/*
- * Copyright 2014 Facebook, Inc.
+ * Copyright 2017 Facebook, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <stdexcept>
#include <vector>
+#include <folly/Executor.h>
+#include <folly/Function.h>
+#include <folly/MicroSpinLock.h>
#include <folly/Optional.h>
-#include <folly/SmallLocks.h>
-
-#include <folly/futures/Try.h>
-#include <folly/futures/Promise.h>
+#include <folly/ScopeGuard.h>
+#include <folly/Try.h>
#include <folly/futures/Future.h>
-#include <folly/Executor.h>
+#include <folly/futures/Promise.h>
#include <folly/futures/detail/FSM.h>
#include <folly/io/async/Request.h>
namespace folly { namespace detail {
-// As of GCC 4.8.1, the std::function in libstdc++ optimizes only for pointers
-// to functions, using a helper avoids a call to malloc.
-template<typename T>
-void empty_callback(Try<T>&&) { }
-
-enum class State {
- Waiting,
- Interruptible,
- Interrupted,
+/*
+ OnlyCallback
+ / \
+ Start Armed - Done
+ \ /
+ OnlyResult
+
+This state machine is fairly self-explanatory. The most important bit is
+that the callback is only executed on the transition from Armed to Done,
+and that transition can happen immediately after transitioning from Only*
+to Armed, if it is active (the usual case).
+*/
+enum class State : uint8_t {
+ Start,
+ OnlyResult,
+ OnlyCallback,
+ Armed,
Done,
};
-/** The shared state object for Future and Promise. */
+/// The shared state object for Future and Promise.
+/// Some methods must only be called by either the Future thread or the
+/// Promise thread. The Future thread is the thread that currently "owns" the
+/// Future and its callback-related operations, and the Promise thread is
+/// likewise the thread that currently "owns" the Promise and its
+/// result-related operations. Also, Futures own interruption, Promises own
+/// interrupt handlers. Unfortunately, there are things that users can do to
+/// break this, and we can't detect that. However if they follow move
+/// semantics religiously wrt threading, they should be ok.
+///
+/// It's worth pointing out that Futures and/or Promises can and usually will
+/// migrate between threads, though this usually happens within the API code.
+/// For example, an async operation will probably make a Promise, grab its
+/// Future, then move the Promise into another thread that will eventually
+/// fulfill it. With executors and via, this gets slightly more complicated at
+/// first blush, but it's the same principle. In general, as long as the user
+/// doesn't access a Future or Promise object from more than one thread at a
+/// time there won't be any problems.
template<typename T>
-class Core : protected FSM<State> {
+class Core final {
+ static_assert(!std::is_void<T>::value,
+ "void futures are not supported. Use Unit instead.");
public:
- // This must be heap-constructed. There's probably a way to enforce that in
- // code but since this is just internal detail code and I don't know how
- // off-hand, I'm punting.
- Core() : FSM<State>(State::Waiting) {}
+ /// This must be heap-constructed. There's probably a way to enforce that in
+ /// code but since this is just internal detail code and I don't know how
+ /// off-hand, I'm punting.
+ Core() : result_(), fsm_(State::Start), attached_(2) {}
+
+ explicit Core(Try<T>&& t)
+ : result_(std::move(t)),
+ fsm_(State::OnlyResult),
+ attached_(1) {}
+
~Core() {
- assert(calledBack_);
- assert(detached_ == 2);
+ DCHECK(attached_ == 0);
}
// not copyable
Core(Core&&) noexcept = delete;
Core& operator=(Core&&) = delete;
+ /// May call from any thread
+ bool hasResult() const noexcept {
+ switch (fsm_.getState()) {
+ case State::OnlyResult:
+ case State::Armed:
+ case State::Done:
+ assert(!!result_);
+ return true;
+
+ default:
+ return false;
+ }
+ }
+
+ /// May call from any thread
+ bool ready() const noexcept {
+ return hasResult();
+ }
+
+ /// May call from any thread
Try<T>& getTry() {
if (ready()) {
return *result_;
}
}
+ /// Call only from Future thread.
