summaryrefslogtreecommitdiffstats
path: root/base/message_loop.cc
diff options
context:
space:
mode:
authordarin@google.com <darin@google.com@0039d316-1c4b-4281-b951-d872f2087c98>2008-08-15 07:31:20 +0000
committerdarin@google.com <darin@google.com@0039d316-1c4b-4281-b951-d872f2087c98>2008-08-15 07:31:20 +0000
commitea15e98a9353494df754a11fb49f0e4b8c0d4789 (patch)
tree9878dbb01f722a4827d6529d994c38960fa7978a /base/message_loop.cc
parent62d22d763ad4d04d274fb156ee58cf546bab2064 (diff)
downloadchromium_src-ea15e98a9353494df754a11fb49f0e4b8c0d4789.zip
chromium_src-ea15e98a9353494df754a11fb49f0e4b8c0d4789.tar.gz
chromium_src-ea15e98a9353494df754a11fb49f0e4b8c0d4789.tar.bz2
rollback portions of r928 to test to see if it impacts perf
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@936 0039d316-1c4b-4281-b951-d872f2087c98
Diffstat (limited to 'base/message_loop.cc')
-rw-r--r--base/message_loop.cc695
1 files changed, 562 insertions, 133 deletions
diff --git a/base/message_loop.cc b/base/message_loop.cc
index b920167..18f4448 100644
--- a/base/message_loop.cc
+++ b/base/message_loop.cc
@@ -27,13 +27,14 @@
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-#include "base/message_loop.h"
-
#include <algorithm>
+#include "base/message_loop.h"
+
#include "base/logging.h"
#include "base/string_util.h"
#include "base/thread_local_storage.h"
+#include "base/win_util.h"
// a TLS index to the message loop for the current thread
// Note that if we start doing complex stuff in other static initializers
@@ -42,10 +43,25 @@
//------------------------------------------------------------------------------
+static const wchar_t kWndClass[] = L"Chrome_MessageLoopWindow";
+
+// Windows Message numbers handled by WindowMessageProc.
+
+// Message sent to get an additional time slice for pumping (processing) another
+// task (a series of such messages creates a continuous task pump).
+static const int kMsgPumpATask = WM_USER + 1;
+
+// Message sent by Quit() to cause our main message pump to terminate as soon as
+// all pending task and message queues have been emptied.
+static const int kMsgQuit = WM_USER + 2;
+
// Logical events for Histogram profiling. Run with -message-loop-histogrammer
// to get an accounting of messages and actions taken on each thread.
-static const int kTaskRunEvent = 0x1;
-static const int kTimerEvent = 0x2;
+static const int kTaskRunEvent = WM_USER + 16; // 0x411
+static const int kSleepingApcEvent = WM_USER + 17; // 0x411
+static const int kPollingSignalEvent = WM_USER + 18; // 0x412
+static const int kSleepingSignalEvent = WM_USER + 19; // 0x413
+static const int kTimerEvent = WM_USER + 20; // 0x414
// Provide range of message IDs for use in histogramming and debug display.
static const int kLeastNonZeroMessageId = 1;
@@ -54,8 +70,26 @@ static const int kNumberOfDistinctMessagesDisplayed = 1100;
//------------------------------------------------------------------------------
-#if defined(OS_WIN)
+#ifndef NDEBUG
+// Force exercise of polling model.
+#define CHROME_MAXIMUM_WAIT_OBJECTS 8
+#else
+#define CHROME_MAXIMUM_WAIT_OBJECTS MAXIMUM_WAIT_OBJECTS
+#endif
+//------------------------------------------------------------------------------
+// A strategy of -1 uses the default case. All strategies are selected as
+// positive integers.
+// static
+int MessageLoop::strategy_selector_ = -1;
+
+// static
+void MessageLoop::SetStrategy(int strategy) {
+ DCHECK(-1 == strategy_selector_);
+ strategy_selector_ = strategy;
+}
+
+//------------------------------------------------------------------------------
// Upon a SEH exception in this thread, it restores the original unhandled
// exception filter.
static int SEHFilter(LPTOP_LEVEL_EXCEPTION_FILTER old_filter) {
@@ -72,22 +106,23 @@ static LPTOP_LEVEL_EXCEPTION_FILTER GetTopSEHFilter() {
return top_filter;
}
-#endif // defined(OS_WIN)
-
//------------------------------------------------------------------------------
MessageLoop::MessageLoop()
#pragma warning(suppress: 4355) // OK, to use |this| in the initializer list.
