dolfin team mailing list archive

[Garth Wells <726487@xxxxxxxxxxxxxxxxxx>]

On 28/02/11 17:51, Johan Hake wrote:
> On Monday February 28 2011 09:28:14 Garth Wells wrote:
>> On 28/02/11 16:29, Johan Hake wrote:
>>> The python LA unit test desperately need a revamp... We have removed the
>>> Matrix slicing as it is broken in parallel and the present Matrix test
>>> more or less just test the slicing.
>>>
>>> However feel free to add more test classes. For example one could be
>>> specific for testing parallel stuff. Something like:
>>>
>>> class ParallelStuff(obejct):
>>>     pass
>>>
>>> ...
>>>
>>> if has_la_backend("PETSc"):
>>>     bases = (AbstractBaseTest, DataNotWorkingTester,\
>>>     
>>>              unittest.TestCase)
>>>     
>>>     bases += (ParallelStuff,) if MPI.num_processes()>1 ()
>>>     PETScTester = type("PETScTester", bases, {"backend":"PETSc"})
>>
>> I would suggest having Matrix.py to test plain matrix functionality,
>> Vector.py to test plain vector functionality, etc. By plain I mean the
>> functions in the C++ classes. We can then have something like
>> Python_la.py to test Python/numpy/scipy magic.
> 
> Sounds good.
> 
> But I suggest we still create a separate parallel test class for testing 
> parallel specific functionality.
> 

Ideally, everything will work in parallel, so a special serial-specific
test would be better ;), if needed.

Garth

> Johan
> 
>> Garth
>>
>>> Johan
>>>
>>> On Monday February 28 2011 05:52:28 Joachim Haga wrote:
>>>> Could certainly add a rectangular matrix to unit/la/python/test.py, but
>>>>
>>>> it looks like the matrix tests are not run at all:
>>>>     # FIXME: Matrix tests need to be updated once interface is sorted
>>>>     out #def test_matrix_with_backend(self):
>>>>     #    self.run_matrix_test(True)
>>>>     
>>>>     #def test_matrix_without_backend(self):
>>>>     #    self.run_matrix_test(False)
>>>>
>>>> ...could easily modify demo/undocumented/block-matrix instead, but that
>>>> would be backend specific...
>

-- 
You received this bug notification because you are a member of DOLFIN
Team, which is subscribed to DOLFIN.
https://bugs.launchpad.net/bugs/726487

Title:
  PETScMatrix vector resize uses wrong dim

Status in DOLFIN:
  New

Bug description:
  The size of vectors resized through PETScMatrix::resize is wrong for
  non-square matrices. The left and right matrices are switched. Patch
  attached.