[ublas] scalar_vector doesn't support assignment of value

From: glen stark (stark_at_[hidden])
Date: 2005-02-17 12:26:31

ublas-request_at_[hidden] wrote:

>>>I think this is a good way to go. If you like the = syntax you can define
>>>	operator= (scalar_vector<double> v, double s)
>>>	{
>>>		v = scalar_vector<double> (v.size(), s);
>>>	}
>>>A good compiler should turn this into efficient code.
>>>      
>>>
>>Unfortunately, operator=() must be a member function.
>>
>>fres
>>    
>>
Well, I copy and pasted the wrong code segment, so it must have been 
late for me too :P...

I meant to post this:

            BOOST_UBLAS_INLINE
                scalar_vector &operator = (const value_type& v){
                        value_ = v;
                        return *this;
                }

Which maybe is more understandable.  Instead I posted some part of the 
vector header I dont' understand myself.  This just overloads = operator 
in an obvious way for a scalar vector.

So, I don't think supporting   scalar_vector=value_type  is a bad idea, 
but I don't mind seing it done as a set(value_type) function, which 
could also be used to set all members of a dense vector (or matrix for 
that matter) to a particular value.  Wasn't there some discussion on the 
list earlier about needing a way to initialize matrix elements anyway?

My two cents worth regarding syntax: I think it would be elegant to 
overload the = operator for all matrix and vector types, where all non 
zero elements are set to rhs.   For banded, diagonal, or sparce 
matricies, I would expect this to set all non zero elements to the rhs. 

I don't have a strong opinion about the syntax, but I think this is 
functionality that should be implemented, and  ought not to be difficult 
to do.  Certainly I could pull it off for scalar and dense vectors.  Who 
makes the executive decision on these things?  Is it Joerg?  I'm willing 
to try  submitting a patch if the syntax is agreed on.

Glen