[Numpy-discussion] 1.4.0 installer fails on OSX 10.6.2
Tue Jan 5 17:01:38 CST 2010
Pierre GM wrote:
> Ah OK, my bad. Now, why should it be that different ? Why rely on a
> second Python to install numpy from a dmg?
OS-X has a way of hard coding paths, so a given installer is designed to
go in one place, and one place only.
The python.org python is the best one to support -- Apple has never
upgraded a python, has often shipped a broken version, and has provided
different versions with each OS-X version. If we support the python.org
python for OS-X 10.4, it can work for everyone with 10.4 - 10.6.
It's changing a bit with OS-X 10.6 -- for the first time, Apple at least
provided an up-to-date python that isn't broken. But it's not really up
to date anymore, anyway (2.6.1 when 2.6.4 is out. I know I've been
bitten by at least one bug that was fixed between 2.6.1 and 2.6.3).
This is a policy followed by other projects as well.
> If it's a matter of
> framework, couldn't we used the Python in /System and create a
> framework in ~/Library ? (I'm just asking out of curiosity...)
The Apple python is fine -- it just isn't the same one, installed in the
same place. If you want to build a binary installer for it, it's easy --
but it will only work on 10.6, and not with an updated Python.
As the 2.6 series is binary compatible, you can build a single installer
that will work with both -- Robin Dunn has done this for wxPython. The
way he's done it is to put wxPython itself into /usr/local, and then put
some *.pth trickery into both of the pythons: /System/... and /Library/...
It works fine, and I've suggested it on this list before, but I guess
folks think it's too much of a hack -- or just no one has taken the time
to do it.
If I get a vote of approval for the approach, I suppose I could do it,
I'm sure I could find Robin's scripts and hack them for numpy.
Christopher Barker, Ph.D.
Emergency Response Division
NOAA/NOS/OR&R (206) 526-6959 voice
7600 Sand Point Way NE (206) 526-6329 fax
Seattle, WA 98115 (206) 526-6317 main reception
More information about the NumPy-Discussion