31.07.14

mfbt now has UniquePtr and MakeUnique for managing singly-owned resources

Managing dynamic memory allocations in C++

C++ supports dynamic allocation of objects using new. For new objects to not leak, they must be deleted. This is quite difficult to do correctly in complex code. Smart pointers are the canonical solution. Mozilla has historically used nsAutoPtr, and C++98 provided std::auto_ptr, to manage singly-owned new objects. But nsAutoPtr and std::auto_ptr have a bug: they can be “copied.”

The following code allocates an int. When is that int destroyed? Does destroying ptr1 or ptr2 handle the task? What does ptr1 contain after ptr2‘s gone out of scope?

typedef auto_ptr<int> auto_int;
{
  auto_int ptr1(new int(17));
  {
    auto_int ptr2 = ptr1;
    // destroy ptr2
  }
  // destroy ptr1
}

Copying or assigning an auto_ptr implicitly moves the new object, mutating the input. When ptr2 = ptr1 happens, ptr1 is set to nullptr and ptr2 has a pointer to the allocated int. When ptr2 goes out of scope, it destroys the allocated int. ptr1 is nullptr when it goes out of scope, so destroying it does nothing.

Fixing auto_ptr

Implicit-move semantics are safe but very unclear. And because these operations mutate their input, they can’t take a const reference. For example, auto_ptr has an auto_ptr::auto_ptr(auto_ptr&) constructor but not an auto_ptr::auto_ptr(const auto_ptr&) copy constructor. This breaks algorithms requiring copyability.

We can solve these problems with a smart pointer that prohibits copying/assignment unless the input is a temporary value. (C++11 calls these rvalue references, but I’ll use “temporary value” for readability.) If the input’s a temporary value, we can move the resource out of it without disrupting anyone else’s view of it: as a temporary it’ll die before anyone could observe it. (The rvalue reference concept is incredibly subtle. Read that article series a dozen times, and maybe you’ll understand half of it. I’ve spent multiple full days digesting it and still won’t claim full understanding.)

Presenting mozilla::UniquePtr

I’ve implemented mozilla::UniquePtr in #include "mozilla/UniquePtr.h" to fit the bill. It’s based on C++11’s std::unique_ptr (not always available right now). UniquePtr provides auto_ptr‘s safety while providing movability but not copyability.

UniquePtr template parameters

Using UniquePtr requires the type being owned and what will ultimately be done to generically delete it. The type is the first template argument; the deleter is the (optional) second. The default deleter performs delete for non-array types and delete[] for array types. (This latter improves upon auto_ptr and nsAutoPtr [and the derivative nsAutoArrayPtr], which fail horribly when used with new[].)

UniquePtr<int> i1(new int(8));
UniquePtr<int[]> arr1(new int[17]());

Deleters are callable values, that are called whenever a UniquePtr‘s object should be destroyed. If a custom deleter is used, it’s a really good idea for it to be empty (per mozilla::IsEmpty<D>) so that UniquePtr<T, D> is as space-efficient as a raw pointer.

struct FreePolicy
{
  void operator()(void* ptr) {
    free(ptr);
  }
};

{
  void* m = malloc(4096);
  UniquePtr<void, FreePolicy> mem(m);
  int* i = static_cast<int*>(malloc(sizeof(int)));
  UniquePtr<int, FreePolicy> integer(i);

  // integer.getDeleter()(i) is called
  // mem.getDeleter()(m) is called
}

Basic UniquePtr construction and assignment

As you’d expect, no-argument construction initializes to nullptr, a single pointer initializes to that pointer, and a pointer and a deleter initialize embedded pointer and deleter both.

