[SciPy-dev] scipy.test() abruptly quits on scipy svn 0.7.0.dev5241

[Peter Norton]

On Thu, Dec 11, 2008 at 1:13 AM, David Cournapeau
<david at ar.media.kyoto-u.ac.jp> wrote:
> Peter Norton wrote:
>> I've had the same result for the last few checkouts. I've been
>> building with numscons, and as far as I can tell no severe errors are
>> coming up in the build. However I get this:
>> [...]
>> This looks like it's bailing out, but it doesn't appear to be exiting
>> with any errors. This was invoked as:
>>
>
> Yes, this is basically a library fatal error (illegal value means that
> the sunperf will call exit I believe, assuming you are using sunperf),
> and I don't think we can catch it in numpy/scipy (it is a bug that the
> error happens at all, though). IOW, it is a crash.
>
> David

I was doing this with ATLAS+Lapack, but I realized that I was having
some other difficulties with ATLAS+gcc (it wasn't obvious that gnu as
wasn't supported for ia32 due to "/" being treated as a comment
character on solaris x86. It's only fixed for 64-bit assembler!).

So I've gone back to scipy + numpy being built with sun's CC and using
libsunperf and numscons. And David, kudos to you because in the past
this was quite difficult. The current SVN head built with only one
minor error, that is the dreaded "o" command being invoked instead of
"CC -o". I've written a wrapper, and this is easier because it now
only seems to get invoked for c++ code.

Anyway the same essential crash is still happening:

...Result may be inaccurate, approximate err = 1.78418260944e-08
...Result may be inaccurate, approximate err = 7.27595761418e-12
.....SSS.........................................................................................................................................................................................EEEFFEEE..........................................................................................................................................................................................................................................................ESSSSSSSSSSSB->ncol
= 1  Bstore->lda = 5  A->nrow = 5
B->Stype = 5  B->Dtype = 0  B->Mtype = 0
** On entry to sgssv, parameter number 7 has an illegal value.
Segmentation Fault (core dumped)

The top of the stack looks like this:

#0  0xfe57da76 in ___pl_Destroy_SuperNode_Matrix () from
/usr/local/python-2.5.1/lib/libsunperf.so.3
#1  0xfb8832c4 in Py_sgssv (self=0x0, args=0x9c11c6c, kwdict=0x0) at
build/scons/scipy/sparse/linalg/dsolve/_ssuperlumodule.c:100
#2  0xfecd6bf2 in PyCFunction_Call (func=0x89cae4c, arg=0x9c11c6c,
kw=0x0) at Objects/methodobject.c:77
#3  0xfed20548 in PyEval_EvalFrameEx (f=0x9bf0cec, throwflag=0) at
Python/ceval.c:3564

Is there any more information I can gather to help resolve this crash?

Aside from this, there seem to be many test errors that come up, most
of them in the *fourier tests. I'm not sure if these are expected.

Thanks,

-Peter