summaryrefslogtreecommitdiffstats
path: root/xpcom/threads/TaskDispatcher.h
diff options
context:
space:
mode:
authorMatt A. Tobin <mattatobin@localhost.localdomain>2018-02-02 04:16:08 -0500
committerMatt A. Tobin <mattatobin@localhost.localdomain>2018-02-02 04:16:08 -0500
commit5f8de423f190bbb79a62f804151bc24824fa32d8 (patch)
tree10027f336435511475e392454359edea8e25895d /xpcom/threads/TaskDispatcher.h
parent49ee0794b5d912db1f95dce6eb52d781dc210db5 (diff)
downloadUXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar
UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar.gz
UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar.lz
UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar.xz
UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.zip
Add m-esr52 at 52.6.0
Diffstat (limited to 'xpcom/threads/TaskDispatcher.h')
-rw-r--r--xpcom/threads/TaskDispatcher.h276
1 files changed, 276 insertions, 0 deletions
diff --git a/xpcom/threads/TaskDispatcher.h b/xpcom/threads/TaskDispatcher.h
new file mode 100644
index 000000000..405c3acfe
--- /dev/null
+++ b/xpcom/threads/TaskDispatcher.h
@@ -0,0 +1,276 @@
+/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim: set ts=8 sts=2 et sw=2 tw=80: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#if !defined(TaskDispatcher_h_)
+#define TaskDispatcher_h_
+
+#include "mozilla/AbstractThread.h"
+#include "mozilla/Maybe.h"
+#include "mozilla/UniquePtr.h"
+#include "mozilla/Unused.h"
+
+#include "nsISupportsImpl.h"
+#include "nsTArray.h"
+#include "nsThreadUtils.h"
+
+#include <queue>
+
+namespace mozilla {
+
+/*
+ * A classic approach to cross-thread communication is to dispatch asynchronous
+ * runnables to perform updates on other threads. This generally works well, but
+ * there are sometimes reasons why we might want to delay the actual dispatch of
+ * these tasks until a specified moment. At present, this is primarily useful to
+ * ensure that mirrored state gets updated atomically - but there may be other
+ * applications as well.
+ *
+ * TaskDispatcher is a general abstract class that accepts tasks and dispatches
+ * them at some later point. These groups of tasks are per-target-thread, and
+ * contain separate queues for several kinds of tasks (see comments below). - "state change tasks" (which
+ * run first, and are intended to be used to update the value held by mirrors),
+ * and regular tasks, which are other arbitrary operations that the are gated
+ * to run after all the state changes have completed.
+ */
+class TaskDispatcher
+{
+public:
+ TaskDispatcher() {}
+ virtual ~TaskDispatcher() {}
+
+ // Direct tasks are run directly (rather than dispatched asynchronously) when
+ // the tail dispatcher fires. A direct task may cause other tasks to be added
+ // to the tail dispatcher.
+ virtual void AddDirectTask(already_AddRefed<nsIRunnable> aRunnable) = 0;
+
+ // State change tasks are dispatched asynchronously always run before regular
+ // tasks. They are intended to be used to update the value held by mirrors
+ // before any other dispatched tasks are run on the target thread.
+ virtual void AddStateChangeTask(AbstractThread* aThread,
+ already_AddRefed<nsIRunnable> aRunnable) = 0;
+
+ // Regular tasks are dispatched asynchronously, and run after state change
+ // tasks.
+ virtual void AddTask(AbstractThread* aThread,
+ already_AddRefed<nsIRunnable> aRunnable,
+ AbstractThread::DispatchFailureHandling aFailureHandling = AbstractThread::AssertDispatchSuccess) = 0;
+
+ virtual void DispatchTasksFor(AbstractThread* aThread) = 0;
+ virtual bool HasTasksFor(AbstractThread* aThread) = 0;
+ virtual void DrainDirectTasks() = 0;
+};
+
+/*
+ * AutoTaskDispatcher is a stack-scoped TaskDispatcher implementation that fires
+ * its queued tasks when it is popped off the stack.