UniquePtr<int> i1;
assert(i1 == nullptr);
UniquePtr<int> i2(new int(8));
assert(i2 != nullptr);
UniquePtr<int, FreePolicy> i3(nullptr, FreePolicy());

Move construction and assignment

All remaining constructors and assignment operators accept only nullptr or compatible, temporary UniquePtr values. These values have well-defined ownership, in marked contrast to raw pointers.

class B
{
    int i;

  public:
    B(int i) : i(i) {}
    virtual ~B() {} // virtual required so delete (B*)(pointer to D) calls ~D()
};

class D : public B
{
  public:
    D(int i) : B(i) {}
};

UniquePtr<B> MakeB(int i)
{
  typedef UniquePtr<B>::DeleterType BDeleter;

  // OK to convert UniquePtr<D, BDeleter> to UniquePtr<B>:
  // Note: For UniquePtr interconversion, both pointer and deleter
  //       types must be compatible!  Thus BDeleter here.
  return UniquePtr<D, BDeleter>(new D(i));
}

UniquePtr<B> b1(MakeB(66)); // OK: temporary value moved into b1

UniquePtr<B> b2(b1); // ERROR: b1 not a temporary, would confuse
                     // single ownership, forbidden

UniquePtr<B> b3;

b3 = b1;  // ERROR: b1 not a temporary, would confuse
          // single ownership, forbidden

b3 = MakeB(76); // OK: return value moved into b3
b3 = nullptr;   // OK: can't confuse ownership of nullptr

What if you really do want to move a resource from one UniquePtr to another? You can explicitly request a move using mozilla::Move() from #include "mozilla/Move.h".

int* i = new int(37);
UniquePtr<int> i1(i);

UniquePtr<int> i2(Move(i1));
assert(i1 == nullptr);
assert(i2.get() == i);

i1 = Move(i2);
assert(i1.get() == i);
assert(i2 == nullptr);

Move transforms the type of its argument into a temporary value type. Move doesn’t have any effects of its own. Rather, it’s the job of users such as UniquePtr to ascribe special semantics to operations accepting temporary values. (If no special semantics are provided, temporary values match only const reference types as in C++98.)

Observing a UniquePtr‘s value

The dereferencing operators (-> and *) and conversion to bool behave as expected for any smart pointer. The raw pointer value can be accessed using get() if absolutely needed. (This should be uncommon, as the only pointer to the resource should live in the UniquePtr.) UniquePtr may also be compared against nullptr (but not against raw pointers).

int* i = new int(8);
UniquePtr<int> p(i);
if (p)
  *p = 42;
assert(p != nullptr);
assert(p.get() == i);
assert(*p == 42);

Changing a UniquePtr‘s value

Three mutation methods beyond assignment are available. A UniquePtr may be reset() to a raw pointer or to nullptr. The raw pointer may be extracted, and the UniquePtr cleared, using release(). Finally, UniquePtrs may be swapped.

int* i = new int(42);
int* i2;
UniquePtr<int> i3, i4;
{
  UniquePtr<int> integer(i);
  assert(i == integer.get());

  i2 = integer.release();
  assert(integer == nullptr);

  integer.reset(i2);
  assert(integer.get() == i2);

  integer.reset(new int(93)); // deletes i2

  i3 = Move(integer); // better than release()

  i3.swap(i4);
  Swap(i3, i4); // mozilla::Swap, that is
}

When a UniquePtr loses ownership of its resource, the embedded deleter will dispose of the managed pointer, in accord with the single-ownership concept. release() is the sole exception: it clears the UniquePtr and returns the raw pointer previously in it, without calling the deleter. This is a somewhat dangerous idiom. (Mozilla’s smart pointers typically call this forget(), and WebKit’s WTF calls this leak(). UniquePtr uses release() only for consistency with unique_ptr.) It’s generally much better to make the user take a UniquePtr, then transfer ownership using Move().

Array fillips

UniquePtr<T> and UniquePtr<T[]> share the same interface, with a few substantial differences. UniquePtr<T[]> defines an operator[] to permit indexing. As mentioned earlier, UniquePtr<T[]> by default will delete[] its resource, rather than delete it. As a corollary, UniquePtr<T[]> requires an exact type match when constructed or mutated using a pointer. (It’s an error to delete[] an array through a pointer to the wrong array element type, because delete[] has to know the element size to destruct each element. Not accepting other pointer types thus eliminates this class of errors.)

struct B {};
struct D : B {};
UniquePtr<B[]> bs;
// bs.reset(new D[17]()); // ERROR: requires B*, not D*
bs.reset(new B[5]());
bs[1] = B();

