Re: AW: AW: [boost] Re: ublas: assigning complex to real matrix

From: Toon Knapen (toon.knapen_at_[hidden])
Date: 2002-02-13 01:18:35

Walter, Jorg wrote:

> Hi Toon,
> 
> you wrote:
>>I've been thinking about this. I actually find the square brackets very 
>>intuitive. On the other hand, using the free functions it's more clear 
>>if you're using a row or a column. Finally I would like to have both 
>>(the square brackets and the free functions).
>>
> 
> OTOH our rationale (for trying ?) to deprecate subscripting was to further
> reduce class interdependencies ...
> 

But was there a problem with circular dependencies ? For the square 
brackets, I suppose Matrix is dependent on Vector. But I guess there's 
no dependency in the other direction.

> 
>>>>>>I patched following files (2 patches follow):
>>>>>>
>>>>>>
>>>>>>
>>>>>[Patch snipped]
>>>>>
>>>>>Based on our earlier discussion 
>>>>>(http://groups.yahoo.com/group/boost/message/21527) we've recently 
>>>>>added expression templates for real(vector/matrix) and imag
>>>>>(vector/matrix). You'll find the changes in boost CVS. If I've 
>>>>>understood your problem correctly, this should help also.
>>>>>
>>>>Great if had added the functionality already. Can you provide a small 
>>>>example to show me how to use it ?
>>>>
>>>Given your example, the corrected solution should be:
>>>
>>>double_matrix[0] = numerics::real(complex_matrix[0]) ;
>>>
>>But here you suppose that you now that the RHS is a complex. I also have 
>>generic algorithms where I don't know the value_type's of the matrices 
>>so they need to be converted automatically.
>>
> 
> No, numerics::real and numerics::imag should accept real matrices, too,
> computing the matrix itself or a zero matrix respective.
> 
But I can't now if the LHS is a real or complex matrix. 

The conversion (if necessary) is thus dependent on the type of the LHS and the RHS. 

Another solutions would e.g. be to something like :
typedef matrix< double > LHS;
LHS lhs;
matrix< complex > rhs ;

lhs = matrix_cast< LHS::value_type >( rhs );

But also here is no generic way to specify if you want to take the real 
or imag part of the RHS (as in my current solution)