SymmetricTensor access returns const reference

[sebproell]

Interestingly, the following compiled:

const SymmetricTensor<2, dim, VectorizedArray<double>> tensor{};
tensor[0][0] = 1.0;

The assignement is performed on a temporary copy returned from the accessor but I obviously wanted to modify the actual tensor. Imo, the correct behavior would be to get a compiler error that tensor is const (in this case by accident).

I don't know whether the decision to return a value rather a reference was intentional. Unless it enables some optimizations(?), I don't see why the const access to a tensor should return something different than a const reference.

[drwells]

The original type seems wrong - if it should be a value then it should be const Number. I'm not sure if there's any advantage to doing this one way or the other.

/rebuild

[kronbichler]

I guess the question is whether we want to expose the underlying memory address (as with a reference) or not (then, the right type would be const Number;). I agree with the current patch.

[kronbichler]

This is clearly forward progress, let us merge this now.

[bangerth]

What an interesting bug! The original design was for tensors with double underlying scalar types. There, if the function just returns a double, it is an rvalue to which you can't assign anything: an rvalue of non-class type can't appear on the left hand side of an assignment.

But if it is a class type like you have here, one can by default. I bet that we have that bug in any number of places where we access elements of arrays of some sort and the const version of operator[] returns the object by value.

There is a way to disallow this, starting in C++11, see https://stackoverflow.com/questions/26809789/how-to-disallow-assigning-to-a-temporary-object-and-implicit-casting-from-other . In essence, we would have to add

VectorizedArray& operator=(const VectorizedArray &) && = delete;

to class VectorizedArray -- note the && at the end of the declaration. I am pretty sure that we actually want to disallow this kind of thing and so if someone wanted to give a patch to this effect a try, I would very much be in favor of it. @sebproell: Want to give this a try?