[Distutils] Need for respect (was: PEP 438, pip and --allow-external)

[Nick Coghlan]

On 15 May 2014 22:05, "Stefan Krah" <stefan-usenet at bytereef.org> wrote:
>
> Nick Coghlan <ncoghlan at gmail.com> wrote:
> > > > I understand you think that is the purpose of PyPI, but I'm trying
to
> > > > tell you that the people that work on PyPI and pip do not share this
> > > > opinion, and as such it can be considered incorrect.
> > >
> > > If only the opinions of the persons working on PyPI and pip matter,
the
> > > logical consequence would be to remove ensurepip from the tree.
> >
> > Stefan, I realise you weren't able to attend the language summit in
2013, but
> > this delegation of authority to distutils-sig is exactly what the PEP 1
process
> > changes and the assignment of myself and Richard Jones as BDFL
delegates that
> > came out of that event was about. python-dev are not the experts on
language
> > level distribution issues for Python - that role belongs to
distutils-sig.
> >
> > While the opinions of core developers do matter, we're also far from
being
> > representative of the wider Python community
>
> It's not only about core developers. The main point is that it's very
hard to
> determine any general opinion of Python users. In fact, already in this
thread
> we have four people stating that their expectations differ from the
"official"
> ones.

Yes, but that's no different from the way python-dev itself works: we're
lacking sufficient objective data (or, more accurately, lacking the time,
inclination and resources to collect and analyse that data), so we instead
trust the collective experience of a group of specialists and a final
decision maker that can arbitrate when there are multiple reasonable
options and a specific decision is needed. In the case of the core language
and standard library, that's python-dev and Guido (or a PEP specific
delegate), in the case of the language level packaging ecosystem, it's
distutils-sig and currently either Richard Jones (for PyPI changes) or me
(for the metadata and packaging format standards).

The recent thread on python-dev triggered a full review and analysis of why
adoption levels for the link spidering system are so low (especially the
style that pip can actually verify properly), why the error messages are so
confusing when it breaks, and what can be done to provide a better user
experience for both publication and installation of Python packages using
the upstream tools.

It turns out the link spidering system is not only overly complicated and
relatively hard to both understand and implement, but also largely
redundant given the support for multiple indexes in the installation tools.
Since the multiple index support is the more powerful and flexible of the
two systems, while also being simpler to implement and easier to
understand, PEP 470 now proposes to standardise on that system for PyPI's
external hosting support, with a few additional enhancements to address the
discoverability issues that would otherwise arise. The link spidering
system will eventually go away and we'll then be left with a fairly
conventional "multiple repository" model, plus a few nice repository
discovery features that most other such systems don't have.

Cheers,
Nick.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/distutils-sig/attachments/20140515/0aa976ee/attachment.html>