Re: proposed python change: removing python 3.7

[Thomas Klausner <wiz%netbsd.org@localhost>]

On Thu, Apr 14, 2022 at 02:02:56PM +0000, nia wrote:
> On Thu, Apr 14, 2022 at 03:45:40PM +0200, Thomas Klausner wrote:
> > On Thu, Apr 14, 2022 at 01:44:00PM +0000, nia wrote:
> > > On Wed, Apr 13, 2022 at 10:14:15AM -0400, Peter Lai wrote:
> > > > Disagree. 3.7 does not EOL until JUN 2023. I recommend we follow
> > > > upstream support schedule.
> > > 
> > > I'm adding another vote to the "disagree" pile, for one reason
> > > alone, and that's that Python 3.7 was the default in pkgsrc
> > > for quite some time. I don't want to break people's scripts
> > > and so on before it's due.
> > 
> > Is that a "Disagree and I'll take care of fixing py37-* packages"?
> >  Thomas
> 
> I'll keep bulk builds working, but I was doing that already.
> No interest in patching code in order to keep up with update churn.

I'm not sure what this means, so let me give you an example.

mef's bulk build from April 1 had nearly no py* packages broken,
because I've been fixing breakage in py* for months now.

mef's bulk build from April 11 had some issues with missing
dependencies (py-setuptools_scm) in py-jsonschema, py-configupdater,
py-ini2toml, py-validate-pyproject affecting all python versions,
which I've fixed, but also the following new breakage because the main
py-numpy package does not support python37 any longer and python37 was
switched to an older version, which has a problem (which is why many
of the following packages are already disabled for python27):

http://www.ki.nu/pkgsrc/reports/current/NetBSD-9.0/20220412.1000/py37-boost-1.78.0/install.log
http://www.ki.nu/pkgsrc/reports/current/NetBSD-9.0/20220412.1000/py37-h5py-3.2.1nb1/build.log
http://www.ki.nu/pkgsrc/reports/current/NetBSD-9.0/20220412.1000/py37-gdal-3.4.2/install.log (not sure about this one, but only broken for python37)
http://www.ki.nu/pkgsrc/reports/current/NetBSD-9.0/20220412.1000/py37-OpenGL-accelerate-3.1.5nb2/build.log
http://www.ki.nu/pkgsrc/reports/current/NetBSD-9.0/20220412.1000/py37-matplotlib-3.4.3nb1/build.log
http://www.ki.nu/pkgsrc/reports/current/NetBSD-9.0/20220412.1000/py37-bottleneck-1.3.2nb2/build.log
http://www.ki.nu/pkgsrc/reports/current/NetBSD-9.0/20220412.1000/py37-numba-0.55.1/build.log
http://www.ki.nu/pkgsrc/reports/current/NetBSD-9.0/20220412.1000/py37-numexpr-2.8.1/build.log
http://www.ki.nu/pkgsrc/reports/current/NetBSD-9.0/20220412.1000/py37-pywavelets-1.0.3nb2/build.log
http://www.ki.nu/pkgsrc/reports/current/NetBSD-9.0/20220412.1000/py37-cftime-1.5.0nb1/build.log

In my eyes, there are two proper solutions here:

* find and fix the issue in the old py-numpy package so it works for
  python37 (and python27 if you're motivated)

* or mark the packages listed above, and the packages using them, as
  not-for-python37.

I've done the second for python27, but I'm just not motivated in doing
it again for python37.

This also means that (as long as neither happens) reading the bulk
build logs will be more work because it means I have to remember to
skip these packages each time, or waste time on them again to identify
the old breakage (or miss new breakage hiding behind this old one).

Looking forward to solutions,
 Thomas