[SciPy-dev] Request to test system info hooks

eric eric at scipy.org
Sat Feb 23 13:39:10 CST 2002

> Could
>   c:\
> be considered as a prefix? Because I see that you are using

Making this switch (which you've already put in the recent CVS I see, solves the
problem for fftw.

>  atlas_library_dirs=['C:\\atlas\\WinNT_PIIISSE1']

atlas_info still doesn't pick up my atlas library.  because it looks like it
doesn't search within the WinNT_PIIISSE1 directory -- only the atlas directory
is searched.

To solve the problem, I've moved all my libraries up into the atlas directory.
Trying to come up with a scheme to calculate the WinNT_PIIISSE1 directory name
stuff for atlas is a little more effort than necessary right now-- but we will
need this info if we're ever to try and automate downloading of the correct
atlas libraries from SciPy.org, etc.

On windows, sticking all libraries in c:\atlas\ isn't that big of a deal because
they are almost always used as stand-alone machines.  On Unix machines where you
working on a shared file system, it might be more of an issue.  Someone might
have multiple atlas versions built for different architectures sitting in their
home directory.  Handling the ATLAS directory naming scheme for different
architectures/processors would be handy for this situation.  Still, I think it
is a second order issue.

> and if os.path.isdir('c:\\') would succeed, then system_info should find
> this atlas library.
> Is is reasonable to search c:\windows\system32 for libraries? Is that the
> place where win32 users install the third party libraries?

This is where most dll's are stored, but it isn't a normal place to search for
.lib, so no, I think it probably isn't a good idea.  The c:\\ you use is as good
of a convention as any.

> Also I am not familiar with Cygwin and Mingw issues? Can they be
> considered to have similar tree structure like in unices? I mean do they
> have directories like /usr/, /usr/local, etc?

Yes, I think so -- as long as the person using the machine mainly uses Cygwin
and stores all their libraries in a standard Cygwin location.  I think people
building with Cygwin will have probably follow this convention.

It looks like system_info works fine on windows after I moved my atlas files.
import is currently failing after I build due to unrelated issues.  As soon as
these are fixed, I'll run the test suite.  I'm betting it will work fine.


More information about the Scipy-dev mailing list