Type Alias mozjs::jsapi::js::detail::UniqueSelector_SingleObject

pub type UniqueSelector_SingleObject = u8;

UniquePtr is a smart pointer that wholly owns a resource. Ownership may be transferred out of a UniquePtr through explicit action, but otherwise the resource is destroyed when the UniquePtr is destroyed.

UniquePtr is similar to C++98’s std::auto_ptr, but it improves upon auto_ptr in one crucial way: it’s impossible to copy a UniquePtr. Copying an auto_ptr obviously can’t copy ownership of its singly-owned resource. So what happens if you try to copy one? Bizarrely, ownership is implicitly transferred, preserving single ownership but breaking code that assumes a copy of an object is identical to the original. (This is why auto_ptr is prohibited in STL containers.)

UniquePtr solves this problem by being movable rather than copyable. Instead of passing a |UniquePtr u| directly to the constructor or assignment operator, you pass |Move(u)|. In doing so you indicate that you’re moving ownership out of |u|, into the target of the construction/assignment. After the transfer completes, |u| contains |nullptr| and may be safely destroyed. This preserves single ownership but also allows UniquePtr to be moved by algorithms that have been made move-safe. (Note: if |u| is instead a temporary expression, don’t use |Move()|: just pass the expression, because it’s already move-ready. For more information see Move.h.)

UniquePtr is also better than std::auto_ptr in that the deletion operation is customizable. An optional second template parameter specifies a class that (through its operator()(T*)) implements the desired deletion policy. If no policy is specified, mozilla::DefaultDelete is used – which will either |delete| or |delete[]| the resource, depending whether the resource is an array. Custom deletion policies ideally should be empty classes (no member fields, no member fields in base classes, no virtual methods/inheritance), because then UniquePtr can be just as efficient as a raw pointer.

Use of UniquePtr proceeds like so:

UniquePtr g1; // initializes to nullptr g1.reset(new int); // switch resources using reset() g1 = nullptr; // clears g1, deletes the int

UniquePtr g2(new int); // owns that int int* p = g2.release(); // g2 leaks its int – still requires deletion delete p; // now freed

struct S { int x; S(int x) : x(x) {} }; UniquePtr g3, g4(new S(5)); g3 = std::move(g4); // g3 owns the S, g4 cleared S* p = g3.get(); // g3 still owns |p| assert(g3->x == 5); // operator-> works (if .get() != nullptr) assert((g3).x == 5); // also operator (again, if not cleared) std::swap(g3, g4); // g4 now owns the S, g3 cleared g3.swap(g4); // g3 now owns the S, g4 cleared UniquePtr g5(std::move(g3)); // g5 owns the S, g3 cleared g5.reset(); // deletes the S, g5 cleared

struct FreePolicy { void operator()(void* p) { free(p); } }; UniquePtr<int, FreePolicy> g6(static_cast<int*>(malloc(sizeof(int)))); int* ptr = g6.get(); g6 = nullptr; // calls free(ptr)

Now, carefully note a few things you can’t do:

UniquePtr b1; b1 = new int; // BAD: can only assign another UniquePtr int* ptr = b1; // BAD: no auto-conversion to pointer, use get()

UniquePtr b2(b1); // BAD: can’t copy a UniquePtr UniquePtr b3 = b1; // BAD: can’t copy-assign a UniquePtr

(Note that changing a UniquePtr to store a direct |new| expression is permitted, but usually you should use MakeUnique, defined at the end of this header.)

A few miscellaneous notes:

UniquePtr, when not instantiated for an array type, can be move-constructed and move-assigned, not only from itself but from “derived” UniquePtr<U, E> instantiations where U converts to T and E converts to D. If you want to use this, you’re going to have to specify a deletion policy for both UniquePtr instantations, and T pretty much has to have a virtual destructor. In other words, this doesn’t work:

struct Base { virtual ~Base() {} }; struct Derived : Base {};

UniquePtr b1; // BAD: DefaultDelete and DefaultDelete don’t interconvert UniquePtr d1(std::move(b));

UniquePtr b2; UniquePtr<Derived, DefaultDelete> d2(std::move(b2)); // okay

UniquePtr is specialized for array types. Specializing with an array type creates a smart-pointer version of that array – not a pointer to such an array.

UniquePtr<int[]> arr(new int[5]); arr[0] = 4;

What else is different? Deletion of course uses |delete[]|. An operator[] is provided. Functionality that doesn’t make sense for arrays is removed. The constructors and mutating methods only accept array pointers (not T*, U* that converts to T*, or UniquePtr<U[]> or UniquePtr) or |nullptr|.

It’s perfectly okay for a function to return a UniquePtr. This transfers the UniquePtr’s sole ownership of the data, to the fresh UniquePtr created in the calling function, that will then solely own that data. Such functions can return a local variable UniquePtr, |nullptr|, |UniquePtr(ptr)| where |ptr| is a |T*|, or a UniquePtr |Move()|’d from elsewhere.

UniquePtr will commonly be a member of a class, with lifetime equivalent to that of that class. If you want to expose the related resource, you could expose a raw pointer via |get()|, but ownership of a raw pointer is inherently unclear. So it’s better to expose a |const UniquePtr&| instead. This prohibits mutation but still allows use of |get()| when needed (but operator-> is preferred). Of course, you can only use this smart pointer as long as the enclosing class instance remains live – no different than if you exposed the |get()| raw pointer.

To pass a UniquePtr-managed resource as a pointer, use a |const UniquePtr&| argument. To specify an inout parameter (where the method may or may not take ownership of the resource, or reset it), or to specify an out parameter (where simply returning a |UniquePtr| isn’t possible), use a |UniquePtr&| argument. To unconditionally transfer ownership of a UniquePtr into a method, use a |UniquePtr| argument. To conditionally transfer ownership of a resource into a method, should the method want it, use a |UniquePtr&&| argument.