Open Stack

On 09/25/2013 12:20 AM, Monty Taylor wrote:
> 
> 
> On 09/24/2013 12:11 PM, Thomas Goirand wrote:
>> Hi Monty,
>>
>> On 09/24/2013 09:44 AM, Monty Taylor wrote:
>>> Instead of:
>>>
>>> "python setup.py install"
>>>
>>> Run:
>>>
>>> "pip install ."
>>
>> No way that this happens on the packaging side. Buildd have no network
>> access (on purpose), and we must not do any network access when building.
>>
>> So I wonder what this post is trying to achieve.
> 
> This post has nothing to do with packaging. The OpenStack project does
> not produce packages.
> 
> This is informational for people who are already running setup.py
> install, suggesting that they use pip install . instead.
> 
> If, however, you are in a packaging situation and you are running
> setup.py install as part of your packaging scripts such as in your
> debian/rules file or your rpm spec file, continuing to use setup.py
> install should have no negative effects, as all of the requirements
> processing should be avoided due to the system packaging having taken
> care of it already, so the evil that is easy_install will not be invoked.
> 
> Further, all debian/rules and rpm spec files that are packaging
> openstack project should really add the SKIP_PIP_INSTALL env var. This
> will turn off the additional pip operations that pbr does - which are
> again pointless in a distro-packaging world.

Hi,

Thanks for this clarification. You got me scared!!! :)

BTW, as I wrote already, I'm not really a fan of adding
SKIP_PIP_INSTALL, because if I do, I might not see the errors due to
missing dependencies. If there's a missing dependency and pip tries to
install it, it will break the build process, which really is what I
want. Or is it that SKIP_PIP_INSTALL still does a check to see if
dependencies are there? If that's not the case, then I would find it
really nice if it was possible to have a mode in which there is a pip
check that does a *hard break* error, and just stops the build process
(no need to wait until dpkg-buildpackage sees that there's an egg-info
folder that shouldn't be there). Any thoughts on this?

Thomas