Re: Problem in fem_wrapper: TypeError: unknown array type

Logan Sorenson wrote:

Hi Niels and Robert,

On Wed, Feb 10, 2010 at 4:08 AM, Robert Cimrman wrote:

...

Hi Niels,

Niels L Ellegaard wrote:

...

I did a fres install of the newest sfepy from git (on debian sid), and I have trouble running ./runTests.py. Can you see if I am doing something wrong?

cd /home/niels/local/src/ && git clone git://git.sympy.org/sfepy.git && cd sfepy && make && ./runTests.py --debug The same command sequence works for me. Could you send me more info? See below.

40 test file(s) executed in 52.13 s, 0 failure(s) of 55 test(s)

I'm also seeing this problem in sidux (based on debian sid). Using git bisect, it looks like the bug was introduced in commit c0e243c9.

Yeah, before this very recent commit, the base function evaluation was in pure Python. Now the evaluation is in C for speed, as I needed it for fast data interpolation between different meshes - this is already implemented in my github repo, I will push it to the master repo today, with the fix of this issue.

For me, I get the following flags (using winpdb). It looks like the problem is in self.mtx_i? Can you confirm if you see the same flags, Niels?

...

eval self.mtx_i.flags C_CONTIGUOUS : False F_CONTIGUOUS : True OWNDATA : True WRITEABLE : True ALIGNED : True UPDATEIFCOPY : False

this one should be C_CONTIGUOUS - it is easy to fix: diff --git a/sfepy/fem/poly_spaces.py b/sfepy/fem/poly_spaces.py index 28ee419..05051b5 100644 --- a/sfepy/fem/poly_spaces.py +++ b/sfepy/fem/poly_spaces.py @@ -232,7 +232,7 @@ class LagrangeSimplexPolySpace(PolySpace): mtx = nm.ones((n_v, n_v), nm.float64) mtx[0:n_v-1,:] = nm.transpose(geometry.coors) - self.mtx_i = nla.inv(mtx) + self.mtx_i = nm.ascontiguousarray(nla.inv(mtx)) self.rhs = nm.ones((n_v,), nm.float64) self.nodes, self.nts, self.node_coors = self._define_nodes() Could you try it, please? With numpy 1.3.0, this was not necessary, though. thanks for debugging! r.