: timer_manager_(this),
+ message_hwnd_(NULL),
exception_restoration_(false),
nestable_tasks_allowed_(true),
- state_(NULL) {
+ dispatcher_(NULL),
+ quit_received_(false),
+ quit_now_(false),
+ task_pump_message_pending_(false),
+ run_depth_(0) {
DCHECK(tls_index_) << "static initializer failed";
DCHECK(!current()) << "should only have one message loop per thread";
ThreadLocalStorage::Set(tls_index_, this);
-#if defined(OS_WIN)
- pump_ = new base::MessagePumpWin();
-#endif
+ InitMessageWnd();
}
MessageLoop::~MessageLoop() {
@@ -100,7 +135,8 @@ MessageLoop::~MessageLoop() {
// OK, now make it so that no one can find us.
ThreadLocalStorage::Set(tls_index_, NULL);
- DCHECK(!state_);
+ DCHECK(!dispatcher_);
+ DCHECK(!quit_received_ && !quit_now_);
// Most tasks that have not been Run() are deleted in the |timer_manager_|
// destructor after we remove our tls index. We delete the tasks in our
@@ -111,6 +147,12 @@ MessageLoop::~MessageLoop() {
DeletePendingTasks();
}
+void MessageLoop::SetThreadName(const std::string& thread_name) {
+ DCHECK(thread_name_.empty());
+ thread_name_ = thread_name;
+ StartHistogrammer();
+}
+
void MessageLoop::AddDestructionObserver(DestructionObserver *obs) {
DCHECK(this == current());
destruction_observers_.AddObserver(obs);
@@ -121,23 +163,26 @@ void MessageLoop::RemoveDestructionObserver(DestructionObserver *obs) {
destruction_observers_.RemoveObserver(obs);
}
+void MessageLoop::AddObserver(Observer *obs) {
+ DCHECK(this == current());
+ observers_.AddObserver(obs);
+}
+
+void MessageLoop::RemoveObserver(Observer *obs) {
+ DCHECK(this == current());
+ observers_.RemoveObserver(obs);
+}
+
void MessageLoop::Run() {
- AutoRunState save_state(this);
- RunHandler();
+ RunHandler(NULL, false);
}
-#if defined(OS_WIN)
-void MessageLoop::Run(base::MessagePumpWin::Dispatcher* dispatcher) {
- AutoRunState save_state(this);
- state_->dispatcher = dispatcher;
- RunHandler();
+void MessageLoop::Run(Dispatcher* dispatcher) {
+ RunHandler(dispatcher, false);
}
-#endif
void MessageLoop::RunAllPending() {
- AutoRunState save_state(this);
- state_->quit_received = true; // Means run until we would otherwise block.
- RunHandler();
+ RunHandler(NULL, true);
}
// Runs the loop in two different SEH modes:
@@ -145,43 +190,94 @@ void MessageLoop::RunAllPending() {
// one that calls SetUnhandledExceptionFilter().
// enable_SEH_restoration_ = true : any unhandled exception goes to the filter
// that was existed before the loop was run.
-void MessageLoop::RunHandler() {
-#if defined(OS_WIN)
+void MessageLoop::RunHandler(Dispatcher* dispatcher, bool non_blocking) {
if (exception_restoration_) {
LPTOP_LEVEL_EXCEPTION_FILTER current_filter = GetTopSEHFilter();
__try {
- RunInternal();
+ RunInternal(dispatcher, non_blocking);
} __except(SEHFilter(current_filter)) {
}
- return;
+ } else {
+ RunInternal(dispatcher, non_blocking);
}
-#endif
-
- RunInternal();
}
//------------------------------------------------------------------------------
+// IF this was just a simple PeekMessage() loop (servicing all passible work
+// queues), then Windows would try to achieve the following order according to
+// MSDN documentation about PeekMessage with no filter):
+// * Sent messages
+// * Posted messages
+// * Sent messages (again)
+// * WM_PAINT messages
+// * WM_TIMER messages
+//
+// Summary: none of the above classes is starved, and sent messages has twice
+// the chance of being processed (i.e., reduced service time).
