Automatic arrays and intrinsic array operations: to use or not to use?

[awvwgk]

@zaikunzhang

We see that diag declares its result as an automatic array, while eye declares it as an allocatable array. Why are they different?

In stdlib we actually prefer automatic arrays where possible, even if this can lead to stack overflows with some compilers. The reason for using an allocatable return value in eye is that the second dimension is optional and we cannot reliably declare the shape of the automatic array anymore.