Re: [eigen] Serious conflict between Dynamic and -1 in Diagonal<>

[Christoph Hertzberg <chtz@xxxxxxxxxxxxxxxxxxxxxxxx>]

On 10.07.2012 10:34, Gael Guennebaud wrote:
> Hi,
>
> as pointed out in this thread:
>
> http://forum.kde.org/viewtopic.php?f=74&t=106803
>
> there is a serious issue in the Diagonal<Expression, DiagIndex> class.
> Recall that Diagonal<> represents an expression of the diagonal
> (DiagIndex==0), a super-diagonal (DiagIndex>0) or a sub-diagonal
> (DiagIndex<0).
>
> Of course, one can also set DiagIndex to Dynamic to pick a given
> super/sub diagonal at runtime. The problem is that since Dynamic==-1,
> we cannot distinguish between the first sub-diagonal (DiagIndex==-1)
> and a runtime selected diagonal.
>
> Since changing the value of Dynamic is not an option, I propose to add
> a special enum for this use case, like RuntimeDiagIndex or
> DynamicDiagIndex that would be set to a very big value.
>
> Opinion? better solution?

Another solution would be to index sub-diagonals by (-idx-1) (i.e. 
~idx), so main-diagonal is 0, 1st sub-diagonal is -2=0xfffffffe, etc.

You could still keep the syntax of m.diagonal<-1>() for the first 
sub-diagonal, and by redefining [Const]DiagonalIndexReturnType you 
almost keep API-compatibility (not if someone directly uses 
Diagonal<MatrixType, Index>).

Btw:
For direct access types the return type of m.diagonal() could be just a 
Map with inner stride of m.outerStride()+m.innerStride(), couldn't it?

Christoph


--
----------------------------------------------
Dipl.-Inf. Christoph Hertzberg
Cartesium 0.049
Universität Bremen
Enrique-Schmidt-Straße 5
28359 Bremen

Tel: +49 (421) 218-64252
----------------------------------------------