template <typename F>
- void setCallback(F func) {
+ void setCallback(F&& func) {
+ bool transitionToArmed = false;
auto setCallback_ = [&]{
- if (callback_) {
- throw std::logic_error("setCallback called twice");
- }
-
context_ = RequestContext::saveContext();
- callback_ = std::move(func);
+ callback_ = std::forward<F>(func);
};
- FSM_START
- case State::Waiting:
- case State::Interruptible:
- case State::Interrupted:
- FSM_UPDATE(state, setCallback_);
+ FSM_START(fsm_)
+ case State::Start:
+ FSM_UPDATE(fsm_, State::OnlyCallback, setCallback_);
break;
- case State::Done:
- FSM_UPDATE2(State::Done,
- setCallback_,
- [&]{ maybeCallback(); });
+ case State::OnlyResult:
+ FSM_UPDATE(fsm_, State::Armed, setCallback_);
+ transitionToArmed = true;
break;
+
+ case State::OnlyCallback:
+ case State::Armed:
+ case State::Done:
+ throw std::logic_error("setCallback called twice");
FSM_END
+
+ // we could always call this, it is an optimization to only call it when
+ // it might be needed.
+ if (transitionToArmed) {
+ maybeCallback();
+ }
}
+ /// Call only from Promise thread
void setResult(Try<T>&& t) {
- FSM_START
- case State::Waiting:
- case State::Interruptible:
- case State::Interrupted:
- FSM_UPDATE2(State::Done,
- [&]{ result_ = std::move(t); },
- [&]{ maybeCallback(); });
+ bool transitionToArmed = false;
+ auto setResult_ = [&]{ result_ = std::move(t); };
+ FSM_START(fsm_)
+ case State::Start:
+ FSM_UPDATE(fsm_, State::OnlyResult, setResult_);
+ break;
+
+ case State::OnlyCallback:
+ FSM_UPDATE(fsm_, State::Armed, setResult_);
+ transitionToArmed = true;
break;
+ case State::OnlyResult:
+ case State::Armed:
case State::Done:
throw std::logic_error("setResult called twice");
FSM_END
- }
- bool ready() const {
- return getState() == State::Done;
+ if (transitionToArmed) {
+ maybeCallback();
+ }
}
- // Called by a destructing Future
+ /// Called by a destructing Future (in the Future thread, by definition)
void detachFuture() {
- if (!callback_) {
- setCallback(empty_callback<T>);
- }
activate();
detachOne();
}
- // Called by a destructing Promise
+ /// Called by a destructing Promise (in the Promise thread, by definition)
void detachPromise() {
- if (!ready()) {
- setResult(Try<T>(exception_wrapper(BrokenPromise())));
+ // detachPromise() and setResult() should never be called in parallel
+ // so we don't need to protect this.
+ if (UNLIKELY(!result_)) {
+ setResult(Try<T>(exception_wrapper(BrokenPromise(typeid(T).name()))));
}
detachOne();
}
+ /// May call from any thread
void deactivate() {
- active_ = false;
+ active_.store(false, std::memory_order_release);
}
+ /// May call from any thread
void activate() {
- active_ = true;
- if (ready()) {
- maybeCallback();
- }
+ active_.store(true, std::memory_order_release);
+ maybeCallback();
}
- bool isActive() { return active_; }
+ /// May call from any thread
+ bool isActive() { return active_.load(std::memory_order_acquire); }
- void setExecutor(Executor* x) {
+ /// Call only from Future thread
+ void setExecutor(Executor* x, int8_t priority = Executor::MID_PRI) {
+ if (!executorLock_.try_lock()) {
+ executorLock_.lock();
+ }
executor_ = x;
+ priority_ = priority;
+ executorLock_.unlock();
}
- void raise(exception_wrapper const& e) {
- FSM_START
- case State::Interruptible:
- FSM_UPDATE2(State::Interrupted,
- [&]{ interrupt_ = folly::make_unique<exception_wrapper>(e); },
- [&]{ interruptHandler_(*interrupt_); });
- break;
+ void setExecutorNoLock(Executor* x, int8_t priority = Executor::MID_PRI) {
+ executor_ = x;
+ priority_ = priority;
+ }
- case State::Waiting:
- case State::Interrupted:
