[SciPy-user] overwriting Numeric on Windows is M$ policy :-)

[Gerard Vermeulen]

Hoping to do an easy install for a student, I tried to install
SciPy on Windows and to port some of my NumPy extensions to Windows.
(I am a Linux user, with very little Windows experience). 

The SciPy windows installer policy of overwriting a few modules,
got me into trouble. This has the side effect that the Numerical
header files (and C-API numbers) are out of sync with the binary modules.
Consequently, my home cooked extensions started to return NumPy arrays of
Complex instead of Float :-)

>From your web site, I got the impression that SciPy for Windows has
been built with Numeric-21.0, but this information looks dated.
So, I have tried several versions (CVS included), but somehow the SciPy
installer refused to overwrite the necessary modules (maybe because
they were newer, or because my registry database is broken -- I am
running Windows 98 with Win4Lin under Linux).

IMHO, it is much more friendly to include a full NumPy (even if it is
not the latest), instead of only a few binary modules.
You are locking authors of NumPy based extension modules out of the
'market' ( I have written a Qt/Qwt based plotting package, see
http://gerard.vermeulen.free.fr ).

MicroSoft has been sued for a similar policy :-)

Regards -- Gerard Vermeulen

PS: your recipy for rebuilding SciPy on Windows looks pretty old.
Did anybody try MinGW-2.O.O? And is MSYS-1.0.7 sufficient to
build ATLAS?

PPS: I am not rich enough to sue you.