[Numpy-discussion] numpy via easy_install on windows

[David Cournapeau]

On Sat, Mar 14, 2009 at 3:15 AM, Jon Wright <wright at esrf.fr> wrote:

> What I want is a simpler way to install things for people to try out our
> programs. We currently have dependencies on at least numpy, matplotlib,
> PIL, Pmw and PyOpenGl and having to go through a series of 6 different
> installations can be a bit intimidating. Any suggestions as to how best
> to distribute such a beast is most welcome.

When distributing things, I see only two solutions: either you
distribute everything separately (ala linux), or you integrate
everything. On windows, integrating is almost always the right
solution: you get the uninstall option, etc...

It depends on how much resource you can spend on it, but if I were to
distribute things on windows, I would build a msi/bdist_wininst of
every package, and wrap this into another installer (which is almost
exactly what the superpack does). That's how every big software on
windows work AFAIK: every MS software installs this way for example. I
don't claim any deep knowledge on the things behind bdist_wininst, but
nsis, which is the open source system I use to build numpy and scipy
so called superpack is powerful, maintained and well documented.
Wrapping all the installer in one would be easy - if you need option,
and in particular to control each installer independently, then it
would become more difficult.

cheers,

David