- FSM_UPDATE(State::Interrupted,
- [&]{ interrupt_ = folly::make_unique<exception_wrapper>(e); });
- break;
+ Executor* getExecutor() {
+ return executor_;
+ }
- case State::Done:
- FSM_BREAK
- FSM_END
+ /// Call only from Future thread
+ void raise(exception_wrapper e) {
+ if (!interruptLock_.try_lock()) {
+ interruptLock_.lock();
+ }
+ if (!interrupt_ && !hasResult()) {
+ interrupt_ = std::make_unique<exception_wrapper>(std::move(e));
+ if (interruptHandler_) {
+ interruptHandler_(*interrupt_);
+ }
+ }
+ interruptLock_.unlock();
}
- void setInterruptHandler(std::function<void(exception_wrapper const&)> fn) {
- FSM_START
- case State::Waiting:
- case State::Interruptible:
- FSM_UPDATE(State::Interruptible,
- [&]{ interruptHandler_ = std::move(fn); });
- break;
+ std::function<void(exception_wrapper const&)> getInterruptHandler() {
+ if (!interruptHandlerSet_.load(std::memory_order_acquire)) {
+ return nullptr;
+ }
+ if (!interruptLock_.try_lock()) {
+ interruptLock_.lock();
+ }
+ auto handler = interruptHandler_;
+ interruptLock_.unlock();
+ return handler;
+ }
- case State::Interrupted:
+ /// Call only from Promise thread
+ void setInterruptHandler(std::function<void(exception_wrapper const&)> fn) {
+ if (!interruptLock_.try_lock()) {
+ interruptLock_.lock();
+ }
+ if (!hasResult()) {
+ if (interrupt_) {
fn(*interrupt_);
+ } else {
+ setInterruptHandlerNoLock(std::move(fn));
+ }
+ }
+ interruptLock_.unlock();
+ }
+
+ void setInterruptHandlerNoLock(
+ std::function<void(exception_wrapper const&)> fn) {
+ interruptHandlerSet_.store(true, std::memory_order_relaxed);
+ interruptHandler_ = std::move(fn);
+ }
+
+ private:
+ // Helper class that stores a pointer to the `Core` object and calls
+ // `derefCallback` and `detachOne` in the destructor.
+ class CoreAndCallbackReference {
+ public:
+ explicit CoreAndCallbackReference(Core* core) noexcept : core_(core) {}
+
+ ~CoreAndCallbackReference() {
+ if (core_) {
+ core_->derefCallback();
+ core_->detachOne();
+ }
+ }
+
+ CoreAndCallbackReference(CoreAndCallbackReference const& o) = delete;
+ CoreAndCallbackReference& operator=(CoreAndCallbackReference const& o) =
+ delete;
+
+ CoreAndCallbackReference(CoreAndCallbackReference&& o) noexcept {
+ std::swap(core_, o.core_);
+ }
+
+ Core* getCore() const noexcept {
+ return core_;
+ }
+
+ private:
+ Core* core_{nullptr};
+ };
+
+ void maybeCallback() {
+ FSM_START(fsm_)
+ case State::Armed:
+ if (active_.load(std::memory_order_acquire)) {
+ FSM_UPDATE2(fsm_, State::Done, []{}, [this]{ this->doCallback(); });
+ }
FSM_BREAK
- case State::Done:
+ default:
FSM_BREAK
FSM_END
}
- private:
- void maybeCallback() {
- assert(ready());
- if (isActive() && callback_) {
- if (!calledBack_.exchange(true)) {
- // TODO(5306911) we should probably try/catch
- Executor* x = executor_;
-
- RequestContext::setContext(context_);
- if (x) {
- MoveWrapper<std::function<void(Try<T>&&)>> cb(std::move(callback_));
- MoveWrapper<folly::Optional<Try<T>>> val(std::move(result_));
- x->add([cb, val]() mutable { (*cb)(std::move(**val)); });
+ void doCallback() {
+ Executor* x = executor_;
+ int8_t priority;
+ if (x) {
+ if (!executorLock_.try_lock()) {
+ executorLock_.lock();
+ }
+ x = executor_;
+ priority = priority_;
+ executorLock_.unlock();
+ }
+
+ if (x) {
+ exception_wrapper ew;
+ // We need to reset `callback_` after it was executed (which can happen
+ // through the executor or, if `Executor::add` throws, below). The
+ // executor might discard the function without executing it (now or
+ // later), in which case `callback_` also needs to be reset.
