Danek:

> On Sat, Nov 29, 2008 at 09:37:53AM -0800, Brian Cameron wrote:
> 
>> So, I think we should safely recommend that modules which ship egg files
>> should not remove them.
> 
> Gosh, they're a bit forceful.

I have been continuing the discussion on distutils-sig, so if you want to
read the thread I'm reposing the link.

http://mail.python.org/pipermail/distutils-sig/2008-November/010556.html

In summary, these files are useful if an end-user tries to build other
modules via setuptools for checking what dependencies are already
installed.  If you remove the files, and a user tries to run ez-install,
it can get confused and try to re-install packages already on the system
if the egg-file metadata is not present.

>> So, Danek, does this address your concerns?  If so, I think I will go
>> ahead and mark the case approved since this is the only outstanding
>> issue.
> 
> It does, but it'd be really useful, now that egg files are going to be
> expected for (at least some) Python projects to have some Solaris-specific
> rules surrounding their inclusion.  That probably would amount to "include
> them, if provided", but also an answer to whether Sun-written Python
> projects should be done as eggs as well as a good example of what the
> actual egg interfaces are -- what of this stuff should people list in their
> interface tables if they're delivering eggs?  The files / pathnames
> themselves are almost certainly not the right answer, since they're not
> consumed directly, but whatever metadata is.  Running a (quick) full case
> with an opinion to set precedent would have been the old way of doing this,
> but perhaps a fast-track would be sufficient these days?

I agree, in the above thread, I posed some questions to get a better idea
of how we should manage egg files.  I think this will require some
discussion with the upstream community, and we might find out that we
need to do some extra work to ensure we are building our Python modules in
the best way to work with setuptools.

I propose that we close this case as approved, and resolve the setuptools
issues separately.  Once I get a full understanding of what the rules should
be, I could submit a separate ARC case to define the interfaces and rules.

Would that be acceptable?

Brian

Reply via email to