[SciPy-user] bvp

[Pauli Virtanen]

Hi Nils,

Fri, 11 Apr 2008 20:03:24 +0200, Nils Wagner wrote:

> I installed bvp using the mercurial repository
> 
> hg clone static-http://www.iki.fi/pav/hg/bvp.hg bvp.hg
> 
> The second example doesn't work for me. Here is the output
> 
> /usr/bin/python -i ex2.py
> 1.0
> unexpected array size: new_size=1, got array with arr_size=0
> Traceback (most recent call last):
>    File "ex2.py", line 61, in ?
>      coarsen_initial_guess_mesh=True)
>    File "/usr/lib/python2.4/site-packages/bvp/colnew.py",
> line 522, in solve
>      vectorized_guess)
> _colnew.error: failed in converting 8th argument `fixpnt' of
> _colnew.colnew to C/Fortran array

Should also be fixed in 0.2.2 and current bvp.hg. (And yep, it was also 
caught by automated tests.)

The cause is that apperently something changed in f2py between numpy 
1.0.4 and 1.0.5: in colnew.pyf I have

	integer, dimension(11), intent(in) :: ipar
	double precision, dimension(ipar[10]), intent(in) :: fixpnt

However, f2py bugs out with the "failed in converting" if ipar[10] == 0 
and fixpnt.size == 0, which I don't think it did in 1.0.3 or 1.0.4. If 
fixed this by making fixpnt a shape = (1,) array even for ipar[10] == 0, 
and it appears to work on numpy 1.0.2, 1.0.4, 1.0.5, even though I don't 
know whether it should.

-- 
Pauli Virtanen