michael@0: /* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
michael@0: /* vim: set ts=8 sts=2 et sw=2 tw=80: */
michael@0: /* This Source Code Form is subject to the terms of the Mozilla Public
michael@0:  * License, v. 2.0. If a copy of the MPL was not distributed with this
michael@0:  * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
michael@0: 
michael@0: /* C++11-style, but C++98-usable, "move references" implementation. */
michael@0: 
michael@0: #ifndef mozilla_Move_h
michael@0: #define mozilla_Move_h
michael@0: 
michael@0: #include "mozilla/TypeTraits.h"
michael@0: 
michael@0: namespace mozilla {
michael@0: 
michael@0: /*
michael@0:  * "Move" References
michael@0:  *
michael@0:  * Some types can be copied much more efficiently if we know the original's
michael@0:  * value need not be preserved --- that is, if we are doing a "move", not a
michael@0:  * "copy". For example, if we have:
michael@0:  *
michael@0:  *   Vector<T> u;
michael@0:  *   Vector<T> v(u);
michael@0:  *
michael@0:  * the constructor for v must apply a copy constructor to each element of u ---
michael@0:  * taking time linear in the length of u. However, if we know we will not need u
michael@0:  * any more once v has been initialized, then we could initialize v very
michael@0:  * efficiently simply by stealing u's dynamically allocated buffer and giving it
michael@0:  * to v --- a constant-time operation, regardless of the size of u.
michael@0:  *
michael@0:  * Moves often appear in container implementations. For example, when we append
michael@0:  * to a vector, we may need to resize its buffer. This entails moving each of
michael@0:  * its extant elements from the old, smaller buffer to the new, larger buffer.
michael@0:  * But once the elements have been migrated, we're just going to throw away the
michael@0:  * old buffer; we don't care if they still have their values. So if the vector's
michael@0:  * element type can implement "move" more efficiently than "copy", the vector
michael@0:  * resizing should by all means use a "move" operation. Hash tables should also
michael@0:  * use moves when resizing their internal array as entries are added and
michael@0:  * removed.
michael@0:  *
michael@0:  * The details of the optimization, and whether it's worth applying, vary
michael@0:  * from one type to the next: copying an 'int' is as cheap as moving it, so
michael@0:  * there's no benefit in distinguishing 'int' moves from copies. And while
michael@0:  * some constructor calls for complex types are moves, many really have to
michael@0:  * be copies, and can't be optimized this way. So we need:
michael@0:  *
michael@0:  * 1) a way for a type (like Vector) to announce that it can be moved more
michael@0:  *    efficiently than it can be copied, and provide an implementation of that
michael@0:  *    move operation; and
michael@0:  *
michael@0:  * 2) a way for a particular invocation of a copy constructor to say that it's
michael@0:  *    really a move, not a copy, and that the value of the original isn't
michael@0:  *    important afterwards (although it must still be safe to destroy).
michael@0:  *
michael@0:  * If a constructor has a single argument of type 'T&&' (an 'rvalue reference
michael@0:  * to T'), that indicates that it is a 'move constructor'. That's 1). It should
michael@0:  * move, not copy, its argument into the object being constructed. It may leave
michael@0:  * the original in any safely-destructible state.
michael@0:  *
michael@0:  * If a constructor's argument is an rvalue, as in 'C(f(x))' or 'C(x + y)', as
michael@0:  * opposed to an lvalue, as in 'C(x)', then overload resolution will prefer the
michael@0:  * move constructor, if there is one. The 'mozilla::Move' function, defined in
michael@0:  * this file, is an identity function you can use in a constructor invocation to
michael@0:  * make any argument into an rvalue, like this: C(Move(x)). That's 2). (You
michael@0:  * could use any function that works, but 'Move' indicates your intention
michael@0:  * clearly.)
michael@0:  *
michael@0:  * Where we might define a copy constructor for a class C like this:
michael@0:  *
michael@0:  *   C(const C& rhs) { ... copy rhs to this ... }
michael@0:  *
michael@0:  * we would declare a move constructor like this:
michael@0:  *
michael@0:  *   C(C&& rhs) { .. move rhs to this ... }
michael@0:  *
michael@0:  * And where we might perform a copy like this:
michael@0:  *
michael@0:  *   C c2(c1);
michael@0:  *
michael@0:  * we would perform a move like this:
michael@0:  *
michael@0:  *   C c2(Move(c1));
michael@0:  *
michael@0:  * Note that 'T&&' implicitly converts to 'T&'. So you can pass a 'T&&' to an
michael@0:  * ordinary copy constructor for a type that doesn't support a special move
michael@0:  * constructor, and you'll just get a copy. This means that templates can use
michael@0:  * Move whenever they know they won't use the original value any more, even if
michael@0:  * they're not sure whether the type at hand has a specialized move constructor.