+ // The `Core` has to be kept alive throughout that time, too. Hence we
+ // increment `attached_` and `callbackReferences_` by two, and construct
+ // exactly two `CoreAndCallbackReference` objects, which call
+ // `derefCallback` and `detachOne` in their destructor. One will guard
+ // this scope, the other one will guard the lambda passed to the executor.
+ attached_ += 2;
+ callbackReferences_ += 2;
+ CoreAndCallbackReference guard_local_scope(this);
+ CoreAndCallbackReference guard_lambda(this);
+ try {
+ if (LIKELY(x->getNumPriorities() == 1)) {
+ x->add([core_ref = std::move(guard_lambda)]() mutable {
+ auto cr = std::move(core_ref);
+ Core* const core = cr.getCore();
+ RequestContextScopeGuard rctx(core->context_);
+ core->callback_(std::move(*core->result_));
+ });
} else {
- callback_(std::move(*result_));
+ x->addWithPriority(
+ [core_ref = std::move(guard_lambda)]() mutable {
+ auto cr = std::move(core_ref);
+ Core* const core = cr.getCore();
+ RequestContextScopeGuard rctx(core->context_);
+ core->callback_(std::move(*core->result_));
+ },
+ priority);
}
+ } catch (const std::exception& e) {
+ ew = exception_wrapper(std::current_exception(), e);
+ } catch (...) {
+ ew = exception_wrapper(std::current_exception());
}
+ if (ew) {
+ RequestContextScopeGuard rctx(context_);
+ result_ = Try<T>(std::move(ew));
+ callback_(std::move(*result_));
+ }
+ } else {
+ attached_++;
+ SCOPE_EXIT {
+ callback_ = {};
+ detachOne();
+ };
+ RequestContextScopeGuard rctx(context_);
+ callback_(std::move(*result_));
}
}
void detachOne() {
- auto d = ++detached_;
- assert(d >= 1);
- assert(d <= 2);
- if (d == 2) {
- // we should have already executed the callback with the value
- assert(calledBack_);
+ auto a = attached_--;
+ assert(a >= 1);
+ if (a == 1) {
delete this;
}
}
+ void derefCallback() {
+ if (--callbackReferences_ == 0) {
+ callback_ = {};
+ }
+ }
+
+ folly::Function<void(Try<T>&&)> callback_;
+ // place result_ next to increase the likelihood that the value will be
+ // contained entirely in one cache line
folly::Optional<Try<T>> result_;
- std::function<void(Try<T>&&)> callback_;
- std::shared_ptr<RequestContext> context_{nullptr};
- std::atomic<bool> calledBack_ {false};
- std::atomic<unsigned char> detached_ {0};
+ FSM<State> fsm_;
+ std::atomic<unsigned char> attached_;
+ std::atomic<unsigned char> callbackReferences_{0};
std::atomic<bool> active_ {true};
- std::atomic<Executor*> executor_ {nullptr};
- std::unique_ptr<exception_wrapper> interrupt_;
- std::function<void(exception_wrapper const&)> interruptHandler_;
+ std::atomic<bool> interruptHandlerSet_ {false};
+ folly::MicroSpinLock interruptLock_ {0};
+ folly::MicroSpinLock executorLock_ {0};
+ int8_t priority_ {-1};
+ Executor* executor_ {nullptr};
+ std::shared_ptr<RequestContext> context_ {nullptr};
+ std::unique_ptr<exception_wrapper> interrupt_ {};
+ std::function<void(exception_wrapper const&)> interruptHandler_ {nullptr};
};
template <typename... Ts>
-struct VariadicContext {
- VariadicContext() : total(0), count(0) {}
- Promise<std::tuple<Try<Ts>... > > p;
- std::tuple<Try<Ts>... > results;
- size_t total;
