Re: problem with fem-fenics

[c.]

On 23 Jul 2013, at 05:32, c. <address@hidden> wrote:

> 
> On 19 Jul 2013, at 10:17, Marco Vassallo <address@hidden> wrote:
> 
>> Hi all,
>> 
>> for the fem-fenics pkg, I have to solve some issues but I can't exactly 
>> figure out how we can do it.
>> 
>> The pkg is supposed to solve a Problem which is defined by the user.
>> 
>> The user defines the problem in a file myproblem.ufl.
>> 
>> Then our pkg should compile it: at the moment, I'm doing it with a script 
>> file which
>> calls the compiler [1] :
>> 
>>    command = "ffc -l dolfin myproblem.ufl"
>>    [output, text] = system (command);
>> 
>> This script generates a .h file, myproblem.h, which contains the definition 
>> of some classes 
>> which I should use in my DLD function.
>> 
>> Then, what I need is to include the header in a function, compile it with 
>> mkoctfile and to execute it.
>> 
>> A sketch of what I'm doing right now is available here [2], but I don't know 
>> if it could work properly inside a pkg. 
>> 
>> [1] 
>> https://sourceforge.net/p/octave/fem-fenics/ci/default/tree/inst/fem_ffc.m
>> [2] 
>> https://sourceforge.net/p/octave/fem-fenics/ci/default/tree/example/fem_func_space.m
>> 
> 
> Hi Marco,
> Sorry for the late reply, I've been quite busy lately as I have an imposrtant 
> meeting tomorrow and the day after.
> Maybe we should arrange to meet in person by the end of this week? will you 
> be in milano on Friday?
> 
> c.
> 

After meeting face-to-face with Marco on Friday it seems this issue was easier 
to 
solve with simple m-files than was expected.
So it seems Marco's project is in very good shape for the mid-term evaluation 
next week.

The main outstandig issue is with converting dolfin sparse matrices to Octave. 
Quite surprisingly there seems to be no documented method to query for the 
number
of nonzeros in a dolfin sparse matrix, so what he is currently doing is to 
allocate
space for a FULL matrix then clean up the extra space later.

This is clearly not acceptable even for medium size 2D problems an completely 
unfeasible for 
quite smal 3D problems. Marco is going to enquire on the dolfin mailing list to 
find out
if there is any way to find out the size of required memory somehow. In the 
meantime I 
was wondering if anyone has an idea about how to incremntally fill and allocate 
a sparse 
matrix in Octave without being too inefficient?

c.