Sage 4.5.2 - sage/numerical/test.py failure on fulvia (32-bit Solaris x86)

[Dr. David Kirkby]

John built a 32-bit Sage 4.5.2.rc1 the other day on fulvia and posted his
results from ptestlong.log here

http://sage.math.washington.edu/home/palmieri/misc/fulvia-ptestlong.log

I've just built 4.5.2 and get the same failures, but with one additional one.

See my log at

http://boxen.math.washington.edu/home/kirkby/fulvia/32-bit/sage-4.5.2/ptestlong.log

The additional failure is

sage -t -long devel/sage/sage/numerical/test.py # 1 doctests failed

which fails as:

sage -t -long devel/sage/sage/numerical/test.py
**********************************************************************
File
"/home/drkirkby/fulvia/32/sage-4.5.2/devel/sage-main/sage/numerical/test.py",
line 28:
sage: e[0]*float(501/pi)**2
Expected:
0.999............
Got:
0.0
**********************************************************************
1 items had failures:
1 of 21 in __main__.example_0
***Test Failed*** 1 failures.
For whitespace errors, see the file /home/drkirkby/.sage//tmp/.doctest_test.py
[14.0 s]

This is the actual test:

#Test arpack
#This matrix is the finite difference approximation to
# the eigenvalue problem
#d^2f/dx^2=\lambda f, on [0,\pi], which boundary values 0
# The lowest eigenvalue calulated should be close to 1
sage: import scipy
sage: from scipy.sparse.linalg.interface import aslinearoperator
sage: n=scipy.zeros((3,500))
sage: n[0,:]=-1
sage: n[1,:]=2
sage: n[2,:]=-1
sage: A=sparse.spdiags(n,[-1,0,1],int(500),int(500))
sage: A=aslinearoperator(A)
sage: e,v=arpack.speigs.ARPACK_eigs(A.matvec,500,6,which='SM')
sage: e[0]*float(501/pi)**2
0.999............
"""

Any ideas? That's a number changed from around 1 to 0, so a dramatic change.

When I run the tests bit by bit, I get the same problem.

32 drkirkby@fulvia:[~/fulvia/32/sage-4.5.2] $ ./sage
----------------------------------------------------------------------
| Sage Version 4.5.2, Release Date: 2010-08-05 |
| Type notebook() for the GUI, and license() for information. |
----------------------------------------------------------------------
sage: from cvxopt.base import *
sage: from cvxopt import umfpack
sage: from scipy import optimize
sage: from scipy import special
sage: from scipy import integrate
sage: from scipy.sparse.linalg import dsolve
sage: from scipy import interpolate
sage: from scipy import sparse
sage: import arpack
sage:
sage: import scipy
sage: from scipy.sparse.linalg.interface import aslinearoperator
sage: n=scipy.zeros((3,500))
sage: n[0,:]=-1
sage: n[1,:]=2
sage: n[2,:]=-1
sage: A=sparse.spdiags(n,[-1,0,1],int(500),int(500))
sage: A=aslinearoperator(A)
sage: e,v=arpack.speigs.ARPACK_eigs(A.matvec,500,6,which='SM')
sage: e[0]*float(501/pi)**2
0.0

Checking I see e[0] is zero.

sage: print e[0]
0.0
sage: float(501/pi)**2
25431.718415410425

I note this reads in cvxopt, and that furthermore that a new version of cvxopt
was merged in 4.5.2.rc1

http://trac.sagemath.org/sage_trac/ticket/9657

But as John has the same version of cvxopt as me, so that seems unlikely to be
the cause.

Any ideas?

Dave