[SciPy-dev] old web site front page
Travis Oliphant
oliphant.travis at ieee.org
Sun Dec 25 01:22:02 EST 2005
Perry Greenfield wrote:
>On Dec 22, 2005, at 10:06 AM, Joe Harrington wrote:
>
>
>I'll second Joe on this issue. My sense is the ususal denizens of
>scipy, being comparatively sophisticated (despite proclamations to the
>contrary :-) in doing installations, are not aware of how an important
>issue this is. I have had many people come to me and say they tried
>scipy before and that they gave up trying to install it. When I tell
>them that numarray is being replaced with scipy the usual reaction is
>of great worry because of the installation issue (I try my best to
>convince them that the core will be as easy to install and so forth).
>You won't hear from most of these people if they try and fail. They
>will just give up and associate bad things with the project. For that
>reason I think it is important not to oversell scipy_core to the
>general community until it has been beaten on for a while,
>installation-wise and shown to be a pretty easy install, and
>comparatively stable (seeing weekly updates will scare them quite a bit
>too).
>
>
We really need to separate scipy_core from full scipy. The lists are
full of "install" problems that are mostly "full scipy" problems or
problems related to confusion regarding the packages that exist. I
think their should be plenty of advertisment for scipy_core. The flux
in scipy_core right now is the naming scheme. This should settle down
in a matter of weeks.
I agree that *a lot* of patience is in order for full scipy. But
scipy_core installs should be settling down. I agree that new users
should be informed of the situation and allowed to choose their poison,
recognizing that if they are the first to try it out on a new platform
their may be issues. It is important that people believe scipy_core
can be installed without difficulty. I think we've made lots of strides
here, but no doubt will learn more as people try it out with strange
configurations.
As a data-point. I was recently on a totally new system with intel's
mkl BLAS and python setup.py install went fine after an SVN checkout and
linked against the mkl BLAS without difficulty (and I didn't need to
edit any configuration files either). So, I was pretty happy with
that. Thanks, Pearu....
Right now, it's just the sub-packages in basic (fft, linalg, and random)
that are being shuffled a bit so that the name-space collisions with
full scipy and the import magic will disappear. I'm *thrilled* to see
the web-page receive the overhaul and point people in the right direction.
-Travis
More information about the SciPy-Dev
mailing list