[Numpy-discussion] Sources more confusing in Python

Nathaniel Smith njs at pobox.com
Sun Apr 7 17:59:10 EDT 2013


On Sun, Apr 7, 2013 at 10:49 PM, Olivier Delalleau <shish at keba.be> wrote:
> 2013/4/7 <josef.pktd at gmail.com>
>>
>> On Sun, Apr 7, 2013 at 5:34 PM, Steve Waterbury <waterbug at pangalactic.us>
>> wrote:
>> > On 04/07/2013 05:30 PM, Nathaniel Smith wrote:
>> >> On Sun, Apr 7, 2013 at 10:25 PM, Steve Waterbury
>> >> <waterbug at pangalactic.us> wrote:
>> >>> On 04/07/2013 05:02 PM, Chris Barker - NOAA Federal wrote:
>> >>>> On Sun, Apr 7, 2013 at 8:06 AM, Daπid <davidmenhur at gmail.com> wrote:
>> >>>>> On 7 April 2013 16:53, Happyman <bahtiyor_zohidov at mail.ru> wrote:
>> >>>>
>> >>>>> $pip install numpy # to install package "numpy"
>> >>>>
>> >>>> as a warning, last I checked pip did not support binary installs  ...
>> >>>
>> >>> Guess you didn't check very recently ;) -- pip does indeed
>> >>> support binary installs.
>> >>
>> >> Binary install in this case means, downloading a pre-built package
>> >> containing .so/.dll files -- very useful if you don't have a working C
>> >> compiler environment on the system you're installing onto.
>> >
>> > Point taken -- just didn't want pip to be sold short.
>> > I'm one of those spoiled Linux people, obviously ... ;)
>>
>> However, pip is really awful on Windows.
>>
>> If you have a virtualenv and you use --upgrade, it wants to upgrade all
>> package dependencies (!), but it doesn't know how (with numpy and scipy).
>>
>> (easy_install was so much nicer.)
>>
>> Josef
>
>
> You can use --no-deps to prevent pip from trying to upgrade dependencies.

This is only a partial workaround, because this also means that if
there *are* new needed dependencies, they get ignored, resulting in a
possibly broken install. IIRC the full workaround is 'pip install
--no-deps --upgrade foo; pip install foo'

The other annoying workaround is to instead of using --upgrade, do
something like 'pip install numpy==1.7.1'. This requires knowing (or
looking up) what the latest version is, but once you've done that it
works.

-n



More information about the NumPy-Discussion mailing list