Inconsistent handling of python3-module vs python3x-module packages

Adam Dinwoodie
Sun Feb 20 16:29:57 GMT 2022

On Thu, Feb 17, 2022 at 03:42:56PM +0000, Jon Turney wrote:
> On 16/02/2022 11:45, marco atzeri wrote:
> > On Wed, Feb 16, 2022 at 12:11 PM Adam Dinwoodie  wrote:
> > > 
> > > While wrangling a bunch of Python packages for my Cygwin installation,
> > > I've noticed an inconsistency about how python3 vs python3x packages are
> > > installed.
> > 
> > Only one ? ;-)
> > 
> > > As an example: the python3 package itself describes itself as a
> > > meta-package; the package itself is almost empty, and the key thing
> > > selecting the package does is depend on the latest python3x package,
> > > currently python39.  The same behaviour is in place with, for example,
> > > python3-tkinter.
> > 
> > Yes, these are real meta packages that pull the latest version
> > 
> > > However the python3-pytest package is marked as obsolete, and is
> > > obsoleted by python36-pytest.  If I select python3-pytest for
> > > installation, the package that's actually installed is python36-pytest,
> > > even though there's a python39-pytest package available.
> > > 
> > > This inconsistency means that someone naively installing python3-tkinter
> > > and python3-pytest will end up with both python3.9 and python3.6
> > > installed, with neither installation having access to both the pytest
> > > and tkinter modules.
> > 
> > This is an artifact of of cygport creating python3-xxxx packages
> > also for packages that never had a real package called like that.
> > We had  a discussion if it was worth to make the  python3-xxxx pulling the
> >   python39-xxxx instead of python39-xxxx and it was decided against it.
> I think this is more the emergent behaviour due to various changes than the
> result of a considered decision.

My assumption here was that this was emergent behaviour, which is
clearly not to say there wasn't a discussion at the time about whether
it was worth finding a better solution.  My hope here is to think about
what would be necessary to improve things, not least because this is
causing me (admittedly very minor) pain now.

> > > I think this inconsistency is liable to cause confusion; it certainly
> > > confused me until I worked out what was going on.  In my ideal world,
> > > we'd be in a situation where I could specify `python3-foo` to an install
> > > script and it'd automatically pick up the latest Python version
> > > available; I could specify `python3x-foo` if I wanted a specific older
> > > release.  But at the very least, I'd really like to see these packages
> > > being handled consistently one way or another.
> > 
> > I expect no one looking for a package will look for obsolete packages.
> It makes 'install python3-foo' a moving target for scripts.

Exactly this.  By default, I'd set up packages I maintain that require
Python packages -- for both build-time and run-time dependencies -- to
look for the latest and greatest Python interpreter and Pyhton modules,
and only pin to a specific Python version where there's a clear reason
to do so.  At the moment I can do that with the python3 package, but not
with python3-pytest.

Most users are probably not looking at obsolete packages, but if you're
trying to automate things by calling setup with the -P option, whether
the package you're looking at is marked as obsolete or not doesn't
actually get considered.  Running `setup.exe -P python3,python3-pytest`
seems reasonable, but will hit the problems I've described above.

> Patches to cygport to make this work better welcome!

I _think_ most -- if not all -- the cygport infrastructure is already in
place: cygport clearly already supports both virtual packages and
"provides" listings, which are the only bits of this that I think are
actually necessary.  The difficult part is just either (a) updating the
cygport files for all the packages that would benefit, or (b) finding a
way to have cygport work out that this is something it should do

I'll have a look at some of the existing cygport files, and cygport
itself, and see if I can work out what it would take...

More information about the Cygwin mailing list