Hi, On Wed, Aug 19, 2026 at 05:51:07PM +0900, Fujii Masao wrote: > On Wed, Aug 19, 2026 at 5:38 PM Daniel Gustafsson <[email protected]> wrote: > > That's another option. I'm not a big fan of adapting new code to bugs of > > the > > past, but at this point in the cycle there are other aspects in play as > > well. > > I'll see what others think for a little bit before reverting. > > +1 > > If I'm in the minority, i.e., the only one who doesn't like adding > data_checksum_version_init and is concerned about the consistency > between pg_control_xxx() and pg_controldata, I'm happy to withdraw my > proposal. >
dc7d70ea05d introduced pg_control_init() and documented from the start as returning cluster initialization state. Later ed308d78379 made the value changing through pg_checksums, without adjusting pg_control_init() or its documentation. So, yeah, this looks like a regression. So I think there is 2 kinds of users: - Users expecting initialization state, as documented. - Users treating it as current state. That contradicts the documented purpose. I'm not sure we should "favor" the latter over the former, so FWIW, I'd vote for keeping the committed fix. Regards, -- Bertrand Drouvot PostgreSQL Contributors Team RDS Open Source Databases Amazon Web Services: https://aws.amazon.com
