Horton

Re: Making empty Bitmapsets always be NULL

Jacob Champion <jchampion@timescale.com>

View thread View on postgr.es
From: Jacob Champion <jchampion@timescale.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: pgsql-hackers@lists.postgresql.org, Soumyadeep Chakraborty <soumyadeep2007@gmail.com>
Date: 2023-03-01T21:35:08Z
Lists: pgsql-hackers 
On Tue, Feb 28, 2023 at 1:59 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> I also discovered that nodeAppend.c is relying on bms_del_members
> not reducing a non-empty set to NULL, because it uses the nullness
> of appendstate->as_valid_subplans as a state boolean.

I seem to recall that Deep and I tripped into this during the zedstore
column projection work. I think we started out using NULL as a
sentinel value for our bitmaps, too, and it looked like it worked,
until it didn't... so +1 to the simplification.

--Jacob

Commits

  1. Remove trailing zero words from Bitmapsets

  2. Remove local optimizations of empty Bitmapsets into null pointers.

  3. Require empty Bitmapsets to be represented as NULL.

  4. Mop up some undue familiarity with the innards of Bitmapsets.

  5. Remove bms_first_member().