michael@0:  * If it doesn't, the 'T&&' will just convert to a 'T&', and the ordinary copy
michael@0:  * constructor will apply.
michael@0:  *
michael@0:  * A class with a move constructor can also provide a move assignment operator.
michael@0:  * A generic definition would run this's destructor, and then apply the move
michael@0:  * constructor to *this's memory. A typical definition:
michael@0:  *
michael@0:  *   C& operator=(C&& rhs) {
michael@0:  *     MOZ_ASSERT(&rhs != this, "self-moves are prohibited");
michael@0:  *     this->~C();
michael@0:  *     new(this) C(Move(rhs));
michael@0:  *     return *this;
michael@0:  *   }
michael@0:  *
michael@0:  * With that in place, one can write move assignments like this:
michael@0:  *
michael@0:  *   c2 = Move(c1);
michael@0:  *
michael@0:  * This destroys c2, moves c1's value to c2, and leaves c1 in an undefined but
michael@0:  * destructible state.
michael@0:  *
michael@0:  * As we say, a move must leave the original in a "destructible" state. The
michael@0:  * original's destructor will still be called, so if a move doesn't
michael@0:  * actually steal all its resources, that's fine. We require only that the
michael@0:  * move destination must take on the original's value; and that destructing
michael@0:  * the original must not break the move destination.
michael@0:  *
michael@0:  * (Opinions differ on whether move assignment operators should deal with move
michael@0:  * assignment of an object onto itself. It seems wise to either handle that
michael@0:  * case, or assert that it does not occur.)
michael@0:  *
michael@0:  * Forwarding:
michael@0:  *
michael@0:  * Sometimes we want copy construction or assignment if we're passed an ordinary
michael@0:  * value, but move construction if passed an rvalue reference. For example, if
michael@0:  * our constructor takes two arguments and either could usefully be a move, it
michael@0:  * seems silly to write out all four combinations:
michael@0:  *
michael@0:  *   C::C(X&  x, Y&  y) : x(x),       y(y)       { }
michael@0:  *   C::C(X&  x, Y&& y) : x(x),       y(Move(y)) { }
michael@0:  *   C::C(X&& x, Y&  y) : x(Move(x)), y(y)       { }
michael@0:  *   C::C(X&& x, Y&& y) : x(Move(x)), y(Move(y)) { }
michael@0:  *
michael@0:  * To avoid this, C++11 has tweaks to make it possible to write what you mean.
michael@0:  * The four constructor overloads above can be written as one constructor
michael@0:  * template like so[0]:
michael@0:  *
michael@0:  *   template <typename XArg, typename YArg>
michael@0:  *   C::C(XArg&& x, YArg&& y) : x(Forward<XArg>(x)), y(Forward<YArg>(y)) { }
michael@0:  *
michael@0:  * ("'Don't Repeat Yourself'? What's that?")
michael@0:  *
michael@0:  * This takes advantage of two new rules in C++11:
michael@0:  *
michael@0:  * - First, when a function template takes an argument that is an rvalue
michael@0:  *   reference to a template argument (like 'XArg&& x' and 'YArg&& y' above),
michael@0:  *   then when the argument is applied to an lvalue, the template argument
michael@0:  *   resolves to 'T &'; and when it is applied to an rvalue, the template
michael@0:  *   argument resolves to 'T &&'. Thus, in a call to C::C like:
michael@0:  *
michael@0:  *      X foo(int);
michael@0:  *      Y yy;
michael@0:  *
michael@0:  *      C(foo(5), yy)
michael@0:  *
michael@0:  *   XArg would resolve to 'X&&', and YArg would resolve to 'Y&'.
michael@0:  *
michael@0:  * - Second, Whereas C++ used to forbid references to references, C++11 defines
michael@0:  *   'collapsing rules': 'T& &', 'T&& &', and 'T& &&' (that is, any combination
michael@0:  *   involving an lvalue reference) now collapse to simply 'T&'; and 'T&& &&'
michael@0:  *   collapses to 'T&&'.
