Tres Seaver wrote:
Stefan H. Holek wrote:
On 19. Sep 2007, at 00:46, Jodok Batlogg wrote:

i'm wondering why i couldn't fine = 3.4.0b1dev_r75415 after 3.4.0 was released a few days ago.
probably someone in [philikon, ctheune, J1m, baijum] removed the egg?
we nailed the version to 3.4.0b1dev_r75415 (and i have still this egg in my cache), but it disappeared from the rest of the world. this should never happen for released eggs, they should be considered read-only imho.

Eggs must not disappear, ever. Now that we have decided on eggs we have to live with the implications. This is one of them.

- -1.

Another implication:  eggs which have '-r[0-9]+' in them should *never*
be released to any location where they might be found by anyone not
explicitly requesting them.  Specifically, this means the cheeseshop as
well as any public 'find-links' or 'index-url' location.

Nobody should ever "nail" a dependency to such an egg without taking
responsibility for hosting it themselves.  Frankly, anybody who *wants*
such an egg should instead be using a checkput + ' develop',
instead;  anything else is an invitation to disaster.

Exactly. I fully agree. 'dev' releases shouldn't have been made in the first place, let alone uploaded to the CheeseShop. That said, I do understand that they might've been an excuse during the egg bootstrapping process. Now that we have eggs all around and they're stabilizing, we should get rid of these 'dev' eggs now (as in "supersede" them). They're a sign of bad (or non-existent) release management. They also usually coincide with bad (or non-existent) release metadata.

I also have the feeling (I can't really point my finger to it) that the 'find-links' mechanism isn't very reliable. At least I know for a fact that it repeatably fails for some people (e.g. when creating a buildout inside a workingenv, and on some Ubuntu systems). I find flat download directories a la a constant pain for human discovery, and they start being a real pain when you have multiple of them *and* are using PyPI. I'd much rather mirror PyPI for backup than maintain this flat list directory.

Our primary goal for the coming weeks should be to release stable eggs wherever possible. This release should happen to PyPI (which requires good release metadata, of course). Having these eggs will also be the requirement for getting Zope 3.4 final out the door. People have already helped: We could use some more help *wink*.

I sure hope that we can ditch the find-links option in our buildout.cfg soon to eliminate a possible source for egg confusion.

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

Reply via email to