+ */
+class AutoTaskDispatcher : public TaskDispatcher
+{
+public:
+ explicit AutoTaskDispatcher(bool aIsTailDispatcher = false)
+ : mIsTailDispatcher(aIsTailDispatcher)
+ {}
+
+ ~AutoTaskDispatcher()
+ {
+ // Given that direct tasks may trigger other code that uses the tail
+ // dispatcher, it's better to avoid processing them in the tail dispatcher's
+ // destructor. So we require TailDispatchers to manually invoke
+ // DrainDirectTasks before the AutoTaskDispatcher gets destroyed. In truth,
+ // this is only necessary in the case where this AutoTaskDispatcher can be
+ // accessed by the direct tasks it dispatches (true for TailDispatchers, but
+ // potentially not true for other hypothetical AutoTaskDispatchers). Feel
+ // free to loosen this restriction to apply only to mIsTailDispatcher if a
+ // use-case requires it.
+ MOZ_ASSERT(!HaveDirectTasks());
+
+ for (size_t i = 0; i < mTaskGroups.Length(); ++i) {
+ DispatchTaskGroup(Move(mTaskGroups[i]));
+ }
+ }
+
+ bool HaveDirectTasks() const
+ {
+ return mDirectTasks.isSome() && !mDirectTasks->empty();
+ }
+
+ void DrainDirectTasks() override
+ {
+ while (HaveDirectTasks()) {
+ nsCOMPtr<nsIRunnable> r = mDirectTasks->front();
+ mDirectTasks->pop();
+ r->Run();
+ }
+ }
+
+ void AddDirectTask(already_AddRefed<nsIRunnable> aRunnable) override
+ {
+ if (mDirectTasks.isNothing()) {
+ mDirectTasks.emplace();
+ }
+ mDirectTasks->push(Move(aRunnable));
+ }
+
+ void AddStateChangeTask(AbstractThread* aThread,
+ already_AddRefed<nsIRunnable> aRunnable) override
+ {
+ EnsureTaskGroup(aThread).mStateChangeTasks.AppendElement(aRunnable);
+ }
+
+ void AddTask(AbstractThread* aThread,
+ already_AddRefed<nsIRunnable> aRunnable,
+ AbstractThread::DispatchFailureHandling aFailureHandling) override
+ {
+ PerThreadTaskGroup& group = EnsureTaskGroup(aThread);
+ group.mRegularTasks.AppendElement(aRunnable);
+
+ // The task group needs to assert dispatch success if any of the runnables
+ // it's dispatching want to assert it.
+ if (aFailureHandling == AbstractThread::AssertDispatchSuccess) {
+ group.mFailureHandling = AbstractThread::AssertDispatchSuccess;
+ }
+ }
+
+ bool HasTasksFor(AbstractThread* aThread) override
+ {
+ return !!GetTaskGroup(aThread) ||
+ (aThread == AbstractThread::GetCurrent() && HaveDirectTasks());
+ }
+
+ void DispatchTasksFor(AbstractThread* aThread) override
+ {
+ for (size_t i = 0; i < mTaskGroups.Length(); ++i) {
+ if (mTaskGroups[i]->mThread == aThread) {
+ DispatchTaskGroup(Move(mTaskGroups[i]));
+ mTaskGroups.RemoveElementAt(i);
+ return;
+ }
+ }
+ }
+
+private:
+
+ struct PerThreadTaskGroup
+ {
+ public:
+ explicit PerThreadTaskGroup(AbstractThread* aThread)
+ : mThread(aThread), mFailureHandling(AbstractThread::DontAssertDispatchSuccess)
+ {
+ MOZ_COUNT_CTOR(PerThreadTaskGroup);
+ }
+
+ ~PerThreadTaskGroup() { MOZ_COUNT_DTOR(PerThreadTaskGroup); }
+
+ RefPtr<AbstractThread> mThread;
+ nsTArray<nsCOMPtr<nsIRunnable>> mStateChangeTasks;
+ nsTArray<nsCOMPtr<nsIRunnable>> mRegularTasks;
+ AbstractThread::DispatchFailureHandling mFailureHandling;
+ };
+
+ class TaskGroupRunnable : public Runnable
+ {
+ public:
+ explicit TaskGroupRunnable(UniquePtr<PerThreadTaskGroup>&& aTasks) : mTasks(Move(aTasks)) {}
+
+ NS_IMETHOD Run() override
+ {
+ // State change tasks get run all together before any code is run, so
+ // that all state changes are made in an atomic unit.
