[Issue 15672] Casting from void[] to T[] is erroneously considered @safe

[via Digitalmars-d-bugs]

https://issues.dlang.org/show_bug.cgi?id=15672

--- Comment #1 from hsteoh at quickfur.ath.cx ---
It's not necessarily safe to cast from void[] to immutable(T)[]. Consider:
-----
int[] a = [ 12345, 54321 ];
void[] b = a;  // any array can implicitly convert to void[]
immutable(Object)[] c = cast(immutable(Object)[]) b; // suppose this was
allowed
b[0].toString(); // illegal pointer dereference
-----

In order to ensure @safety, we cannot allow reinterpreting *anything* as a
pointer, that wasn't already a pointer of the same type, and with the same
attributes.

Note that it's not @safe even to convert from a pointer of the same type but
different attributes. For instance:
-----
alias safeFunc = void function() @safe;
alias unsafeFunc = void function() @system;

void main() @safe {
    unsafeFunc[] unsafePtrs = [ &unsafeFunc ];
    void[] voidPtrs = unsafePtrs; // OK, everything converts to void[]
implicitly
    auto arr = cast(immutable(safeFunc)[]) voidPtrs; // OK to convert func ptrs
to func ptrs, right?
    arr[0](); // oops, we just called a @system function from @safe code
}
-----

The void[] step is not necessary, but illustrates the danger of allowing
conversions from void[] to immutable(T)[].

--