// Copyright 2011 The Chromium Authors // Use of this source code is governed by a BSD-style license that can be // found in the LICENSE file. #ifndef BASE_FUNCTIONAL_BIND_H_ #define BASE_FUNCTIONAL_BIND_H_ #include <functional> #include <memory> #include <type_traits> #include <utility> #include "base/compiler_specific.h" #include "base/functional/bind_internal.h" // IWYU pragma: export #include "base/memory/raw_ptr.h" #include "build/build_config.h" // ----------------------------------------------------------------------------- // Usage documentation // ----------------------------------------------------------------------------- // // Overview: // base::BindOnce() and base::BindRepeating() are helpers for creating // base::OnceCallback and base::RepeatingCallback objects respectively. // // For a runnable object of n-arity, the base::Bind*() family allows partial // application of the first m arguments. The remaining n - m arguments must be // passed when invoking the callback with Run(). // // // The first argument is bound at callback creation; the remaining // // two must be passed when calling Run() on the callback object. // base::OnceCallback<long(int, long)> cb = base::BindOnce( // [](short x, int y, long z) { return x * y * z; }, 42); // // When binding to a method, the receiver object must also be specified at // callback creation time. When Run() is invoked, the method will be invoked on // the specified receiver object. // // class C : public base::RefCounted<C> { void F(); }; // auto instance = base::MakeRefCounted<C>(); // auto cb = base::BindOnce(&C::F, instance); // std::move(cb).Run(); // Identical to instance->F() // // See //docs/callback.md for the full documentation. // // ----------------------------------------------------------------------------- // Implementation notes // ----------------------------------------------------------------------------- // // If you're reading the implementation, before proceeding further, you should // read the top comment of base/functional/bind_internal.h for a definition of // common terms and concepts. namespace base { // Bind as OnceCallback. template <typename Functor, typename... Args> inline auto BindOnce(Functor&& functor, Args&&... args) { … } // Bind as RepeatingCallback. template <typename Functor, typename... Args> inline auto BindRepeating(Functor&& functor, Args&&... args) { … } // Overloads to allow nicer compile errors when attempting to pass the address // an overloaded function to `BindOnce()` or `BindRepeating()`. Otherwise, clang // provides only the error message "no matching function [...] candidate // template ignored: couldn't infer template argument 'Functor'", with no // reference to the fact that `&` is being used on an overloaded function. // // These overloads to provide better error messages will never be selected // unless template type deduction fails because of how overload resolution // works; per [over.ics.rank/2.2]: // // When comparing the basic forms of implicit conversion sequences (as defined // in [over.best.ics]) // - a standard conversion sequence is a better conversion sequence than a // user-defined conversion sequence or an ellipsis conversion sequence, and // - a user-defined conversion sequence is a better conversion sequence than // an ellipsis conversion sequence. // // So these overloads will only be selected as a last resort iff template type // deduction fails. BindFailedCheckPreviousErrors BindOnce(...); BindFailedCheckPreviousErrors BindRepeating(...); // Unretained(), UnsafeDangling() and UnsafeDanglingUntriaged() allow binding a // non-refcounted class, and to disable refcounting on arguments that are // refcounted. The main difference is whether or not the raw pointers will be // checked for dangling references (e.g. a pointer that points to an already // destroyed object) when the callback is run. // // It is _required_ to use one of Unretained(), UnsafeDangling() or // UnsafeDanglingUntriaged() for raw pointer receivers now. For other arguments, // it remains optional. If not specified, default behavior is Unretained(). // Unretained() pointers will be checked for dangling pointers when the // callback is run, *if* the callback has not been cancelled. // // Example of Unretained() usage: // // class Foo { // public: // void func() { cout << "Foo:f" << endl; } // }; // // // In some function somewhere. // Foo foo; // OnceClosure foo_callback = // BindOnce(&Foo::func, Unretained(&foo)); // std::move(foo_callback).Run(); // Prints "Foo:f". // // Without the Unretained() wrapper on |&foo|, the above call would fail // to compile because Foo does not support the AddRef() and Release() methods. // // Unretained() does