Frederick, 

Remember that C/C++ slogant is to keep backward compatibility for 50 years or more.

That is, the processors that use segmentation addressing, like real mode intel CPU's should be supported, other processors like the Z80 which uses 16bit segmented addressing mode....etc.

On the other side, "relaxed" is better than "strict" rules, because it will englobe old and new architectures, especially if it works fine.

But there is cost for that.

Look at this scenario. 

template<typename T>

struct A{

T* ptr = &m_d;

T m_d;

};

If on some architecture where sizeof(T*) is so small that it cannot point to a memory segment (for example the stack segment), then you will be faced with a problem where you cannot allocate the above structure in that memory segment.

Expl: for T = char and sizeof(char*) ==2, and the architecture is 32bit address space.

Then you will never be able to create a stack variable of type A<char>, because A::ptr cannot point to a 32bit memory address in the stack segment,

 yet this scenario is totally fine.

 How?

Behind the scene, the compiler could promote sizeof(char*) from 2 to 4, so it can access the whole address space, or it can use a segment and an index, or some other technique that I don't know of....all these are extra CPU instructions and bookkeeping.

In conclusion,  what you are asking for is sharp and decisive,  but the state of the world dectates otherwise.