test failure: devel/sage/sage/calculus/interpolators.pyx

[kiwifb]

sage -t -force_lib "devel/sage/sage/calculus/interpolators.pyx"
**********************************************************************
File "/opt/sage/devel/sage/sage/calculus/interpolators.pyx", line 53:
    sage: show(m.plot_colored() + m.plot_spiderweb())
Exception raised:
    Traceback (most recent call last):
      File "/opt/sage/local/bin/ncadoctest.py", line 1231, in run_one_test
        self.run_one_example(test, example, filename, compileflags)
      File "/opt/sage/local/bin/sagedoctest.py", line 38, in run_one_example
        OrigDocTestRunner.run_one_example(self, test, example, filename, compileflags)
      File "/opt/sage/local/bin/ncadoctest.py", line 1172, in run_one_example
        compileflags, 1) in test.globs
      File "", line 1, in 
        show(m.plot_colored() + m.plot_spiderweb())###line 53:
    sage: show(m.plot_colored() + m.plot_spiderweb())
      File "riemann.pyx", line 728, in sage.calculus.riemann.Riemann_Map.plot_spiderweb (sage/calculus/riemann.c:8750)
      File "", line 1, in 
        m = Riemann_Map([lambda x: ps.value(x)], [lambda x: ps.derivative(x)],Integer(0))###line 52:
    sage: m = Riemann_Map([lambda x: ps.value(x)], [lambda x: ps.derivative(x)],0)
      File "interpolators.pyx", line 125, in sage.calculus.interpolators.PSpline.value (sage/calculus/interpolators.c:1270)
    ValueError: cannot convert float NaN to integer
**********************************************************************
File "/opt/sage/devel/sage/sage/calculus/interpolators.pyx", line 183:
    sage: show(m.plot_colored() + m.plot_spiderweb())
Exception raised:
    Traceback (most recent call last):
      File "/opt/sage/local/bin/ncadoctest.py", line 1231, in run_one_test
        self.run_one_example(test, example, filename, compileflags)
      File "/opt/sage/local/bin/sagedoctest.py", line 38, in run_one_example
        OrigDocTestRunner.run_one_example(self, test, example, filename, compileflags)
      File "/opt/sage/local/bin/ncadoctest.py", line 1172, in run_one_example
        compileflags, 1) in test.globs
      File "", line 1, in 
        show(m.plot_colored() + m.plot_spiderweb())###line 183:
    sage: show(m.plot_colored() + m.plot_spiderweb())
      File "riemann.pyx", line 728, in sage.calculus.riemann.Riemann_Map.plot_spiderweb (sage/calculus/riemann.c:8750)
      File "", line 1, in 
        m = Riemann_Map([lambda x: cs.value(x)], [lambda x: cs.derivative(x)], Integer(0))###line 182:
    sage: m = Riemann_Map([lambda x: cs.value(x)], [lambda x: cs.derivative(x)], 0)
      File "interpolators.pyx", line 277, in sage.calculus.interpolators.CCSpline.value (sage/calculus/interpolators.c:2390)                  
    IndexError: index out of bounds                                                                                                           
**********************************************************************

[kiwifb]

This is on x86. Related to the problem in strata.py?

[kiwifb]

Hard to track down. Not related to strata.py.
Not seen on ppc.
Not seen on my ~x86 box.
So the issue is confined to my home box on x86.

[kiwifb]

ATLAS again is responsible for that one but I am not sure where it acts.
My thought was that I linked gsl with an external clas (with the useflag cblas and
more recently cblas-external). Rebuilding gsl to use its own cblas (gslcblas) did not solve
the problem however even after rebuilding sage. Switching cblas from ATLAS to gsl (and
lapack-atlas to lapack-reference otherwise it doesn't work) solved the problem straight away.

[kiwifb]

Writing gave me an idea. Just switching from lapack-atlas to lapack-reference was enough to solve the problem. No parts of sage links directly to lapack so finding the faulty bits will be tricky.

[kiwifb]

There are multiple issues with ATLAS-3.9.xx in cblas and lapack. My position on this is that we should recommend to users to use another implementation. My favorite is openblas+lapack-reference from bicatali's overlay. It needs a fix though.

I need to write to Sebastien again about openblas (we already talked about it in the past).

[cschwan]

Good idea. What do you think of an ewarn message which is triggered if atlas is the default blas implementation ?

[kiwifb]

A possibility. But we also need something in the README so that people paying attention don't come to the last ebuild before being greeted by "danger you are using ATLAS-3.9.xx".

[cschwan]

Right! I remember atlas was one of the packages taking hours to compile ...

[cschwan]

Added a warning both to our README file (commit 0e89d1b) and to pkg_setup in sage itself (commit f1d4995).