+ for (size_t i = 0; i < mTasks->mStateChangeTasks.Length(); ++i) {
+ mTasks->mStateChangeTasks[i]->Run();
+ }
+
+ // Once the state changes have completed, drain any direct tasks
+ // generated by those state changes (i.e. watcher notification tasks).
+ // This needs to be outside the loop because we don't want to run code
+ // that might observe intermediate states.
+ MaybeDrainDirectTasks();
+
+ for (size_t i = 0; i < mTasks->mRegularTasks.Length(); ++i) {
+ mTasks->mRegularTasks[i]->Run();
+
+ // Scope direct tasks tightly to the task that generated them.
+ MaybeDrainDirectTasks();
+ }
+
+ return NS_OK;
+ }
+
+ private:
+ void MaybeDrainDirectTasks()
+ {
+ AbstractThread* currentThread = AbstractThread::GetCurrent();
+ if (currentThread) {
+ currentThread->TailDispatcher().DrainDirectTasks();
+ }
+ }
+
+ UniquePtr<PerThreadTaskGroup> mTasks;
+ };
+
+ PerThreadTaskGroup& EnsureTaskGroup(AbstractThread* aThread)
+ {
+ PerThreadTaskGroup* existing = GetTaskGroup(aThread);
+ if (existing) {
+ return *existing;
+ }
+
+ mTaskGroups.AppendElement(new PerThreadTaskGroup(aThread));
+ return *mTaskGroups.LastElement();
+ }
+
+ PerThreadTaskGroup* GetTaskGroup(AbstractThread* aThread)
+ {
+ for (size_t i = 0; i < mTaskGroups.Length(); ++i) {
+ if (mTaskGroups[i]->mThread == aThread) {
+ return mTaskGroups[i].get();
+ }
+ }
+
+ // Not found.
+ return nullptr;
+ }
+
+ void DispatchTaskGroup(UniquePtr<PerThreadTaskGroup> aGroup)
+ {
+ RefPtr<AbstractThread> thread = aGroup->mThread;
+
+ AbstractThread::DispatchFailureHandling failureHandling = aGroup->mFailureHandling;
+ AbstractThread::DispatchReason reason = mIsTailDispatcher ? AbstractThread::TailDispatch
+ : AbstractThread::NormalDispatch;
+ nsCOMPtr<nsIRunnable> r = new TaskGroupRunnable(Move(aGroup));
+ thread->Dispatch(r.forget(), failureHandling, reason);
+ }
+
+ // Direct tasks. We use a Maybe<> because (a) this class is hot, (b)
+ // mDirectTasks often doesn't get anything put into it, and (c) the
+ // std::queue implementation in GNU libstdc++ does two largish heap
+ // allocations when creating a new std::queue.
+ mozilla::Maybe<std::queue<nsCOMPtr<nsIRunnable>>> mDirectTasks;
+
+ // Task groups, organized by thread.
+ nsTArray<UniquePtr<PerThreadTaskGroup>> mTaskGroups;
+
+ // True if this TaskDispatcher represents the tail dispatcher for the thread
+ // upon which it runs.
+ const bool mIsTailDispatcher;
+};
+
+// Little utility class to allow declaring AutoTaskDispatcher as a default
+// parameter for methods that take a TaskDispatcher&.
+template<typename T>
+class PassByRef
+{
+public:
+ PassByRef() {}
+ operator T&() { return mVal; }
+private:
+ T mVal;
+};
+
+} // namespace mozilla
+
+#endif