Enhancement #623

Allow inversion rules inside MathVector<T>

Added by M. Rolf almost 13 years ago. Updated over 11 years ago.

Status:ClosedStart date:10/12/2011
Priority:LowDue date:
Assignee:M. Rolf% Done:

0%

Category:-
Target version:NemoMath 0.4

Description

==> Do not have to specify full inversion rules like

leftAddInverse<MathVector<TYPE> >

every time a new TYPE is used inside MathVector

History

#1 Updated by M. Rolf almost 13 years ago

  • Target version set to NemoMath 0.2

#2 Updated by M. Rolf over 12 years ago

  • Target version changed from NemoMath 0.2 to NemoMath 0.4

#3 Updated by M. Rolf over 12 years ago

Reduce compilation time by NOT instanciating Vectors/Matrices over several types T before absolutely necessary.

Possible solution:
Call with typename-inference leftAddInverse(MathVector<T>) and provide implementations inside MathVector.
The compiler should find them as 'specializations' by ADL, similar to the swap(ReadWriteRef) implementation in MathVector.

#4 Updated by M. Rolf over 11 years ago

  • Status changed from New to Closed

Something like this inside MathVector<T> was intended

friend MathVector<T> leftAddInverse(const MathVector<T> &value){return -value;}

Yet, this particular piece of code doesn't, because it is not recognized as template specialization of leftAddInverse<D>.

This version tries to do that specialization, but doesn't compile inside MathVector:

template<> MathVector<T> rightAddInverse<MathVector<T> >(const MathVector<T> &value){return -value;}

GCC says:
error: explicit specialization in non-namespace scope ‘class nemo::MathVector<T>’

You cannot specialize in a different namespace (here class-namespace) that the generic declaration (::nemo space).
C++03 specification:
An explicit specialization shall be declared in the namespace of which the template is a member, or, for 
member templates, in the namespace of which the enclosing class or enclosing class template is a member.

Doesn't seem to be possible.

#5 Updated by M. Rolf over 11 years ago

In order to reduce the compilation overhead, now only two types inside MathVector come with ready-made inversion rules:
  1. int
  2. double

These are also the two types that have ready-made typedefs IntVector and RealVector.

All other integer types can still be conveniently supplied with inversion rules (for addition/subtraction) by stating 

_nemo_vector_inversion_(TYPE)

in the client code.
The same holds for floating point types (addition/subtraction/multiplication/division) with 
_nemo_vector_cwise_inversion_(TYPE)

Also available in: Atom PDF