Should we call it quits on support for 32-bit

[Greg Troxel]

"Regina Obe" <lr at pcorp.us> writes:

> I feel a little uneasy about declaring 32-bit systems dead, but
> apt.postgresql.org (and I assume Debian upstream), is about to declare no
> longer supporting 32-bit systems.

There's a difference between the source code not being buildable on an
ILP32 machine, requiring LP64, and Debian packaging.

> How do people feel about this.  On the one hand I feel like some of those
> 32-bit failures deserve further investigation.

I think giving up on ILP32 CPUs is crazy.  postgresql and postgis are
not super small, but for small workloads should be able to run fine in
even 1G.

> Poor berrie has been crashing on garden tests ST_MakePolygon(Nan ...) for a
> while, I'm sure it's fixable, and it's our fault and we should do something
> about it, but berrie64 is doing fine on the same tests.

Is the machine crashing, or the postgis processes?

> And this - https://trac.osgeo.org/postgis/ticket/5461  
>
> And I can't remember the last time I saw 32-bit gitlab happy -
> https://gitlab.com/postgis/postgis/-/jobs/7550985291
>
> Part of me worries, are these true errors we are only seeing because of the
> smaller bit space or have we just gone beyond 32-bit that we just shouldn't
> care.

Or because we have types wrong and it only works if different types
happen to match.


My general view is that testing on different systems and CPUs is very
valuable because it finds real bugs.