And a mozilla::MakeUnique helper function

Typing out new T every time a UniquePtr is created or initialized can get old. We’ve added a helper function, MakeUnique<T>, that combines new object (or array) creation with creation of a corresponding UniquePtr. The nice thing about MakeUnique is that it’s in some sense foolproof: if you only create new objects in UniquePtrs, you can’t leak or double-delete unless you leak the UniquePtr‘s owner, misuse a get(), or drop the result of release() on the floor. I recommend always using MakeUnique instead of new for single-ownership objects.

struct S { S(int i, double d) {} };

UniquePtr<S> s1 = MakeUnique<S>(17, 42.0);   // new S(17, 42.0)
UniquePtr<int> i1 = MakeUnique<int>(42);     // new int(42)
UniquePtr<int[]> i2 = MakeUnique<int[]>(17); // new int[17]()


// Given familiarity with UniquePtr, these work particularly
// well with C++11 auto: just recognize MakeUnique means new,
// T means single object, and T[] means array.
auto s2 = MakeUnique<S>(17, 42.0); // new S(17, 42.0)
auto i3 = MakeUnique<int>(42);     // new int(42)
auto i4 = MakeUnique<int[]>(17);   // new int[17]()

MakeUnique<T>(...args) computes new T(...args). MakeUnique of an array takes an array length and constructs the correspondingly-sized array.

In the long run we probably should expect everyone to recognize the MakeUnique idiom so that we can use auto here and cut down on redundant typing. In the short run, feel free to do whichever you prefer.

Update: Beware! Due to compiler limitations affecting gcc less than 4.6, passing literal nullptr as an argument to a MakeUnique call will fail to compile only on b2g-ics. Everywhere else will pass. You have been warned. The only alternative I can think of is to pass static_cast<T*>(nullptr) instead, or assign to a local variable and pass that instead. Love that b2g compiler!

Conclusion

UniquePtr was a free-time hacking project last Christmas week, that I mostly finished but ran out of steam on when work resumed. Only recently have I found time to finish it up and land it, yet we already have a couple hundred uses of it and MakeUnique. Please add more uses, and make our existing new code safer!

A final note: please use UniquePtr instead of mozilla::Scoped. UniquePtr is more standard, better-tested, and better-documented (particularly on the vast expanses of the web, where most unique_ptr documentation also suffices for UniquePtr). Scoped is now deprecated — don’t use it in new code!

08.12.11

Party like it’s 1999: <stdint.h> comes to Mozilla!

tl;dr

Need an integer type with guaranteed size? If you’re not defining a cross-file interface, #include "mozilla/StdInt.h"#include "mozilla/StandardInteger.h" and use uint32_t or any other type defined by <stdint.h>. (If you are defining an interface, use PRUint32 and similar — for now.) mozilla/StdInt.hmozilla/StandardInteger.h is a cross-platform implementation of <stdint.h>‘s functionality usable in any code.

Embedders may find that the mozilla/StdInt.hmozilla/StandardInteger.h typedefs conflict with ones they have already been using. To work around this conflict, write a stdint.h compatible with the embedding’s typedefs (more likely: adapt an existing implementation), then set the preprocessor variable MOZ_CUSTOM_STDINT_H to a quoted path to that reimplementation when mozilla/StdInt.hmozilla/StandardInteger.h is included. It may be simplest to add this to command line flags when invoking the compiler.

Fixed-size integer types

Fixed-size integer types are signed or unsigned types with exactly N bits. They contrast with the built-in C and C++ types (char, short, int, &c.) with compiler-dependent sizes. Fixed-size integer types are quite useful:

  • They work well when serializing an object to a sequence of bytes, where the size of a serialized item must be constant for correctness.
  • They minimize memory use in classes and structs, also making padding-based waste more obvious.
  • They work well in cross-platform APIs, eliminating the challenge of implementing correct behavior when types have different sizes across platforms.

Fixed-size integer types are useful in the same way size_t, off_t, and other non-built-in types are: they fit some problem domains better than built-in types.

C99 and C++11 finally standardized fixed-size integer types in <stdint.h> and <cstdint>. They define {u,}int{8,16,32,64}_t types, plus useful constants for their limits (INT8_MIN, INT8_MAX, UINT8_MAX, INT16_MIN, &c.). One would expect projects to quickly use these types, but it didn’t happen.

