This is more of a bug report/feature request than a real pull request, since I'm certain the solution (and possibly the problem description) here is incomplete at best. The additional unit test shows the situation :

        const T data[] = {1, 2, 3};
        TensorMap<const T, 3> mdata{data};
        Tensor<T, 3> tdata = mdata;

Prior to this patch, the above will not compile. That is, we cannot assign a TensorMap wrapping an array of const qualified elements to a Tensor containing non-const elements.

For background, I came across this issue when wanting to copy the (3-dimensional) positions of the 4 nodes in a tetrahedron into a 4x3 tensor. Those node positions are passed as const-qualified (non-Fastor) vectors, which I attempted to TensorMap then assign each in turn to a slice of my Tensor. But of course that didn't work.

This pull request does the very minimum to get the above situation working. I'm happy for you to take it over or replace it, or I can do minor tweaks as requested. I started using Fastor only a couple of days ago, so I'm probably not the best person to be fiddling ;)

A brief summary of changes:

• Use destination type (e.g. non-const) to construct SIMDVector in eval functions, otherwise we get "assignment of read-only location" errors.
• Don't const-qualify destination tensor in assign, otherwise overload resolution of called trivial_assign fails.
• Missing inline declaration for a maskstore overload causing "multiple definition of Fastor::maskstore" errors (this is unrelated to the issue of assigning const data, but caused compilation errors for me - should probably be a separate PR/issue).