- std::atomic<size_t> count;
+struct CollectAllVariadicContext {
+ CollectAllVariadicContext() {}
+ template <typename T, size_t I>
+ inline void setPartialResult(Try<T>& t) {
+ std::get<I>(results) = std::move(t);
+ }
+ ~CollectAllVariadicContext() {
+ p.setValue(std::move(results));
+ }
+ Promise<std::tuple<Try<Ts>...>> p;
+ std::tuple<Try<Ts>...> results;
typedef Future<std::tuple<Try<Ts>...>> type;
};
-template <typename... Ts, typename THead, typename... Fs>
-typename std::enable_if<sizeof...(Fs) == 0, void>::type
-whenAllVariadicHelper(VariadicContext<Ts...> *ctx, THead&& head, Fs&&... tail) {
- head.setCallback_([ctx](Try<typename THead::value_type>&& t) {
- std::get<sizeof...(Ts) - sizeof...(Fs) - 1>(ctx->results) = std::move(t);
- if (++ctx->count == ctx->total) {
- ctx->p.setValue(std::move(ctx->results));
- delete ctx;
+template <typename... Ts>
+struct CollectVariadicContext {
+ CollectVariadicContext() {}
+ template <typename T, size_t I>
+ inline void setPartialResult(Try<T>& t) {
+ if (t.hasException()) {
+ if (!threw.exchange(true)) {
+ p.setException(std::move(t.exception()));
+ }
+ } else if (!threw) {
+ std::get<I>(results) = std::move(t);
+ }
+ }
+ ~CollectVariadicContext() noexcept {
+ if (!threw.exchange(true)) {
+ p.setValue(unwrapTryTuple(std::move(results)));
}
- });
+ }
+ Promise<std::tuple<Ts...>> p;
+ std::tuple<folly::Try<Ts>...> results;
+ std::atomic<bool> threw {false};
+ typedef Future<std::tuple<Ts...>> type;
+};
+
+template <template <typename...> class T, typename... Ts>
+void collectVariadicHelper(const std::shared_ptr<T<Ts...>>& /* ctx */) {
+ // base case
}
-template <typename... Ts, typename THead, typename... Fs>
-typename std::enable_if<sizeof...(Fs) != 0, void>::type
-whenAllVariadicHelper(VariadicContext<Ts...> *ctx, THead&& head, Fs&&... tail) {
- head.setCallback_([ctx](Try<typename THead::value_type>&& t) {
- std::get<sizeof...(Ts) - sizeof...(Fs) - 1>(ctx->results) = std::move(t);
- if (++ctx->count == ctx->total) {
- ctx->p.setValue(std::move(ctx->results));
- delete ctx;
- }
+template <template <typename ...> class T, typename... Ts,
+ typename THead, typename... TTail>
+void collectVariadicHelper(const std::shared_ptr<T<Ts...>>& ctx,
+ THead&& head, TTail&&... tail) {
+ using ValueType = typename std::decay<THead>::type::value_type;
+ std::forward<THead>(head).setCallback_([ctx](Try<ValueType>&& t) {
+ ctx->template setPartialResult<
+ ValueType,
+ sizeof...(Ts) - sizeof...(TTail)-1>(t);
});
// template tail-recursion
- whenAllVariadicHelper(ctx, std::forward<Fs>(tail)...);
+ collectVariadicHelper(ctx, std::forward<TTail>(tail)...);
}
-template <typename T>
-struct WhenAllContext {
- WhenAllContext() : count(0) {}
- Promise<std::vector<Try<T> > > p;
- std::vector<Try<T> > results;
- std::atomic<size_t> count;
-};
-
-template <typename T>
-struct WhenAnyContext {
- explicit WhenAnyContext(size_t n) : done(false), ref_count(n) {};
- Promise<std::pair<size_t, Try<T>>> p;
- std::atomic<bool> done;
- std::atomic<size_t> ref_count;
- void decref() {
- if (--ref_count == 0) {
- delete this;
- }
- }
-};
-
}} // folly::detail
projects / folly.git / blobdiff