Horton

Re: Incorrect result of bitmap heap scan.

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andres Freund <andres@anarazel.de>
Cc: Matthias van de Meent <boekewurm+postgres@gmail.com>, vignesh C <vignesh21@gmail.com>, Peter Geoghegan <pg@bowt.ie>, Andrew Dunstan <andrew@dunslane.net>, Melanie Plageman <melanieplageman@gmail.com>, Thomas Munro <thomas.munro@gmail.com>, Konstantin Knizhnik <knizhnik@garret.ru>, PostgreSQL Hackers <pgsql-hackers@postgresql.org>, Alexander Kuzmenkov <akuzmenkov@timescale.com>
Date: 2025-04-02T18:17:01Z
Lists: pgsql-hackers 
Andres Freund <andres@anarazel.de> writes:
> Does anybody have an opinion about how non-invasive to be in the
> back-branches? The minimal version is something like this diff:

Minimal is good -- less chance of breaking anything.

> - Should we commit the test showing that the naive implementation of
>   index-only-bitmap-heapscan is broken, in case somebody wants to re-introduce
>   it?

Seems like a good idea.  Agreed on HEAD-only for that.

> - We have some superfluous includes in nodeBitmapHeapscan.c - but I think
>   that's not actually the fault of this patch. Seems the read-stream patch
>   should have removed the at least the includes of visibilitymap.h, bufmgr.h
>   and spccache.h?  And b09ff53667f math.h...

Meh, let's leave that for the next round of IWYU hacking.

			regards, tom lane

Commits

  1. Add test for HeapBitmapScan's broken skip_fetch optimization

  2. Remove HeapBitmapScan's skip_fetch optimization

  3. Allow bitmap scans to operate as index-only scans when possible.