[arch-general] [python-scipy] in community-testing WAS scipy issue

Angel Velásquez angvp at archlinux.com.ve
Tue Aug 10 14:37:02 EDT 2010


I've uploaded a new release of python-scipy with this change applied
(read above Fabrizio's e-mail).

I did some tests, but I don't know certainly if this change will break
other stuff...

I would like to have some feedback from SciPy users.


---------- Forwarded message ----------
From: Fabrizio Castellano <castella at phys.ethz.ch>
Date: Mon, Aug 9, 2010 at 8:01 PM
Subject: scipy issue
To: angvp at archlinux.com.ve

Hello Angel

I'm Fabrizio, a happy user of the python-scipy package on x86_64 platform,
anyway today I noticed a problem.

In scipy-0.8.0/scipy/special/cephes/mconf.h, line 111 there is:

#define UNK 1

this should set the arithmetic used by scipy.special to "unknown".
In fact in the file scipy-0.8.0/scipy/special/cephes/airy.c there is:

#ifdef UNK
#define MAXAIRY 25.77
#ifdef DEC
#define MAXAIRY 25.77
#ifdef IBMPC
#define MAXAIRY 103.892
#ifdef MIEEE
#define MAXAIRY 103.892

this limits evaluation of Airy functions to 25.77 for UNK arithmetic,
while leaving 103.892
for the IBMPC arithmetic (the one used in x86 processors I think).

I think they do this because they don't want to introduce
architecture-specific code, but
this is quite a limitation in the use of the airy function in scipy, and
probably other
things in scipy suffer from this UNK too.

I manually recompiled scipy changing the #define UNK into #define IBMPC
and it works.

I think the #define UNK should be changed to a #define IBMPC by default in
the arch repository.
This would make the arch package match better with the x86 architectures.
What do you think?

Fabrizio Castellano

Angel Velásquez
angvp @ irc.freenode.net
Arch Linux Developer / Trusted User
Linux Counter: #359909

More information about the arch-general mailing list