Re: Add bms_offset_members() function for bitshifting Bitmapsets

On Thu, 9 Jul 2026 at 11:55, David Rowley <dgrowleyml(at)gmail(dot)com> wrote:
> I pushed a change to make seed int64 instead. It's still possible to
> specify the full range of possible values that GetCurrentTimestamp()
> can (or could ever) return with the int64 type. This way, there's no
> longer a mismatch in the format specifier for elog and the type we're
> formatting. That's an improvement, so thanks.

I was just in this area again and noticed that the nmembers
calculation does not take into account the min_value parameter. The
bad code is:

nmembers = pg_prng_uint32(&state) % max_range + min_value;

That should really be subtracting the min_value, as I had intended
nmembers to do was populate a random set of members in the range
specified by the function. While there, I noticed that the parameter
names are not very consistently named and having the max_value before
the min_value seems weird too. I understand test_random_operations()
has it that way. I didn't adjust that function. Maybe that could be
done separately since it's also in v19, whereas
test_random_offset_operations() is new to v20.

I've attached a patch which fixes this plus a few more things that
Claude code pointed out to me when I asked it to review my changes.
Those are primarily around overflow risk if max_value was given as
INT32_MAX.

David