Thanks Michael.
I'm trying two different paths here:
1) see if I come up with a build.conf that is correctly recognized (before
it looked it wasn't, but that may be because of my wrong syntax in name the
libs). I'm trying first with the lapack/blas that I got time ago when
installing Xcode. My build.conf seems to be working now. Build says
[...]
**************** Build summary ****************
User-configured LAPACK and BLAS
Great. The install does not complain either (not that it did say much
before either). However, the test fails miserably again
$python -c 'import kwant; kwant.test()'
Traceback (most recent call last):
File "<string>", line 1, in <module>
File "kwant/__init__.py", line 14, in <module>
exec 'from . import {0}'.format(module)
File "<string>", line 1, in <module>
File "kwant/system.py", line 17, in <module>
from . import physics, _system
File "kwant/physics/__init__.py", line 14, in <module>
exec 'from . import {0}'.format(module)
File "<string>", line 1, in <module>
File "kwant/physics/leads.py", line 16, in <module>
from .. import linalg as kla
File "kwant/linalg/__init__.py", line 10, in <module>
from . import lapack
ImportError: cannot import name lapack
Should I use some extra compile/linking arguments in the build.con file? if
so, which ones?
My build.conf was
[lapack]
libraries = f77lapack blas
library_dirs = /usr/lib/
2) The second path is installing it via a virtual machine, as Christoph
mentions in his reply.
On Mon, Feb 9, 2015 at 4:46 PM, Michael Wimmer wrote: Dear Pol, very hard to say what the problem is. I couldn't find a Mac with OSX
10.6, so I cannot reproduce the error. Have you tried to install kwant using homebrew? That still supports OSX
10.6, I believe. Otherwise, you could still try to use a different blas or lapack, as you
tried originally, and see if the problem goes away. Note that build.conf
must be in the same directory as setup.py. Also, the line libraries
should only contain the library name without the "lib" in front, and
without the trailing ".dylib. Then you should be able to use a custom
BLAS and LAPACK installation Best, Michael On 09.02.2015 14:47, Pol Nasam wrote: Thanks Michael, Here the ouput with verbosity=2 $python -c 'import kwant; kwant.test(verbose=2)' >&
Kwant-test-verbose2.txt
$cat Kwant-test-verbose2.txt /Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/kwant/solvers/sparse.py:94: RuntimeWarning: The installed SciPy does not use UMFPACK. Instead,
SciPy will use the version of SuperLu it is shipped with.
Performance can be very poor in this case.
"Performance can be very poor in this case.", RuntimeWarning)
kwant.graph.tests.test_core.test_empty ... ok
kwant.graph.tests.test_core.test_num_nodes ... ok
kwant.graph.tests.test_core.test_large ... ok
kwant.graph.tests.test_core.test_small ... ok
kwant.graph.tests.test_core.test_negative_node_ids ... ok
kwant.graph.tests.test_core.test_add_edges ... ok
kwant.graph.tests.test_core.test_edge_ids ... ok
kwant.graph.tests.test_slicer.test_rectangle ... ok
kwant.graph.tests.test_utils.test_make_undirected ... ok
kwant.graph.tests.test_utils.test_remove_duplicates ... ok
kwant.graph.tests.test_utils.test_induced_subgraph ... ok
kwant.linalg.tests.test_linalg.test_gen_eig ... ok
kwant.linalg.tests.test_linalg.test_lu ... ok
kwant.linalg.tests.test_linalg.test_rcond_from_lu ... ok
kwant.linalg.tests.test_linalg.test_schur ... ok
kwant.linalg.tests.test_linalg.test_convert_r2c_schur ... ok
kwant.linalg.tests.test_linalg.test_order_schur ... ok
kwant.linalg.tests.test_linalg.test_evecs_from_schur ... ok
kwant.linalg.tests.test_linalg.test_gen_schur ... ok
kwant.linalg.tests.test_linalg.test_convert_r2c_gen_schur ... ok
kwant.linalg.tests.test_linalg.test_order_gen_schur ... ok
kwant.linalg.tests.test_linalg.test_evecs_from_gen_schur ... ok
kwant.linalg.tests.test_lll.test_lll ... ok
kwant.linalg.tests.test_lll.test_cvp ... ok
kwant.linalg.tests.test_mumps.test_lu_with_dense ... SKIP: Skipping
test: test_lu_with_dense: Test skipped due to test condition
kwant.linalg.tests.test_mumps.test_schur_complement_with_dense ...
SKIP: Skipping test: test_schur_complement_with_dense: Test skipped
due to test condition
Test if MUMPSError -9 is properly caught by increasing memory ...
SKIP: Skipping test: test_error_minus_9: Test skipped due to test
condition
kwant.physics.tests.test_dispersion.test_band_energies ... ok
kwant.physics.tests.test_leads.test_analytic_numeric ... ok
Selfenergy with an invertible hopping matrix, and degenerate bands.
... ok
This is a testcase with invertible hopping matrices, ... ok
This testcase features a rectangular (and hence singular) ... ok
This testcase features a singular, square hopping matrices ... ok
This testcase features a rectangular (and hence singular) ... ok
This testcase features a rectangular (and hence singular) ... ok
kwant.physics.tests.test_leads.test_singular_h_and_t ... And here it segfaults... On Sun, Feb 8, 2015 at 2:24 PM, wimmer Hi Pol, can you run kwant.test as 'kwant.test(verbose=2)' to see at which
test the segfault occurs? I don't remember all the details, but MacOSX 10.6 was a bit special
in that there was some mix of 32bit and 64bit. It might be some
tricky stuff ... Best, Michael On , Pol Nasam wrote: Thanks Michael. Well, changing directory did make a difference.
However, I get now a segmentation fault: $python -c 'import kwant; kwant.test()' /Library/Frameworks/Python.__framework/Versions/2.7/lib/__python2.7/site-packages/kwant/__solvers/sparse.py:94: RuntimeWarning: The installed SciPy does not use UMFPACK.
Instead,
SciPy will use the version of SuperLu it is shipped with.
Performance can be very poor in this case. "Performance can be very poor in this case.", RuntimeWarning) ........................SSS...__.....Segmentation fault Right now I don't care about the MUPS lib. Thinking on having first kwant running ok, before reinstalling with MUPS. It seems to be there is some minor, even obvious, but essential step that I may be missing. The
not testing things in the same directory where kwant was untarred
should come in the install readme,
for instance. But what else could I be missing now? Cheers, On Sun, Feb 8, 2015 at 11:34 AM, Michael Wimmer
import kwant