Old projects predating <stdint.h> have been particularly slow to adopt it. Many such projects already rolled their own non-<stdint.h>-named fixed-size integer types; switching would be a hassle. And not all compilers shipped <stdint.h>: Visual Studio didn’t have it until 2010! (ಠ_ಠ) Projects implementing their own <stdint.h>-compatible types posed another problem, because different projects’ implementations might be incompatible.

New projects fare better, but not always. Sometimes their dependence on old projects anchors them to the old, pre-<stdint.h> world.

Fixed-size integer types in Mozilla

Mozilla sits squarely in the old-project category, facing every rationale noted above for using its own types. These have long been NSPR‘s PR{Ui,I}nt{8,16,32,64} and SpiderMonkey’s {u,}int{8,16,32,64} types. But recently the landscape has changed.

As Mozilla has imported more external code, fixed-size integer types have proliferated. Most imported code uses <stdint.h>, with fallback typedefs for Visual Studio; some code (IPC code from Chromium) defines and uses {u,}int{8,16,32,64}. As type definitions have proliferated, surprising problems have arisen.

The woes of multiplicity

#include-order issues are the simplest problem. For example, the IPC uint32-style definitions are incompatible with SpiderMonkey’s definitions with some compilers. #include IPC and SpiderMonkey headers in the wrong order, and the compiler will error on incompatible typedefs. This problem is easy to diagnose but harder to resolve, and sometimes it causes considerable pain. Mass refactorings that add #includes have fallen afoul of this, with the least bad solution usually being to fix the first error, recompile, and repeat until done (twenty-odd times in one instance).

Worse than #include mis-ordering and conflict are linking problems. int and long could be 32-bit integers yet appear different when linked. Suppose a method taking an int32 argument defined int32 = int during compilation, but a user of it saw int32 = long during compilation. Each alone would compile. But beneath typedefs they’d be incompatible and wouldn’t link.

As we’ve imported more code in Mozilla, more and more developers have been bitten by these problems. We’ve reached a breaking point. We could use PRUint32, JSUint32, and other types which never trample upon each other. Yet no one likes them given the standardized types, and it’s not possible to change “upstream” code to such a scheme. Thus a second solution: use <stdint.h> definitions for everything.

Switching to <stdint.h>

Using the <stdint.h> types in Mozilla code is now as simple as #include "mozilla/StdInt.h"#include "mozilla/StandardInteger.h". mozilla/StdInt.hmozilla/StandardInteger.h implements the <stdint.h> interface even in the edge cases: for compilers not supporting it, and for embedders who can’t use the regular definitions. It works as follows:

  1. If the preprocessor definition MOZ_CUSTOM_STDINT_H is defined, then #include MOZ_CUSTOM_STDINT_H. Embedders who can’t use the default definitions should use this to adapt. (MOZ_CUSTOM_STDINT_H may also be passed into the Mozilla build system using an environment variable. Note that while the preprocessor definition must be a quoted path, the environment variable must be an unquoted path.)
  2. Otherwise, if the compiler doesn’t provide <stdint.h>, use a custom implementation. This is currently limited to Visual Studio prior to 2010, using an implementation imported from msinttypes on Google Code.
  3. Otherwise use <stdint.h>.

We’re only providing these types now, but shortly we’ll start switching code using non-<stdint.h> fixed-size integer types to use them. Adding mozilla/StdInt.hmozilla/StandardInteger.h is merely the first step toward removing the other fixed-size integer types (except when they’re necessary to interact with external libraries).

Conclusion

It’s been a dozen years since <stdint.h> was standardized. Now it finally comes to Mozilla. Let’s lower a barrier to hackability in Mozilla and start using it.

26.11.11

Introducing MOZ_FINAL: prevent inheriting from a class, or prevent overriding a virtual function

Tags: , , , , , , , — Jeff @ 09:17

The inexorable march of progress continues in the Mozilla Framework Based on Templates with the addition of MOZ_FINAL, through which you can limit various forms of inheritance in C++.

