Re: [eigen] two decisions to take

[Benoît Jacob <jacob@xxxxxxxxxxxxxxx>]

On Wednesday 05 September 2007 19:30:08 Schleimer, Ben wrote:
> I was think of infix as x.dot(y) as opposed to prefix which is dot(x,y).

OK.

> Whether the methods are named or are overloaded operators doesn't matter.
> Actually it would be very nice if for every overloaded operator, there is
> the named version too just so if we want our expressions to be clearer,
> it's possible.

OK, I'll keep this in mind. Of course not for +(add) -(sub) *(mul) as these 3 
operators are really standard. But for other operators, yes.

> Yippy! Fun with priorities. Just the thing to make or break eigen2. We need
> named methods...

OK I get your point.

> Sounds like something to leave until someone asks for that feature. They
> both sound useful but outside of tensor operations, I can't imagine what it
> could be used for.

I do see a common use case. At least I know I'd use it in libavogadro. Often 
you want to build a matrix with prescribed columns. That is actually the most 
natural way to build a matrix, because from the point of view of linear maps 
it amounts to build a map f from the values f(e) where e ranges over the 
vectors in the standard basis. So that's a use case for operator&.

The same thing with rows (hence operator&&) is useful when dealing with 
orthogonal maps because then inverse=transpose and hence you can directly 
build the inverse by assembling rows. Again I do so in libavogadro.

Cheers,
Benoit

Attachment: signature.asc
Description: This is a digitally signed message part.