Rob Browning <
r...@defaultvalue.org> writes:
> Greg Troxel <
g...@lexort.com> writes:
>
>> There's a hardcoded list. pkgsrc's approach to python instability is to
>> expect everything that uses python to bind to a specific version, so
>> there's no 'python' and no 'python3', as both are unsound late binding.
>>
>> The fix is to add 3.14 and 3.13. I'm surprised to see 3.6 still.
>
> This script is intentionally very generous since it should and is
> intended to work with "most any python", and it starts with python3
> because
https://peps.python.org/pep-0394/ says that python3 should be
> expected to work.
>
> No worries if pkgsrc doesn't follow that PEP, but I suspect many systems
> do.
Sure - PEP 0394 is a difference of opinion :-) and I expect many do.
Having python3 present means things use it, and then when python3
changes where it points, you get UB in anything using it because the set
of installed modules changes compared to what was probed when it was
built. pkgsrc will install that symlink if you have "pkg_alternatives"
installed, but I don't on that box, and some people object to these
symlinks. (There are vast numbers of packages that need newer and that
can't use newer so we end up with multiple versions installed usually.
That's messy, but python compat isn't good enough to avoid it.)
As for old versions, I see the point but if bup has a documented
"python3 >= x" then I'd align this.
Are you amenable to adding 3.14 and 3.13 - should I send a diff or is it
just easier for you to do that?