Re: [eigen] Bugs in Mat::Random, reductions

[Benoît Jacob <jacob@xxxxxxxxxxxxxxx>]

I just thought of why my idea with #include<Eigen/Eigen> doesn't work !
Suppose that the user #includes two headers A.h and B.h and that both A.h and 
B.h #include Eigen/Eigen. Then the second one triggers the #error "you must 
include Eigen/Eigen last" !

After A.h is included, class MatrixBase is defined, no way that B.h may 
include its own modules...

So the best that we can do is improve documentation wrt the #includes.

Cheers,
Benoit


On Saturday 25 October 2008 20:10:33 Gael Guennebaud wrote:
> ah, now I'm puzzled again. indeed, in that case the user even don't
> have to read the doc:
>
> first attempt:
>
> #include<Eigen/Eigen>
> m.cwise().sin();
>
> => error including: you_must_include_array_module_first
>
> second attempt:
>
> #include<Eigen/Eigen>
> #include<Eigen/Array>
> m.cwise().sin();
>
> error: Eigen/Array must be included before Eigen/Eigen header
>
> third attempt:
>
> => the lesson is learned  !!
>
>
> so why not.
>
> Then there is a practical issue because it is not possible to keep
> both approaches for a smooth transition. So I guess we should only
> enforce a correct order with a warning and mark Eigen/Core as
> deprecated. And then wait some times before doing the real change.
>
> gael.
>
> On Sat, Oct 25, 2008 at 7:44 PM, Benoît Jacob <jacob@xxxxxxxxxxxxxxx> wrote:
> > I mostly agree with that, but I would like to put one more evolution of
> > this solution on the table.
> >
> > Instead of requiring the user to do
> > #define EIGEN_USING_ARRAY_MODULE
> > #define EIGEN_USING_GEOMETRY_MODULE
> > #include <Eigen/Eigen>
> >
> > we could do
> >
> > #include <Eigen/Array>
> > #include <Eigen/Geometry>
> > #include <Eigen/Eigen>
> >
> > In other words, the only difference from the current situation, from the
> > user's point of view, is that he must always include Eigen/Eigen last.
> >
> > Eigen/Array would just do:
> > 1) check that Eigen/Eigen is not already included (#error otherwise)
> > 2) define EIGEN_USING_ARRAY_MODULE
> >
> > Let me explain why it might still be worth doing that. We brought this
> > conversation from the topic of error messages when the user forgets to
> > #include a module. But this is a broader issue. The real question is, is
> > it potentially useful to let Eigen know globally which modules are
> > #included? The issue of error message is just one example of where it is
> > useful to know that. There might be more...?
> >
> > So, I have the feeling that it might be worth asking the user to
> > #include<Eigen/Eigen> last.
> >
> > Opinions?
> >
> > Benoit
> >
> >> Since no one complained, I guess nobody is against that solution. On
> >> the other hand nobody seems to be excited by it, and actually, neither
> >> do I. I mean, currently we have an intuitive mechanism to include
> >> modules, but some weird link error messages if someone forgot to
> >> include the right module.  And the we would change for more intuitive
> >> error messages but with an unintuitive inclusion mechanism. So
> >> eventually that's probably not worth it. In both cases we have to
> >> compensate some unintuitiveness  by big warning in the documentation.
> >> So let's keep the current state, and improve the doc. isn't it ?
> >>
> >> gael.
> >>
> >> ---
> >
> > ---



---