On Fri, Jan 29, 2021 at 6:33 PM Michael Paquier <michael@paquier.xyz> wrote:
> Those 32-bit modules are still being sold actively by the RPI
> foundation, and used as cheap machines for education purposes, so I
> think that it is still useful for Postgres to have active buildfarm
> members for 32-bit architectures.
But I'm not arguing against that. I'm merely arguing that it is okay
to regress 32-bit platforms (within reason) in order to make them more
like 64-bit platforms. This makes them less prone to subtle
portability bugs that the regression tests won't catch, so even 32-bit
Postgres may well come out ahead, in a certain sense.
--
Peter Geoghegan
