Re: [eigen] unpleasant surprise mit math functions

[Christoph Hertzberg <chtz@xxxxxxxxxxxxxxxxxxxxxxxx>]

On 23.09.2012 15:24, Gael Guennebaud wrote:
> thank you Christoph and Karli, that makes perfect sense. So I thought
> the solution would simply to put the "internal" base classes into
> their own namespace. That worked for some expressions, but not for

How about putting the math-functions into their own namespace? And why 
are they required at all?
I remember once users were required to specify ei_sin, etc if they had 
custom scalar types and this changed to internal::sin (no pun intended) 
in Eigen 3. However, I think it would make much more sense to make them 
declare that in the namespace of their class (or in std:: or global, if 
they prefer) and this seems to be sufficient at the moment.
Or is it to prevent users from calling sin on integer arrays? I would 
say this goes into the category "shot yourself in the foot" and does not 
necessarily need to be prohibited by Eigen. Btw: what should happen if I 
call sin for a rational number class?

If you can't/don't want to remove the internal::sin function then you 
could provide a function Eigen::sin(const Eigen::EigenBase<Derived>&); 
which statically asserts that Derived is an Array-Expression. In 
general, hoping that a namespace is hidden does not seem a good way to 
prohibit function calls. After all, the user should always be allowed to 
declare using Eigen::internal; without producing illegal code.


> all. Typically if a template class Expr<Func> defined in namespace A
> is instantiated with a class defined in internal then, all the generic
> functions of internal become readily available for the class Expr.
>
> [...]
>
> Did anyone be aware of such a rule? It does not make sense to me, and
> looks rather dangerous.

Well, the biggest problem with ADL is that hardly anyone really knows 
completely what it does ...
As said above, I would never rely on functions in other namespaces not 
being visible.

Christoph



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

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