small things jumped out.

nit: in bitmapset.c there is a new line added above bms_add_range()

+ * Arguments:
+ *	arg1: optional random seed, or < 0 to use a random seed.
+ *  arg2: the number of operations to perform.
+ *  arg3: the maximum bitmapset member number to use in the random set.
+ *  arg4: the minimum bitmapset member number to use in the random set.

nit: whitespace ahead of arg1, also should be "NULL" not "< 0"

in test_bitmapset.sql

+-- perform some random test on bms_offset_members()

nit: "tests"

Also, I think the random testing will likely cover these, but here are a few more explicit tests for odd corner cases.

-- shift that shrinks nwords
SELECT test_bms_offset_members('(b 64 65 66)', -64);  -- drops into word 0

-- shift that drops some low members and keeps others
SELECT test_bms_offset_members('(b 0 1 2 10)', -2);   -- expect (b 0 8)

-- entire set shifts below zero -> empty
SELECT test_bms_offset_members('(b 1 2 3)', -10);     -- expect empty

-- word-aligned positive and negative shifts
SELECT test_bms_offset_members('(b 1 2 3)', 64);
SELECT test_bms_offset_members('(b 65 66 67)', -64);

-- INT_MIN boundary
SELECT test_bms_offset_members('(b 1)', -2147483648);


I like the functionality and the reduction of repeated code that you've identified and fixed.

best.

-greg