On Wed, Oct 7, 2026 at 1:31 PM Álvaro Herrera <[email protected]> wrote:
> On 2026-Oct-07, Baji Shaik wrote:
>
> Oh, it was? I don't remember that -- where is that?
That was commit a120ecf5498 ("Fix parsing of REPACK options"), which
made REPACK honor the last value when an option is repeated.
> I don't think last-wins is our norm, or is it?
>
Right, most commands that take a DefElem option list reject duplicates
via errorConflictingDefElem() (COPY, WAIT FOR, CREATE/ALTER ROLE and
SUBSCRIPTION, CREATE DATABASE, etc.).
my thought process was like these are maintenance-style commands
(VACUUM, ANALYZE, EXPLAIN, CHECKPOINT, REPACK) and are last-wins, so
I went in that direction thinking of them as one group that behaves the
same way.
So if the goal is to unify across the board, I'm happy to go with
rejection instead and make the maintenance commands reject duplicates too,
which matches the majority. I can redo the patch that way.
Thanks,
Baji Shaik.
