Re: [ublas] problem of "reference to reference" in vc6

From: Adrien Schvalberg (adrien.schvalberg_at_[hidden])
Date: 2005-08-01 09:35:11

Adrien Schvalberg a écrit :
> Hi all again,
> 
> I have a problem of reference to reference in my code with VC++6
> I'm using boost 1.32.0
> 
> I try to make the product of a matrix of doubles with a vector of points.
> As the +=, + and *(on left with double) operators are defined with my 
> point class, this should work.
> 
> See mvproduct.cpp for example code and vc6_output.txt for compiler error 
> output.
> 
> this code works with g++ 3.4.4 but not with VC++6.
> 
> in functional.hpp, line 182 there is a hint, but I don't understand how 
> to use it:
>         // ISSUE Remove reference to avoid reference to reference problems
> 
> does someone has any idea about this phenomenon?
> 
> Adrien Schvalberg
> 
> PS: I have to use VC6 for my work, so I understand that changing it 
> would be great (and I'm stuck with many other limitations of VC6), but I 
> can't ;-(
> 
> 

OK, self-answering, after some researchs, I've found the 2 following points:

1. I use BOOST_TT_BROKEN_COMPILER_SPEC(FakePoint) as a workaround to VC6 lacking 
partial specialization of templates

2. I do a specialization of boost::numeric::ublas::promote_traits<T1, T2>

template<> struct boost::numeric::ublas::promote_traits<double, FakePoint> 

{ typedef FakePoint promote_type; };

Does anyboday have a good idea about why T1 is the default return type in 
promote_traits<T1, T2>?
Will it break something if I replace T1 by T2?

I do the following operation: W = M*V
with M: (mxn)-double matrix
      V: n-Thing vector
Thing can be a point, a quaternion, anything you want such that double*Thing -> 
Thing, Thing + Thing -> Thing, and Thing(0) (equals to the neutral element of Thing 
addition law) are defined.
W should be a m-Thing vector, so the promote type should be Thing, ie T2

Adrien Schvalberg