Traditional C++ inheritance mostly can’t be controlled

In C++98 any class can be inherited. (An extremely obscure pattern will prevent this, but it has down sides.) Sometimes this makes sense: it’s natural to subclass an abstract List class as LinkedList, or LinkedList as CircularLinkedList. But sometimes this doesn’t make sense. StringBuilder certainly could inherit from Vector<char>, but doing so might expose many Vector methods that don’t belong in the StringBuilder concept. It would be more sensible for StringBuilder to contain a private Vector<char> which StringBuilder member methods manipulated. Preventing Vector from being used as a base class would be one way (not necessarily the best one) to avoid this conceptual error. But C++98 doesn’t let you easily do that.

Even when inheritance is desired, sometimes you don’t want completely-virtual methods. Sometimes you’d like a class to implement a virtual method (virtual perhaps because its base declared it so) which derived classes can’t override. Perhaps you want to rely on that method being implemented only by your base class, or perhaps you want it “fixed” as an optimization. Again in C++98, you can’t do this: public virtual functions are overridable.

C++11’s contextual final keyword

C++11 introduces a contextual final keyword for these purposes. To prevent a class from being inheritable, add final to its definition just after the class name (the class can’t be unnamed).

struct Base1 final
{
  virtual void foo();
};

// ERROR: can't inherit from B.
struct Derived1 : public Base1 { };

struct Base2 { };

// Derived classes can be final too.
struct Derived2 final : public Base2 { };

Similarly, a virtual member function can be marked as not overridable by placing the contextual final keyword at the end of its declaration, before a terminating semicolon, body, or = 0.

struct Base
{
  virtual void foo() final;
};

struct Derived : public Base
{
  // ERROR: Base::foo was final.
  virtual void foo() { }
};

Introducing MOZ_FINAL

mfbt now includes support for marking classes and virtual member functions as final using the MOZ_FINAL macro in mozilla/Attributes.h. Simply place it in the same position as final would occur in the C++11 syntax:

#include "mozilla/Attributes.h"

class Base
{
  public:
    virtual void foo();
};

class Derived final : public Base
{
  public:
    /*
     * MOZ_FINAL and MOZ_OVERRIDE are composable; as a matter of
     * style, they should appear in the order MOZ_FINAL MOZ_OVERRIDE,
     * not the other way around.
     */
    virtual void foo() MOZ_FINAL MOZ_OVERRIDE { }
    virtual void bar() MOZ_FINAL;
    virtual void baz() MOZ_FINAL = 0;
};

MOZ_FINAL expands to the C++11 syntax or its compiler-specific equivalent whenever possible, turning violations of final semantics into compile-time errors. The same compilers that usefully expand MOZ_OVERRIDE also usefully expand MOZ_FINAL, so misuse will be quickly noted.

One interesting use for MOZ_FINAL is to tell the compiler that one worrisome C++ trick sometimes isn’t. This is the virtual-methods-without-virtual-destructor trick. It’s used when a class must have virtual functions, but code doesn’t want to pay the price of destruction having virtual-call overhead.

class Base
{
  public:
    virtual void method() { }
    ~Base() { }
};

void destroy(Base* b)
{
  delete b; // may cause a warning
}

Some compilers warn when they instantiate a class with virtual methods but without a virtual destructor. Other compilers only emit this warning when a pointer to a class instance is deleted. The reason is that in C++, behavior is undefined if the static type of the instance being deleted isn’t the same as its runtime type and its destructor isn’t virtual. In other words, if ~Base() is non-virtual, destroy(new Base) is perfectly fine, but destroy(new DerivedFromBase) is not. The warning makes sense if destruction might miss a base class — but if the class is marked final, it never will! Clang silences its warning if the class is final, and I hope that MSVC will shortly do the same.

What about NS_FINAL_CLASS?

As with MOZ_OVERRIDE we had a gunky XPCOM static-analysis NS_FINAL_CLASS macro for final classes. (We had no equivalent for final methods.) NS_FINAL_CLASS too was misplaced as far as C++11 syntax was concerned, and it too has been deprecated. Almost all uses of NS_FINAL_CLASS have now been removed (the one remaining use I’ve left for the moment due to an apparent Clang bug I haven’t tracked down yet), and it shouldn’t be used.

(Side note: In replacing NS_FINAL_CLASS with MOZ_FINAL, I discovered that some of the existing annotations have been buggy for months! Clearly no one’s done static analysis builds in awhile. The moral of the story: compiler static analyses that happen for every single build are vastly superior to user static analyses that happen only in special builds.)

