Bug 8956 – Ability to break typesystem with constructor/postblit/destructor (e.g. modify immutable)

As was mentioned in Issue 4338 and Issue 4867 one is able to break typesystem using postblit/destructor. But that's not all. Constructor also can be used for this. Example: --- int* p1, p2, p3; struct S { int* p; this(int* p) { p1 = this.p = p; } this(this) { p2 = p; } ~this() { p3 = p; } } void main() { immutable s = immutable S(new int); // call constructor { immutable tmp = s; } // call postblit and destructor assert(p1 is s.p && p2 is s.p && p3 is s.p); } --- The worst is that it can be done accidentally. E.g.: --- struct S { int* p; this(int* p) { this.p = p; } // Looks really innocuous, isn't it? } void main() { int i; immutable s = immutable S(&i); // Feel constructor's destructive power! assert(&i is s.p); } --- Please, create enhancement requests as other issues and add e.g. `Depends on` links here.

The first fix proposal: Issue 8958

this(int* i) has no const/immutable qualifier. As such, if I remember correctly, the compiler should not be allowing it to be used to construct an immutable instance. In any case, it's a bug that it accepts the call to it to construct an immutable without having first checked that it's still valid when the this pointer is immutable.

It was fixed in 2.063 (now ctor attributes should correspond to those in variable declarations).