[SciPy-dev] SciPy 0.4.4 release progress

[Robert Kern]

Ed Schofield wrote:

> 1. I uninstalled Python 2.4 and installed Python 2.3, figuring that  
> SciPy binaries built against Python 2.3 would be upwardly compatible,  
> but perhaps not vice versa.  Is this true?  I noticed you built  
> separate binaries for NumPy for Python 2.3 and 2.4 on Windows but  
> just one set on Linux -- why?

You need to compile separate versions for Python 2.3 and 2.4. Python 2.x
binaries are not compatible for different "x". Python 2.x.y binaries *are*
compatible for the same "x" but different "y".

> 2. I've been getting various LAPACK errors since I moved away my  
> custom-rolled ATLAS/LAPACK libraries, but things are looking better  
> now.  I presume that the Linux binaries should be built without an  
> ATLAS dependency?  If I link against /ed's/own/atlas/library.so, the  
> binaries will be useless to anyone but me, right?  ...

Pretty much. If you use static libraries, then they would be useful, but
generally Linux binaries aren't terribly useful if they aren't tailored for a
specific distribution.

> It makes me wonder if we need Linux binaries at all.  I've noticed  
> that numarray only distributes Windows binaries, probably since the  
> source tarball is generally far more useful, and because Linux  
> distributors will soon package SciPy themselves anyway.  Could we do  
> this too?

Probably.

> 3. Are there any Windows linking traps to be aware of?  I presume we  
> link SciPy against an ATLAS DLL that gets rolled up into the .exe  
> file ...

You should use the static libraries provided here:

http://www.scipy.org/download/atlasbinaries/

-- 
Robert Kern
robert.kern at gmail.com

"In the fields of hell where the grass grows high
 Are the graves of dreams allowed to die."
  -- Richard Harter