Jodok Batlogg wrote:
the current practice (at least here at lovely systems, and presumably a lot of other developers) is to add (or a mirror of it) to your find-links. recently some people started registering packages (and uploading eggs ) to cheeseshop. this makes totally sense for zc.buildout, lovely.buildouthttp,... but i'm not sure about zope packages.

We're certainly uploading a lot of Zope3's own eggs to PyPI *IF*

- they have a decent description and long_description
- they have decent metadata otherwise (Trove classifiers, changelog)
- they're actual stable releases

When packages fulfil these criteria, then I think it's very good to have them on PyPI: they're a great way to make them discoverable for other developers, Zope and non-Zope. For example, I can browse the 'Zope3' framework category and find a lot of Zope3-related software now. I wish more software were aptly classified with this Trove classifier so I could find more (all!) Zope3 software tehre.

because of this mix you might end up in getting the wrong egg. or not finding a egg you downloaded a few days ago. in case pypi is down you're totally stuck.

You're stuck either way, even if you'd be using The key is to mirror the PyPI simple index. That's what ppix (or actually it's success zc.mirrorpypislashsimple) is all about.

This is a non-issue I think.

well, i understand that the stability of pypi is much better lately and the simple interface, the and ppix mirrors make the index lookup much faster as well.
but for me there are still two remaining issues:
- the eggs are hosted on cheesehop as well, it's not easy (commandline) to host the egg externally or to specify a mirror to use when pulling the eggs. that's not acceptable for production deployments.

I think it's really easy. Just use Jim's mirror software and point buildout to that different index. That's one line in buildout.cfg.

Again, seems like a non-issue.

- by default " sdist bdist_egg register upload" hides the old releases, which is not acceptable as we nail all versions for deployments. if someone releases a new version, the older ones disappear and buildout (with nailed versions) stops and complains not finding the egg.

It doesnt' disappear from the simple index. It just disappears from the "human" index, which I think is a good idea. It doesn't make sense to show people a billion releases. setuptools and zc.buildout (which both nowadays use the simple index) will obviously need to know about the odl releases.

This is a non-issue :).

before we don't have a solution for the two points above i have objections to register the packages on pypi and prefer to add to my find-links.
if there is an easy solution for it, enlighten me :)

Human discoverability is *very* important. If you have doubts about depending on PyPI, fine. I won't mind a backup location (though I hate It has piled up so many crappy eggs by now with no way of deleting them...).

But please don't take the human discoverability away from it all. I thikn that's what PyPI is great for. Provided you actually make it worthwhile and have meaningful package metadata in

Bis morgen :)

-- -- Professional Zope documentation and training
Zope3-dev mailing list

Reply via email to