Myujth

I write a function that takes two DenseBase as arguments.

The function uses .derived().array() to convert both Array and Matrix to Array.

I got tired of writing derived for many times and use auto.

But auto leads to strange error. Eigen complains that x2 and y2 don't have same shape.

If I don't want to write .derived().array() for many times, what can I use?

Eigen is from https://github.com/eigenteam/eigen-git-mirror.git

#include <Eigen/Eigen>
int main() 
 Eigen::ArrayXf x(3);
 Eigen::ArrayXf y(3);
 x << 1, 2, 3;
 y << 4, 5, 6;
 // x.derived().array() * y.derived().array();
 auto x2 = x.derived().array();
 auto y2 = y.derived().array();
 y2 = x2 * y2; 

Run time error:

CwiseBinaryOp.h:110: ...

Assertion `aLhs.rows() == aRhs.rows() 
 && aLhs.cols() == aRhs.cols()' failed.

You can fix the runtime issue with auto x2 = x.array().derived();, that is: reverse array and derived. But auto is not desirable here. Here is why. Say you have:

template <typename T> void foo(DenseBase<T> &x);

If T is an Array<> then x.array().derived() is an Array<> and x2 will be a deep copy of x. In this case you would like to use auto& x2 = ....

If T is something else, e.g., a Matrix<>, then auto x2 = x.array().derived(); is perfectly fine, but not auto& x2 = ....

So what you really want is something as complicated as:

internal::ref_selector<std::decay<decltype(x.array().derived())>::type>::non_const_type
 x2 = x.array().derived();

Not nice :(

A simpler solution is to not bother and create an ArrayWrapper even for inputs that are already in the array world:

ArrayWrapper<T> x2(x.derived());

Yet another simple solution is to enforce the caller to pass expressions in the array world:

template <typename T> void foo(ArrayBase<T> &x) 
 T& x2(x.derived());
 ...