Is the current layout composition algorithm incomplete?

[admbbs]

The current layout composition algorithm depends on left-distributive and stride-divisibility-condition. It seems to have some corners not covered. Consider layout A, B and their composition as follows,

auto A = Layout<Shape<_2, _3>, Stride<_3, _1>>{};
auto B = Layout<Shape<_2, _2>, Stride<_3, _2>>{};

// auto C = composition(A, B); // stride-divisibility-condition violated.
auto C = Layout<Shape<_2, _2>, Stride<_4, _1>>{};

static_assert(C(Int<0>{}) == A(B(Int<0>{})));
static_assert(C(Int<1>{}) == A(B(Int<1>{})));
static_assert(C(Int<2>{}) == A(B(Int<2>{})));
static_assert(C(Int<3>{}) == A(B(Int<3>{})));

I got 'static assertion failed with "Stride Divisibility Condition"' when I tried to cute::compose layout A and B. But the not-commented layout C constructed by-hand clearly is a valid composition, which is also supported by these static assertions.

Are cases like this deliberately omitted? Is it mathmatically impossible to handle them, or because the code would cost to much to compile or run?

Thanks in advance for any answers or clues.