michael@0:  *
michael@0:  *   Thus, in the call above, 'XArg&&' is 'X&& &&', collapsing to 'X&&'; and
michael@0:  *   'YArg&&' is 'Y& &&', which collapses to 'Y &'. Because the arguments are
michael@0:  *   declared as rvalue references to template arguments, the rvalue-ness
michael@0:  *   "shines through" where present.
michael@0:  *
michael@0:  * Then, the 'Forward<T>' function --- you must invoke 'Forward' with its type
michael@0:  * argument --- returns an lvalue reference or an rvalue reference to its
michael@0:  * argument, depending on what T is. In our unified constructor definition, that
michael@0:  * means that we'll invoke either the copy or move constructors for x and y,
michael@0:  * depending on what we gave C's constructor. In our call, we'll move 'foo()'
michael@0:  * into 'x', but copy 'yy' into 'y'.
michael@0:  *
michael@0:  * This header file defines Move and Forward in the mozilla namespace. It's up
michael@0:  * to individual containers to annotate moves as such, by calling Move; and it's
michael@0:  * up to individual types to define move constructors and assignment operators
michael@0:  * when valuable.
michael@0:  *
michael@0:  * (C++11 says that the <utility> header file should define 'std::move' and
michael@0:  * 'std::forward', which are just like our 'Move' and 'Forward'; but those
michael@0:  * definitions aren't available in that header on all our platforms, so we
michael@0:  * define them ourselves here.)
michael@0:  *
michael@0:  * 0. This pattern is known as "perfect forwarding".  Interestingly, it is not
michael@0:  *    actually perfect, and it can't forward all possible argument expressions!
michael@0:  *    There are two issues: one that's a C++11 issue, and one that's a legacy
michael@0:  *    compiler issue.
michael@0:  *
michael@0:  *    The C++11 issue is that you can't form a reference to a bit-field.  As a
michael@0:  *    workaround, assign the bit-field to a local variable and use that:
michael@0:  *
michael@0:  *      // C is as above
michael@0:  *      struct S { int x : 1; } s;
michael@0:  *      C(s.x, 0); // BAD: s.x is a reference to a bit-field, can't form those
michael@0:  *      int tmp = s.x;
michael@0:  *      C(tmp, 0); // OK: tmp not a bit-field
michael@0:  *
michael@0:  *    The legacy issue is that when we don't have true nullptr and must emulate
michael@0:  *    it (gcc 4.4/4.5), forwarding |nullptr| results in an |int| or |long|
michael@0:  *    forwarded reference.  But such a reference, even if its value is a null
michael@0:  *    pointer constant expression, is not itself a null pointer constant
michael@0:  *    expression.  This causes -Werror=conversion-null errors and pointer-to-
michael@0:  *    integer comparison errors.  Until we always have true nullptr, users of
michael@0:  *    forwarding methods must not pass |nullptr| to them.
michael@0:  */
michael@0: 
michael@0: /**
michael@0:  * Identical to std::Move(); this is necessary until our stlport supports
michael@0:  * std::move().
michael@0:  */
michael@0: template<typename T>
michael@0: inline typename RemoveReference<T>::Type&&
michael@0: Move(T&& a)
michael@0: {
michael@0:   return static_cast<typename RemoveReference<T>::Type&&>(a);
michael@0: }
michael@0: 
michael@0: /**
michael@0:  * These two overloads are identical to std::forward(); they are necessary until
michael@0:  * our stlport supports std::forward().
michael@0:  */
michael@0: template<typename T>
michael@0: inline T&&
michael@0: Forward(typename RemoveReference<T>::Type& a)
michael@0: {
michael@0:   return static_cast<T&&>(a);
michael@0: }
michael@0: 
michael@0: template<typename T>
michael@0: inline T&&
michael@0: Forward(typename RemoveReference<T>::Type&& t)
michael@0: {
michael@0:   static_assert(!IsLvalueReference<T>::value,
michael@0:                 "misuse of Forward detected!  try the other overload");
michael@0:   return static_cast<T&&>(t);
michael@0: }
michael@0: 
michael@0: /** Swap |t| and |u| using move-construction if possible. */
michael@0: template<typename T>
michael@0: inline void
michael@0: Swap(T& t, T& u)
michael@0: {
michael@0:   T tmp(Move(t));
michael@0:   t = Move(u);
michael@0:   u = Move(tmp);
michael@0: }
michael@0: 
michael@0: } // namespace mozilla
michael@0: 
michael@0: #endif /* mozilla_Move_h */