not allow dangling pointers, e.g.: // class MyClass { // public: // OnError(int error); // private: // scoped_refptr<base::TaskRunner> runner_; // std::unique_ptr<AnotherClass> obj_; // }; // // void MyClass::OnError(int error) { // // the pointer (which is also the receiver here) to `AnotherClass` // // might dangle depending on when the task is invoked. // runner_->PostTask(FROM_HERE, base::BindOnce(&AnotherClass::OnError, // base::Unretained(obj_.get()), error)); // // one of the way to solve this issue here would be: // // runner_->PostTask(FROM_HERE, // // base::BindOnce(&AnotherClass::OnError, // // base::Owned(std::move(obj_)), error)); // delete this; // } // // the above example is a BAD USAGE of Unretained(), which might result in a // use-after-free, as `AnotherClass::OnError` might be invoked with a dangling // pointer as receiver. template <typename T> inline auto Unretained(T* o) { … } template <typename T, RawPtrTraits Traits> inline auto Unretained(const raw_ptr<T, Traits>& o) { … } template <typename T, RawPtrTraits Traits> inline auto Unretained(raw_ptr<T, Traits>&& o) { … } template <typename T, RawPtrTraits Traits> inline auto Unretained(const raw_ref<T, Traits>& o) { … } template <typename T, RawPtrTraits Traits> inline auto Unretained(raw_ref<T, Traits>&& o) { … } // Similar to `Unretained()`, but allows dangling pointers, e.g.: // // class MyClass { // public: // DoSomething(HandlerClass* handler); // private: // void MyClass::DoSomethingInternal(HandlerClass::Id id, // HandlerClass* handler); // // std::unordered_map<HandlerClass::Id, HandlerClass*> handlers_; // scoped_refptr<base::SequencedTaskRunner> runner_; // base::Lock lock_; // }; // void MyClass::DoSomething(HandlerClass* handler) { // runner_->PostTask(FROM_HERE, // base::BindOnce(&MyClass::DoSomethingInternal, // base::Unretained(this), // handler->id(), // base::Unretained(handler))); // } // void MyClass::DoSomethingInternal(HandlerClass::Id id, // HandlerClass* handler) { // base::AutoLock locker(lock_); // if (handlers_.find(id) == std::end(handlers_)) return; // // Now we can use `handler`. // } // // As `DoSomethingInternal` is run on a sequence (and we can imagine // `handlers_` being modified on it as well), we protect the function from // using a dangling `handler` by making sure it is still contained in the // map. // // Strongly prefer `Unretained()`. This is useful in limited situations such as // the one above. // // When using `UnsafeDangling()`, the receiver must be of type MayBeDangling<>. template <typename T> inline auto UnsafeDangling(T* o) { … } template <typename T, RawPtrTraits Traits> auto UnsafeDangling(const raw_ptr<T, Traits>& o) { … } template <typename T, RawPtrTraits Traits> auto UnsafeDangling(raw_ptr<T, Traits>&& o) { … } template <typename T, RawPtrTraits Traits> auto UnsafeDangling(const raw_ref<T, Traits>& o) { … } template <typename T, RawPtrTraits Traits> auto UnsafeDangling(raw_ref<T, Traits>&& o) { … } // Like `UnsafeDangling()`, but used to annotate places that still need to be // triaged and either migrated to `Unretained()` and safer ownership patterns // (preferred) or `UnsafeDangling()` if the correct pattern to use is the one // in the `UnsafeDangling()` example above for example. // // Unlike `UnsafeDangling()`, the receiver doesn't have to be MayBeDangling<>. template <typename T> inline auto UnsafeDanglingUntriaged(T* o) { … } template <typename T, RawPtrTraits Traits> auto UnsafeDanglingUntriaged(const raw_ptr<T, Traits>& o) { … } template <typename T, RawPtrTraits Traits> auto UnsafeDanglingUntriaged(raw_ptr<T, Traits>&& o) { … } template <typename T, RawPtrTraits Traits> auto UnsafeDanglingUntriaged(const raw_ref<T, Traits>& o) { … } template <typename T, RawPtrTraits Traits> auto UnsafeDanglingUntriaged(raw_ref<T, Traits>&& o) { … } // RetainedRef() accepts a ref counted object and retains a reference to it. // When the callback is called, the object is passed as a raw pointer. // // EXAMPLE OF RetainedRef(): // // void foo(RefCountedBytes* bytes) {} // // scoped_refptr<RefCountedBytes> bytes = ...; // OnceClosure callback = BindOnce(&foo, base::RetainedRef(bytes)); // std::move(callback).Run(); // // Without RetainedRef, the scoped_refptr would try to implicitly convert to // a raw pointer and fail compilation: // // OnceClosure callback = BindOnce(&foo, bytes); // ERROR! template <typename T> inline internal::RetainedRefWrapper<T> RetainedRef(T* o) { … } template <typename T> inline internal::RetainedRefWrapper<T> RetainedRef(scoped_refptr<T> o) { … } // Owned() transfers ownership of an object to the callback resulting from // bind; the object will be deleted when the