Summary

If you don’t want a class to be inheritable, add MOZ_FINAL to its definition after the class name. If you don’t want a virtual member function to be overridden in derived classes, add MOZ_FINAL at the end of its declaration. Some compilers will then enforce your wishes, and you can rely on these requirements rather than hope for the best.

16.11.11

Introducing MOZ_OVERRIDE to annotate virtual functions which override base-class virtual functions

Tags: , , , , , , , — Jeff @ 09:55

Overriding inherited virtual functions

One way C++ supports code reuse is through inheritance. One base class implements common functionality. Then other classes inherit from it, essentially copying functionality from it. These other classes can add their own new functionality, or, more powerfully, they can override the base class functionality.

class Base
{
  public:
    virtual const char* type() { return "Base"; }
};
class Derived : public Base
{
  public:
    virtual const char* type() { return "Derived"; }
};

Overriding base class functionality is simple. Keeping such overrides working correctly is sometimes harder. The problem is that the override relationship is implicit: if the override doesn’t exactly match the signature of the desired function in the base class, it may not work correctly.

class Base
{
  public:
    // Perhaps as part of an incomplete refactoring,
    // the base class's function changed its name.
    virtual const char* kind() { return "Base"; }
};
class DerivedIncorrectly : public Base
{
  public:
    virtual const char* type() { return "Derived"; }
};

// BAD: code expecting kind() to work and sometimes
// indicate Derived-ness no longer will.

Making the override relationship explicit

Some languages (Scala, C#, probably others) provide the ability to mark a derived class’s function as an override of an inherited function. C++98 included no such ability, but C++11 does, through the contextual override keyword. When override is used, that virtual member function must override one found on a base class. If it does not, it is a compile error.

class Base
{
  public:
    virtual const char* kind() { return "Base"; }
};
class DerivedIncorrectly : public Base
{
  public:
    // This will cause a compile error: there's no type()
    // method on Base that this overrides.
    virtual const char* type() override { return "Derived"; }

    // This will work as intended.
    virtual const char* kind() override { return "Derived"; }
};

Introducing MOZ_OVERRIDE

The Mozilla Framework Based on Templates now includes support for the C++11 contextual override keyword, encapsulated in the MOZ_OVERRIDE macro in mozilla/Types.hmozilla/Attributes.h. Simply place it at the end of the declaration of the relevant method, before any = 0 or method body, like so:

#include "mozilla/Types.h" // MOZ_OVERRIDE has since moved...
#include "mozilla/Attributes.h" // ...to here

class Base
{
  public:
    virtual void f() = 0;
};
class Derived1 : public Base
{
  public:
    virtual void f() MOZ_OVERRIDE;
};
class Derived2 : public Base
{
  public:
    virtual void f() MOZ_OVERRIDE = 0;
};
class Derived3 : public Base
{
  public:
    virtual void f() MOZ_OVERRIDE { }
};

MOZ_OVERRIDE will expand to use the C++11 construct in compilers which support it. Thus in such compilers misuse of MOZ_OVERRIDE is an error. Even better, some of the compilers used by tinderbox support override, so in many cases tinderbox will detect misuse. (Specifically, MSVC++ 2005 and later support it, so errors in cross-platform and Windows code won’t pass tinderbox . Much more recent versions of GCC and Clang support it as well, but these versions are too new for tinderbox to have picked them up yet — in the case of GCC too new to even have been released yet. :-) )

What about NS_OVERRIDE?

It turns out there’s already a macro annotation to indicate an override relationship: NS_OVERRIDE. This gunky XPCOM macro expands to a user attribute under gcc-like compilers. It’s only used by static analysis right now, so its value is limited. Unfortunately its position is different — necessarily so, because in the C++11 override position it would attach to the return value of the method:

class OldAndBustedDerived : public Base
{
  public:
    NS_OVERRIDE virtual void f(); // annotates the method
    __attribute__(...) virtual void g(); // its expansion
};
class Derived2 : public Base
{
  public:
    // But in the MOZ_OVERRIDE position, it would annotate
    // f()'s return value.
    virtual void f() __attribute__(...);
};

NS_OVERRIDE is now deprecated and should be replaced with MOZ_OVERRIDE. With a little work, static analysis with new-enough compilers can likely look for MOZ_OVERRIDE just as easily as for NS_OVERRIDE. And since MOZ_OVERRIDE works in non-static analysis builds, it’s arguably better in the majority of cases anyway. If you’re looking for an easy way to improve Mozilla code, changing NS_OVERRIDE uses to use MOZ_OVERRIDE would be a simple way to help.