+
+void MessageLoop::RunInternal(Dispatcher* dispatcher, bool non_blocking) {
+ // Preserve ability to be called recursively.
+ ScopedStateSave save(this); // State is restored on exit.
+ dispatcher_ = dispatcher;
+ StartHistogrammer();
-void MessageLoop::RunInternal() {
DCHECK(this == current());
+ //
+ // Process pending messages and signaled objects.
+ //
+ // Flush these queues before exiting due to a kMsgQuit or else we risk not
+ // shutting down properly as some operations may depend on further event
+ // processing. (Note: some tests may use quit_now_ to exit more swiftly,
+ // and leave messages pending, so don't assert the above fact).
+ RunTraditional(non_blocking);
+ DCHECK(non_blocking || quit_received_ || quit_now_);
+}
- StartHistogrammer();
+void MessageLoop::RunTraditional(bool non_blocking) {
+ for (;;) {
+ // If we do any work, we may create more messages etc., and more work
+ // may possibly be waiting in another task group. When we (for example)
+ // ProcessNextWindowsMessage(), there is a good chance there are still more
+ // messages waiting (same thing for ProcessNextObject(), which responds to
+ // only one signaled object; etc.). On the other hand, when any of these
+ // methods return having done no work, then it is pretty unlikely that
+ // calling them again quickly will find any work to do.
+ // Finally, if they all say they had no work, then it is a good time to
+ // consider sleeping (waiting) for more work.
+ bool more_work_is_plausible = ProcessNextWindowsMessage();
+ if (quit_now_)
+ return;
-#if defined(OS_WIN)
- if (state_->dispatcher) {
- pump_win()->RunWithDispatcher(this, state_->dispatcher);
- return;
+ more_work_is_plausible |= ProcessNextDeferredTask();
+ more_work_is_plausible |= ProcessNextObject();
+ if (more_work_is_plausible)
+ continue;
+
+ if (quit_received_)
+ return;
+
+ // Run any timer that is ready to run. It may create messages etc.
+ if (ProcessSomeTimers())
+ continue;
+
+ // We run delayed non nestable tasks only after all nestable tasks have
+ // run, to preserve FIFO ordering.
+ if (ProcessNextDelayedNonNestableTask())
+ continue;
+
+ if (non_blocking)
+ return;
+
+ // We service APCs in WaitForWork, without returning.
+ WaitForWork(); // Wait (sleep) until we have work to do again.
}
-#endif
-
- pump_->Run(this);
}
//------------------------------------------------------------------------------
// Wrapper functions for use in above message loop framework.
bool MessageLoop::ProcessNextDelayedNonNestableTask() {
- if (state_->run_depth != 1)
+ if (run_depth_ != 1)
return false;
if (delayed_non_nestable_queue_.Empty())
@@ -203,12 +299,41 @@ bool MessageLoop::ProcessSomeTimers() {
//------------------------------------------------------------------------------
void MessageLoop::Quit() {
- DCHECK(current() == this);
- if (state_) {
- state_->quit_received = true;
- } else {
- NOTREACHED() << "Must be inside Run to call Quit";
+ EnsureMessageGetsPosted(kMsgQuit);
+}
+
+bool MessageLoop::WatchObject(HANDLE object, Watcher* watcher) {
+ DCHECK(this == current());
+ DCHECK(object);
+ DCHECK_NE(object, INVALID_HANDLE_VALUE);
+
+ std::vector<HANDLE>::iterator it = find(objects_.begin(), objects_.end(),
+ object);
+ if (watcher) {
+ if (it == objects_.end()) {
+ static size_t warning_multiple = 1;
+ if (objects_.size() >= warning_multiple * MAXIMUM_WAIT_OBJECTS / 2) {
+ LOG(INFO) << "More than " << warning_multiple * MAXIMUM_WAIT_OBJECTS / 2
+ << " objects being watched";
+ // This DCHECK() is an artificial limitation, meant to warn us if we
+ // start creating too many objects. It can safely be raised to a higher
+ // level, and the program is designed to handle much larger values.
+ // Before raising this limit, make sure that there is a very good reason
+ // (in your debug testing) to be watching this many objects.
