Re: Bump default Python version to 3.14?

[Greg Troxel <gdt%lexort.com@localhost>]

Benny Siegert <bsiegert%netbsd.org@localhost> writes:

> I wanted to ask if there are any technical reasons why Python 3.13 is
> still the default version. 3.14 has some nice performance improvements
> AIUI.

The reason to not bump has almost always been that too many packages
don't work with the proposed new version.  While individual packages are
built with multiple versions, it gets messy if anything with many
depending packages fails to build with the default version.

> Who is usually doing the Python version bumps?

I am not sure there is a usual.


I just looked over the 2026Q3 build, for 11.0 and x86_64, taking the
list of py313 packages, seding 313 to 314 (calling that expected), and
"diff -U0"ing that with the list of 314 packages.

--- PYTHON.census.py314-expected        2026-10-02 12:04:33.857384276 +0000
+++ PYTHON.census.py314 2026-10-02 12:04:34.664028729 +0000
@@ -162 +161,0 @@
-py314-backports.zstd
@@ -1142,0 +1142 @@
+py314-pikepdf
@@ -1652 +1651,0 @@
-py314-torch

So

  pikepdf is built for 314 and not 313 (not a problem)

  backports.zstd is missing; the package is marked that it's included in
  314 (not a problem)

  torch is missing (problem)

torch is marked incompatible with 314.  That doesn't make sense in the
larger ecosystem scheme of things -- surely upstream works with all
released python versions.   Sure enough, our torch is 2.8.0 and upstream
is at 2.14.1.  2.8.0 is from 2025-08-06, so we're not that out of date
in time.

There aren't other python packages that depend on torch, since they'd be
missing too, and they aren't.

I'm therefore in favor of

  change default to 314 right now

  someone(tm) should fix/upgrade torch, async with the change, which is
  true whether or not we change the default