On Sun, Jul 12, 2026 at 5:34 AM Jelte Fennema-Nio <[email protected]> wrote: > On Wed, 8 Jul 2026 at 19:30, Jacob Champion > <[email protected]> wrote: > > If we did that, Python 3.9 looks within reach for PG20, which seems > > reasonable considering it'll already have been dead for two years. (It > > looks like LTS support for a spectrum of Pythons is now the norm > > rather than the exception, and we should start making decisions > > accordingly.) That'd only be a roadbump for our SUSE buildfarm owners, > > who'd need to install an additional package beside the base Python 3.6 > > in order to run Python scripts. > > You seem to be walking back on your previous email here by applying > that logic only to SUSE/SLES. If we applied that logic to all > operating systems, we'd end up on Python 3.11 (based on your list).
I switched which type of dependency I was talking about between the two emails you quoted, which was confusing, sorry. My argument is that we should work to split our "Python minimum version" into three separate things: - runtime dependency: what actually gets linked against and tested, e.g. with PL/Python. We should be conservative here. - build-time dependency: what our build scripts (and Meson) can be written against. For most platforms this should be the same as the runtime dependency to simplify our mental models, but Peter's recent work allows them to be different. - test-time dependency: what our tests can be written against. We can potentially be the most aggressive with this, as long as we don't leave our buildfarm volunteers to twist in the wind. I think the runtime dependency for "supported LTS platforms" should generally be the base Python version that is shipped, so that our packagers aren't splitting their ecosystems by default. We don't want our supported platforms to have to navigate package dependency hell for core Postgres. (If packagers actually think it'd be fine for a specific platform, they can tell us and then we can bump those up. But I don't think we should assume that's the case.) So in the email you replied to, I'm saying that 3.9 looks like a reasonable floor for the *build-time* dependency to me, with the caveat that SLES testers and packagers will have a 3.6 runtime version and a 3.9 build-time version that we will need to navigate. If that's not acceptable for some reason -- it's been a while since I've had to author RPM specs, but I thought I remembered that they allow for pretty good dependency separation? -- then IMO the floor would need to drop to 3.6. Copied Devrim since he'd probably have opinions re: SLES. Thanks, --Jacob