+ DCHECK(2 <= warning_multiple);
+ ++warning_multiple;
+ }
+ objects_.push_back(object);
+ watchers_.push_back(watcher);
+ } else {
+ watchers_[it - objects_.begin()] = watcher;
+ }
+ } else if (it != objects_.end()) {
+ std::vector<HANDLE>::difference_type index = it - objects_.begin();
+ objects_.erase(it);
+ watchers_.erase(watchers_.begin() + index);
}
+ return true;
}
// Possibly called on a background thread!
@@ -227,23 +352,88 @@ void MessageLoop::PostTaskInternal(Task* task) {
// directly, as it could starve handling of foreign threads. Put every task
// into this queue.
- scoped_refptr<base::MessagePump> pump;
+ // Local stack variables to use IF we need to process after releasing locks.
+ HWND message_hwnd;
{
- AutoLock locked(incoming_queue_lock_);
-
+ AutoLock lock1(incoming_queue_lock_);
bool was_empty = incoming_queue_.Empty();
incoming_queue_.Push(task);
if (!was_empty)
return; // Someone else should have started the sub-pump.
- pump = pump_;
+ // We may have to start the sub-pump.
+ AutoLock lock2(task_pump_message_lock_);
+ if (task_pump_message_pending_)
+ return; // Someone else continued the pumping.
+ task_pump_message_pending_ = true; // We'll send one.
+ message_hwnd = message_hwnd_;
+ } // Release both locks.
+ // We may have just posted a kMsgQuit, and so this instance may now destroyed!
+ // Do not invoke non-static methods, or members in any way!
+
+ // PostMessage may fail, as the hwnd may have vanished due to kMsgQuit.
+ PostMessage(message_hwnd, kMsgPumpATask, 0, 0);
+}
+
+void MessageLoop::InitMessageWnd() {
+ HINSTANCE hinst = GetModuleHandle(NULL);
+
+ WNDCLASSEX wc = {0};
+ wc.cbSize = sizeof(wc);
+ wc.lpfnWndProc = WndProcThunk;
+ wc.hInstance = hinst;
+ wc.lpszClassName = kWndClass;
+ RegisterClassEx(&wc);
+
+ message_hwnd_ = CreateWindow(kWndClass, 0, 0, 0, 0, 0, 0, HWND_MESSAGE, 0,
+ hinst, 0);
+ DCHECK(message_hwnd_);
+}
+
+// static
+LRESULT CALLBACK MessageLoop::WndProcThunk(
+ HWND hwnd, UINT message, WPARAM wparam, LPARAM lparam) {
+ DCHECK(MessageLoop::current());
+ return MessageLoop::current()->WndProc(hwnd, message, wparam, lparam);
+}
+
+LRESULT MessageLoop::WndProc(
+ HWND hwnd, UINT message, WPARAM wparam, LPARAM lparam) {
+ if (hwnd == message_hwnd_) {
+ switch (message) {
+ case kMsgPumpATask: {
+ ProcessPumpReplacementMessage(); // Avoid starving paint and timer.
+ if (!nestable_tasks_allowed_)
+ return 0;
+ PumpATaskDuringWndProc();
+ return 0;
+ }
+
+ case WM_TIMER:
+ ProcessSomeTimers(); // Give the TimerManager a tickle.
+ DidChangeNextTimerExpiry(); // Maybe generate another WM_TIMER.
+ return 0;
+
+ case kMsgQuit: {
+ // TODO(jar): bug 1300541 The following assert should be used, but
+ // currently too much code actually triggers the assert, especially in
+ // tests :-(.
+ // Discarding a second quit will cause a hang.
+ //CHECK(!quit_received_);
+ quit_received_ = true;
+ return 0;
+ }
+ }
}
- // Since the incoming_queue_ may contain a task that destroys this message
- // loop, we cannot exit incoming_queue_lock_ until we are done with |this|.
- // We use a stack-based reference to the message pump so that we can call
- // ScheduleWork outside of incoming_queue_lock_.