callback is deleted. // // EXAMPLE OF Owned(): // // void foo(int* arg) { cout << *arg << endl } // // int* pn = new int(1); // RepeatingClosure foo_callback = BindRepeating(&foo, Owned(pn)); // // foo_callback.Run(); // Prints "1" // foo_callback.Run(); // Prints "1" // *pn = 2; // foo_callback.Run(); // Prints "2" // // foo_callback.Reset(); // |pn| is deleted. Also will happen when // // |foo_callback| goes out of scope. // // Without Owned(), someone would have to know to delete |pn| when the last // reference to the callback is deleted. template <typename T> inline internal::OwnedWrapper<T> Owned(T* o) { … } template <typename T, typename Deleter> inline internal::OwnedWrapper<T, Deleter> Owned( std::unique_ptr<T, Deleter>&& ptr) { … } // OwnedRef() stores an object in the callback resulting from // bind and passes a reference to the object to the bound function. // // EXAMPLE OF OwnedRef(): // // void foo(int& arg) { cout << ++arg << endl } // // int counter = 0; // RepeatingClosure foo_callback = BindRepeating(&foo, OwnedRef(counter)); // // foo_callback.Run(); // Prints "1" // foo_callback.Run(); // Prints "2" // foo_callback.Run(); // Prints "3" // // cout << counter; // Prints "0", OwnedRef creates a copy of counter. // // Supports OnceCallbacks as well, useful to pass placeholder arguments: // // void bar(int& ignore, const std::string& s) { cout << s << endl } // // OnceClosure bar_callback = BindOnce(&bar, OwnedRef(0), "Hello"); // // std::move(bar_callback).Run(); // Prints "Hello" // // Without OwnedRef() it would not be possible to pass a mutable reference to an // object owned by the callback. template <typename T> internal::OwnedRefWrapper<std::decay_t<T>> OwnedRef(T&& t) { … } // Passed() is for transferring movable-but-not-copyable types (eg. unique_ptr) // through a RepeatingCallback. Logically, this signifies a destructive transfer // of the state of the argument into the target function. Invoking // RepeatingCallback::Run() twice on a callback that was created with a Passed() // argument will CHECK() because the first invocation would have already // transferred ownership to the target function. // // Note that Passed() is not necessary with BindOnce(), as std::move() does the // same thing. Avoid Passed() in favor of std::move() with BindOnce(). // // EXAMPLE OF Passed(): // // void TakesOwnership(std::unique_ptr<Foo> arg) { } // std::unique_ptr<Foo> CreateFoo() { return std::make_unique<Foo>(); // } // // auto f = std::make_unique<Foo>(); // // // |cb| is given ownership of Foo(). |f| is now NULL. // // You can use std::move(f) in place of &f, but it's more verbose. // RepeatingClosure cb = BindRepeating(&TakesOwnership, Passed(&f)); // // // Run was never called so |cb| still owns Foo() and deletes // // it on Reset(). // cb.Reset(); // // // |cb| is given a new Foo created by CreateFoo(). // cb = BindRepeating(&TakesOwnership, Passed(CreateFoo())); // // // |arg| in TakesOwnership() is given ownership of Foo(). |cb| // // no longer owns Foo() and, if reset, would not delete Foo(). // cb.Run(); // Foo() is now transferred to |arg| and deleted. // cb.Run(); // This CHECK()s since Foo() already been used once. // // We offer 2 syntaxes for calling Passed(). The first takes an rvalue and is // best suited for use with the return value of a function or other temporary // rvalues. The second takes a pointer to the scoper and is just syntactic sugar // to avoid having to write Passed(std::move(scoper)). // // Both versions of Passed() prevent T from being an lvalue reference. The first // via use of enable_if, and the second takes a T* which will not bind to T&. // // DEPRECATED - Do not use in new code. See https://crbug.com/1326449 template <typename T> requires(!std::is_lvalue_reference_v<T>) inline internal::PassedWrapper<T> Passed(T&& scoper) { … } template <typename T> inline internal::PassedWrapper<T> Passed(T* scoper) { … } // IgnoreResult() is used to adapt a function or callback with a return type to // one with a void return. This is most useful if you have a function with, // say, a pesky ignorable bool return that you want to use with PostTask or // something else that expect a callback with a void return. // // EXAMPLE OF IgnoreResult(): // // int DoSomething(int arg) { cout << arg << endl; } // // // Assign to a callback with a void return type. // OnceCallback<void(int)> cb = BindOnce(IgnoreResult(&DoSomething)); // std::move(cb).Run(1); // Prints "1". // // // Prints "2" on |ml|. // ml->PostTask(FROM_HERE, BindOnce(IgnoreResult(&DoSomething), 2); template <typename T> inline internal::IgnoreResultHelper<T> IgnoreResult(T data) { … } } // namespace base #endif // BASE_FUNCTIONAL_BIND_H_