Summary

If you’ve overridden an inherited virtual member function and you’re worried that that override might silently break at some point, annotate your override with MOZ_OVERRIDE. This will cause some compilers to enforce an override relationship, making it much less likely that your intended relationship will break.

09.11.11

Introducing MOZ_DELETE, a macro improving upon the deliberately-unimplemented method idiom

C++ default operators and the sole-ownership idiom

Often a C++ class will solely manage some value: for example, a GtkWindow* or a void* for malloc‘d memory. The class will then release ownership in its destructor as appropriate. It would be extremely problematic to release ownership multiple times — think security-vulnerability-problematic. C++ copy construction and default assignment exacerbate this issue, because C++ automatically generates these methods for all classes, even when the default behavior breaks sole-ownership. The C++98 idiom solving this is to privately declare a copy constructor and a default assignment operator, then never define them:

struct Struct
{
  private:
    Struct(const Struct& other);
    void operator=(const Struct& other);
};

Declaring the methods privately prevents any code but friends of Struct from calling them. And by never defining them, even such friends will cause a link-time error if they try.

Disabling the default operators in C++11

Once you’re familiar with this idiom it’s not too bad. But initially, it’s pretty unclear. And nothing prevents someone from actually defining these methods. (They could only be used by Struct or friends of Struct, to be sure, but for sufficiently complex code it’s possible someone might make a mistake.) C++11 improves upon this trick by introducing deleted function syntax:

struct Struct
{
  private: // no longer necessary, but doesn't hurt
    Struct(const Struct& other) = delete;
    void operator=(const Struct& other) = delete;
};

Deleted functions are effectively removed from name lookup: using, defining, or referring to a deleted function produces a compile error — far better than a link error or, even worse, no error at all.

= delete support in mfbt

The Mozilla Framework Based on Templates now includes support for declaring a function only to prevent its use (or use of an inherited version). The MOZ_DELETE macro encapsulates this support:

#include "mozilla/Types.h" // MOZ_DELETE has since moved...
#include "mozilla/Attributes.h" // ...to here

struct Struct
{
  private:
    Struct(const Struct& other) MOZ_DELETE;
    void operator=(const Struct& other) MOZ_DELETE;
};

MOZ_DELETE isn’t as readable or understandable as = delete, but it’s searchable, and the comment next to its definition will clarify matters. If the declarations are private, MOZ_DELETE is just as good as the traditional idiom, and in compilers supporting C++11 deleted functions it’s better.

Which compilers support C++11 deleted functions? I’m aware of GCC since 4.4, Clang since 2.9, and ICC since 12.0. Rightly, if unfortunately, you must specify -std=c++0x or similar to use deleted function syntax without causing a warning. For various reasons Mozilla can’t do that yet, so MOZ_DELETE only produces the C++11 syntax when compiling with Clang (where we can pass -Wno-c++0x-extensions to disable the warning). I’d love to see it use C++11 syntax in GCC and ICC as well, but I don’t have the time to solve the -std=c++0x problem now, or to figure out another workaround. I’ve filed bug 701183 for this problem — help there is much appreciated.

Summary

Use MOZ_DELETE when declaring any method you will intentionally not implement. It’ll work better, and produce better errors, in some compilers. Those compilers don’t include GCC or ICC yet, but with your help they could. Any takers?

Update, evening of November 10, 2011: I just landed further changes to make MOZ_DELETE use C++11 syntax with GCC when compiling with -std=c++0x (which we apparently do more often than I’d thought), so you should now get its goodness in GCC as well — most of the time. In some “exotic” situations we don’t compile anything with -std=c++0x, so you won’t get any benefit there. Also, the JavaScript engine is never compiled with it. So if you want this to work fully, everywhere, you should use Clang.

Older »