+ return ::DefWindowProc(hwnd, message, wparam, lparam);
+}
+
+void MessageLoop::WillProcessMessage(const MSG& msg) {
+ FOR_EACH_OBSERVER(Observer, observers_, WillProcessMessage(msg));
+}
- pump->ScheduleWork();
+void MessageLoop::DidProcessMessage(const MSG& msg) {
+ FOR_EACH_OBSERVER(Observer, observers_, DidProcessMessage(msg));
}
void MessageLoop::SetNestableTasksAllowed(bool allowed) {
@@ -252,7 +442,7 @@ void MessageLoop::SetNestableTasksAllowed(bool allowed) {
if (!nestable_tasks_allowed_)
return;
// Start the native pump if we are not already pumping.
- pump_->ScheduleWork();
+ EnsurePumpATaskWasPosted();
}
}
@@ -260,7 +450,215 @@ bool MessageLoop::NestableTasksAllowed() const {
return nestable_tasks_allowed_;
}
+bool MessageLoop::ProcessNextWindowsMessage() {
+ MSG msg;
+ if (PeekMessage(&msg, NULL, 0, 0, PM_REMOVE)) {
+ return ProcessMessageHelper(msg);
+ }
+ return false;
+}
+
+bool MessageLoop::ProcessMessageHelper(const MSG& msg) {
+ HistogramEvent(msg.message);
+
+ if (WM_QUIT == msg.message) {
+ // Repost the QUIT message so that it will be retrieved by the primary
+ // GetMessage() loop.
+ quit_now_ = true;
+ PostQuitMessage(static_cast<int>(msg.wParam));
+ return false;
+ }
+
+ // While running our main message pump, we discard kMsgPumpATask messages.
+ if (msg.message == kMsgPumpATask && msg.hwnd == message_hwnd_)
+ return ProcessPumpReplacementMessage();
+
+ WillProcessMessage(msg);
+
+ if (dispatcher_) {
+ if (!dispatcher_->Dispatch(msg))
+ quit_now_ = true;
+ } else {
+ TranslateMessage(&msg);
+ DispatchMessage(&msg);
+ }
+
+ DidProcessMessage(msg);
+ return true;
+}
+
+bool MessageLoop::ProcessPumpReplacementMessage() {
+ MSG msg;
+ bool have_message = (0 != PeekMessage(&msg, NULL, 0, 0, PM_REMOVE));
+ DCHECK(!have_message || kMsgPumpATask != msg.message
+ || msg.hwnd != message_hwnd_);
+ {
+ // Since we discarded a kMsgPumpATask message, we must update the flag.
+ AutoLock lock(task_pump_message_lock_);
+ DCHECK(task_pump_message_pending_);
+ task_pump_message_pending_ = false;
+ }
+ return have_message && ProcessMessageHelper(msg);
+}
+
+// Create a mini-message-pump to force immediate processing of only Windows
+// WM_PAINT messages.
+void MessageLoop::PumpOutPendingPaintMessages() {
+ // Don't provide an infinite loop, but do enough peeking to get the job done.
+ // Actual common max is 4 peeks, but we'll be a little safe here.
+ const int kMaxPeekCount = 20;
+ int peek_count;
+ bool win2k(true);
+ if (win_util::GetWinVersion() > win_util::WINVERSION_2000)
+ win2k = false;
+ for (peek_count = 0; peek_count < kMaxPeekCount; ++peek_count) {
+ MSG msg;
+ if (win2k) {
+ if (!PeekMessage(&msg, NULL, WM_PAINT, WM_PAINT, PM_REMOVE))
+ break;
+ } else {
+ if (!PeekMessage(&msg, NULL, 0, 0, PM_REMOVE | PM_QS_PAINT))
+ break;
+ }
+ ProcessMessageHelper(msg);
+ if (quit_now_ ) // Handle WM_QUIT.
+ break;
+ }
+ // Histogram what was really being used, to help to adjust kMaxPeekCount.
+ DHISTOGRAM_COUNTS(L"Loop.PumpOutPendingPaintMessages Peeks", peek_count);
+}
+
//------------------------------------------------------------------------------
+// If we handle more than the OS limit on the number of objects that can be
+// waited for, we'll need to poll (sequencing through subsets of the objects
+// that can be passed in a single OS wait call). The following is the polling
+// interval used in that (unusual) case. (I don't have a lot of justifcation
+// for the specific value, but it needed to be short enough that it would not
+// add a lot of latency, and long enough that we wouldn't thrash the CPU for no
+// reason... especially considering the silly user probably has a million tabs
+// open, etc.)
