Re: proposed python change: removing python 3.7

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

On Wed, Apr 13, 2022 at 10:55:10AM -0400, Greg Troxel wrote:
> 
> Thomas Klausner <wiz%NetBSD.org@localhost> writes:
> 
> > Python 3.7 is on its way out, modules are starting to desupport it.
> > The 3.x versions are generally quite easy to switch between, and I
> > don't think the effort in supporting Python 3.7 is well-spent.
> 
> As always a plea for clarity on removal requests.
> 
> (I'm going to assume 2.7 doesn't exist for this discussion.)
> 
> First, "on its way out" is vague, and Peter Lai has said that the EOL
> date is June of 2023.

Yes. I meant that some important python modules do not work with
python 3.7 any longer.  So while 3.7 is not officially EOL, module
maintainers start treating it that way.

> Now, pkgsrc is not an LTS, and the only reason we have multiple python
> versions in the first place is that python upstream and the community of
> package upstreams that are written in python are collectively not good
> at compatibility: In an ideal world, we'd just update the single python
> version when there was a new stable branch released.  But each new
> version won't run some python code, and lots of python code needs a
> newish version, so here we are.  That's a long way of saying: just
> because 3.7 isn't formally EOL, we don't have to keep it.
> 
> Now, the questions are:
> 
>   Are there any packages in pkgsrc that only work on 3.7?

No.

There are two packages in wip that are marked this way (wip/py-ana,
wip/glide) but on first glance these look incorrect and probably just
wanted to forbid python 2.7.

>   Are there any python programs people run, not in pkgsrc, that only
>   work on 3.7?

I'm not aware of any.

>   Do we think any actual people would have problems with 3.7 removal?
>   Surely some might still be using 3.7, even though they are very
>   overdue for updating, and that doesn't IMHO really count.  So it's
>   more about the first two questions.
> 
>   What's the work in keeping 3.7?  It seems it's
>     - updating to new micros
>     - having to add PYTHON_VERSIONS_INCOMPATIBLE=37 to various things
>     - worse, having to mark things not-37 if they depend on something
>       that doesn't do 37, to avoid breaking esp. pbulk scan (which is
>       IMHO a pbulk bug, but that's how it seems to be)

Yes, this is the point which made me suggest the removal.

We currently have to maintain the recursive limitations of
PYTHON_VERSIONS_INCOMPATIBLE manually. This is a lot of effort and
regularly breaks pbulk builds, and I don't think it's worth our time.

> It seems to me that there are no good reasons to run 3.7, and that 3.7
> is crufty enough that a non-trivial amount of things need to be marked
> not-for-37.  So that argues for dropping it now.

Thanks for your comments.

I'll wait a bit longer for other feedback on the proposal, but I still
intend on removing python37 before the next branch.
 Thomas