PyClaw's Controller class loses Fortran order of frames

[Aron Ahmadia]

All,

I've filing a bug against this here: https://github.com/clawpack/pyclaw/issues/259

It's enough of a gotcha that I thought I would send a general warning to PyClaw developers and users.

The following clawpack controller code 'loses' the imposed Fortran order passed in from the state object when a frame gets passed in:

        if self.keep_copy:

            self.frames.append(copy.deepcopy(self.solution))

due to this bug in numpy's deepcopy semantics: https://github.com/numpy/numpy/pull/2699

This means that output frames are of default numpy type (C contiguous).

In fact, the output solution does not appear to have either C or Fortran contiguous ordering, I'm not even sure what that means.

In [80]: claw.solution.q.flags

Out[80]:

  C_CONTIGUOUS : False

  F_CONTIGUOUS : False

  OWNDATA : False

  WRITEABLE : True

  ALIGNED : True

  UPDATEIFCOPY : False

This won't affect anybody who's not interfacing into other languages, but I was quite surprised to see this behavior, especially from a library as robust as numpy.

Cheers,

Aron