+static const int kMultipleWaitPollingInterval = 20;
+
+void MessageLoop::WaitForWork() {
+ bool original_can_run = nestable_tasks_allowed_;
+ int wait_flags = original_can_run ? MWMO_ALERTABLE | MWMO_INPUTAVAILABLE
+ : MWMO_INPUTAVAILABLE;
+
+ bool use_polling = false; // Poll if too many objects for one OS Wait call.
+ for (;;) {
+ // Do initialization here, in case APC modifies object list.
+ size_t total_objs = original_can_run ? objects_.size() : 0;
+
+ int delay;
+ size_t polling_index = 0; // The first unprocessed object index.
+ do {
+ size_t objs_len =
+ (polling_index < total_objs) ? total_objs - polling_index : 0;
+ if (objs_len >= CHROME_MAXIMUM_WAIT_OBJECTS) {
+ objs_len = CHROME_MAXIMUM_WAIT_OBJECTS - 1;
+ use_polling = true;
+ }
+ HANDLE* objs = objs_len ? polling_index + &objects_.front() : NULL;
+
+ // Only wait up to the time needed by the timer manager to fire the next
+ // set of timers.
+ delay = timer_manager_.GetCurrentDelay();
+ if (use_polling && delay > kMultipleWaitPollingInterval)
+ delay = kMultipleWaitPollingInterval;
+ if (delay < 0) // Negative value means no timers waiting.
+ delay = INFINITE;
+
+ DWORD result;
+ result = MsgWaitForMultipleObjectsEx(static_cast<DWORD>(objs_len), objs,
+ delay, QS_ALLINPUT, wait_flags);
+
+ if (WAIT_IO_COMPLETION == result) {
+ HistogramEvent(kSleepingApcEvent);
+ // We'll loop here when we service an APC. At it currently stands,
+ // *ONLY* the IO thread uses *any* APCs, so this should have no impact
+ // on the UI thread.
+ break; // Break to outer loop, and waitforwork() again.
+ }
+
+ // Use unsigned type to simplify range detection;
+ size_t signaled_index = result - WAIT_OBJECT_0;
+ if (signaled_index < objs_len) {
+ SignalWatcher(polling_index + signaled_index);
+ HistogramEvent(kSleepingSignalEvent);
+ return; // We serviced a signaled object.
+ }
+
+ if (objs_len == signaled_index)
+ return; // A WM_* message is available.
+
+ DCHECK_NE(WAIT_FAILED, result) << GetLastError();
+
+ DCHECK(!objs || result == WAIT_TIMEOUT);
+ if (!use_polling)
+ return;
+ polling_index += objs_len;
+ } while (polling_index < total_objs);
+ // For compatibility, we didn't return sooner. This made us do *some* wait
+ // call(s) before returning. This will probably change in next rev.
+ if (!delay || !timer_manager_.GetCurrentDelay())
+ return; // No work done, but timer is ready to fire.
+ }
+}
+
+// Note: MsgWaitMultipleObjects() can't take a nil list, and that is why I had
+// to use SleepEx() to handle APCs when there were no objects.
+bool MessageLoop::ProcessNextObject() {
+ if (!nestable_tasks_allowed_)
+ return false;
+
+ size_t total_objs = objects_.size();
+ if (!total_objs) {
+ return false;
+ }
+
+ size_t polling_index = 0; // The first unprocessed object index.
+ do {
+ DCHECK(polling_index < total_objs);
+ size_t objs_len = total_objs - polling_index;
+ if (objs_len >= CHROME_MAXIMUM_WAIT_OBJECTS)
+ objs_len = CHROME_MAXIMUM_WAIT_OBJECTS - 1;
+ HANDLE* objs = polling_index + &objects_.front();
+
+ // Identify 1 pending object, or allow an IO APC to be completed.
+ DWORD result = WaitForMultipleObjectsEx(static_cast<DWORD>(objs_len), objs,
+ FALSE, // 1 signal is sufficient.
+ 0, // Wait 0ms.
+ false); // Not alertable (no APC).
+
+ // Use unsigned type to simplify range detection;
+ size_t signaled_index = result - WAIT_OBJECT_0;
+ if (signaled_index < objs_len) {
+ SignalWatcher(polling_index + signaled_index);
+ HistogramEvent(kPollingSignalEvent);
+ return true; // We serviced a signaled object.
