[Catalog-sig] API search by python version (or classifier)

Alexis Métaireau alexis at notmyidea.org
Thu Jan 27 02:35:43 CET 2011

Hi Tres,

Le 26/01/2011 23:23, Tres Seaver a écrit :
> I think the SIG's response in September was a good example of "the
> perfect is the ene good":
> - - A miniscule fraction of all PyPI releases actually declare PEP 345
>   metadata.  I see no evidence that the number is growing[1].

I do agree. By now, the PEP 345 is not adopted. Probably that's mainly
because no installer currently manages those PEP 345 metadata format.
We're working on it with distutils2.

> - - A very significant number of the packages on PyPI declare
>   dependencies via setuptools'requires.txt';  "practicality beats
>   purity."

It's the better way to provide metadata by now, IMO.

> - - Of the packages which were made with setuptools metadata, the number
>   for which one distribution of a given release has different
>   dependencies than another of the same release is likely insignificant.

The fact is that for those (tiny) proportion of projects that relies on
this feature, the metadata provided by PyPI will be incorrect. I'm not
sure it's acceptable. Also, there is no way to know that the metadata
are incorrect, or partially correct.

I'm not sure the best thing to do is to provide those wrong metadata.
Probably this could ease the way to get distribution dependencies, but
it's surely *not* something we can rely on for packaging purposes.

> Exposing even imperfect dependency data while we await PEP 345 adoption
> would have been a useful, pragmatic thing to do.

We can go for something like that, but what I'm affraid of is that we
propose a wrong source of informations, whereas we are working on a way
to provide the same informations the right way. I do think the tradeoff
to wait a bit more is right. Afterall, this feature is not vital, am I
right ?


More information about the Catalog-SIG mailing list