+ }
+
+ // If an handle is invalid, it will be WAIT_FAILED.
+ DCHECK_EQ(WAIT_TIMEOUT, result) << GetLastError();
+ polling_index += objs_len;
+ } while (polling_index < total_objs);
+ return false; // We serviced nothing.
+}
+
+bool MessageLoop::SignalWatcher(size_t object_index) {
+ BeforeTaskRunSetup();
+ DCHECK(objects_.size() > object_index);
+ // On reception of OnObjectSignaled() to a Watcher object, it may call
+ // WatchObject(). watchers_ and objects_ will be modified. This is
+ // expected, so don't be afraid if, while tracing a OnObjectSignaled()
+ // function, the corresponding watchers_[result] is inexistant.
+ watchers_[object_index]->OnObjectSignaled(objects_[object_index]);
+ // Signaled objects tend to be removed from the watch list, and then added
+ // back (appended). As a result, they move to the end of the objects_ array,
+ // and this should make their service "fair" (no HANDLEs should be starved).
+ AfterTaskRunRestore();
+ return true;
+}
bool MessageLoop::RunTimerTask(Timer* timer) {
HistogramEvent(kTimerEvent);
@@ -307,15 +705,16 @@ bool MessageLoop::QueueOrRunTask(Task* new_task) {
// Execute oldest task.
while (!work_queue_.Empty()) {
Task* task = work_queue_.Pop();
- if (task->nestable() || state_->run_depth == 1) {
+ if (task->nestable() || run_depth_ == 1) {
RunTask(task);
// Show that we ran a task (Note: a new one might arrive as a
// consequence!).
return true;
+ } else {
+ // We couldn't run the task now because we're in a nested message loop
+ // and the task isn't nestable.
+ delayed_non_nestable_queue_.Push(task);
}
- // We couldn't run the task now because we're in a nested message loop
- // and the task isn't nestable.
- delayed_non_nestable_queue_.Push(task);
}
// Nothing happened.
@@ -344,12 +743,54 @@ void MessageLoop::AfterTaskRunRestore() {
nestable_tasks_allowed_ = true;
}
+void MessageLoop::PumpATaskDuringWndProc() {
+ // TODO(jar): Perchance we should check on signaled objects here??
+ // Signals are generally starved during a native message loop. Even if we
+ // try to service a signaled object now, we wouldn't automatically get here
+ // (i.e., the native pump would not re-start) when the next object was
+ // signaled. If we really want to avoid starving signaled objects, we need
+ // to translate them into Tasks that can be passed in via PostTask.
+ // If these native message loops (and sub-pumping activities) are short
+ // lived, then the starvation won't be that long :-/.
+
+ if (!ProcessNextDeferredTask())
+ return; // Nothing to do, so lets stop the sub-pump.
+
+ // We ran a task, so make sure we come back and try to run more tasks.
+ EnsurePumpATaskWasPosted();
+}
+
+void MessageLoop::EnsurePumpATaskWasPosted() {
+ {
+ AutoLock lock(task_pump_message_lock_);
+ if (task_pump_message_pending_)
+ return; // Someone else continued the pumping.
+ task_pump_message_pending_ = true; // We'll send one.
+ }
+ EnsureMessageGetsPosted(kMsgPumpATask);
+}
+
+void MessageLoop::EnsureMessageGetsPosted(int message) const {
+ const int kRetryCount = 30;
+ const int kSleepDurationWhenFailing = 100;
+ for (int i = 0; i < kRetryCount; ++i) {
+ // Posting to our own windows should always succeed. If it doesn't we're in
+ // big trouble.
+ if (PostMessage(message_hwnd_, message, 0, 0))
+ return;
+ Sleep(kSleepDurationWhenFailing);
+ }
+ LOG(FATAL) << "Crash with last error " << GetLastError();
+ int* p = NULL;
+ *p = 0; // Crash.
+}
+
void MessageLoop::ReloadWorkQueue() {
// We can improve performance of our loading tasks from incoming_queue_ to
- // work_queue_ by waiting until the last minute (work_queue_ is empty) to
- // load. That reduces the number of locks-per-task significantly when our
- // queues get large. The optimization is disabled on threads that make use
- // of the priority queue (prioritization requires all our tasks to be in the
+ // work_queue_ by wating until the last minute (work_queue_ is empty) to load.
+ // That reduces the number of locks-per-task significantly when our queues get
+ // large. The optimization is disabled on threads that make use of the
+ // priority queue (prioritization requires all our tasks to be in the
// work_queue_ ASAP).
if (!work_queue_.Empty() && !work_queue_.use_priority_queue())
return; // Wait till we *really* need to lock and load.
@@ -392,78 +833,42 @@ void MessageLoop::DeletePendingTasks() {
}
void MessageLoop::DidChangeNextTimerExpiry() {
+#if defined(OS_WIN)
+ //
+ // We would *like* to provide high resolution timers. Windows timers using
+ // SetTimer() have a 10ms granularity. We have to use WM_TIMER as a wakeup
+ // mechanism because the application can enter modal windows loops where it
+ // is not running our MessageLoop; the only way to have our timers fire in
+ // these cases is to post messages there.
+ //
+ // To provide sub-10ms timers, we process timers directly from our run loop.
+ // For the common case, timers will be processed there as the run loop does
+ // its normal work. However, we *also* set the system timer so that WM_TIMER
+ // events fire. This mops up the case of timers not being able to work in
+ // modal message loops. It is possible for the SetTimer to pop and have no
+ // pending timers, because they could have already been processed by the
+ // run loop itself.
+ //
+ // We use a single SetTimer corresponding to the timer that will expire
+ // soonest. As new timers are created and destroyed, we update SetTimer.
+ // Getting a spurrious SetTimer event firing is benign, as we'll just be
+ // processing an empty timer queue.
+ //
int delay = timer_manager_.GetCurrentDelay();
- if (delay == -1)
- return;
-
- // Simulates malfunctioning, early firing timers. Pending tasks should only
- // be invoked when the delay they specify has elapsed.
- if (timer_manager_.use_broken_delay())
- delay = 10;
-
- pump_->ScheduleDelayedWork(TimeDelta::FromMilliseconds(delay));
-}
-
-bool MessageLoop::DoWork() {
- if (ProcessNextDeferredTask()) {
- // TODO(darin): Temporarily moved into MessagePumpWin::HandleWorkMessage.
-#if 0
- // Let the MessagePump know that we may have more work to do. It is
- // tempting to only call this function if our work queue is not empty, but
- // doing so is insufficient. The issue is that ReloadWorkQueue does not
- // look at the incoming queue if the work queue is not empty, and we may
- // have just processed the last task on the work queue. ScheduleWork is
- // very low cost (in the case where it is redundant), so we don't worry
- // about optimizing away spurious ScheduleWork calls.
- pump_->ScheduleWork();
-#endif
- return true;
- }
- return false;
-}
-
-bool MessageLoop::DoDelayedWork() {
- bool did_work = ProcessSomeTimers();
-
- // We may not have run any timers, but we may still have future timers to
- // run, so we need to inform the pump again of pending timers.
- DidChangeNextTimerExpiry();
-
- return did_work;
-}
-
-bool MessageLoop::DoIdleWork() {
- if (ProcessNextDelayedNonNestableTask())
- return true;
-
- if (state_->quit_received)
- pump_->Quit();
-
- return false;
-}
-
-//------------------------------------------------------------------------------
-// MessageLoop::AutoRunState
-
-MessageLoop::AutoRunState::AutoRunState(MessageLoop* loop) : loop_(loop) {
- // Make the loop reference us.
- previous_state_ = loop_->state_;
- if (previous_state_) {
- run_depth = previous_state_->run_depth + 1;
+ if (delay == -1) {
+ KillTimer(message_hwnd_, reinterpret_cast<UINT_PTR>(this));
} else {
- run_depth = 1;
+ if (delay < USER_TIMER_MINIMUM)
+ delay = USER_TIMER_MINIMUM;
+ // Simulates malfunctioning, early firing timers. Pending tasks should only
+ // be invoked when the delay they specify has elapsed.
+ if (timer_manager_.use_broken_delay())
+ delay = 10;
+ // Create a WM_TIMER event that will wake us up to check for any pending
+ // timers (in case we are running within a nested, external sub-pump).
+ SetTimer(message_hwnd_, reinterpret_cast<UINT_PTR>(this), delay, NULL);
}
- loop_->state_ = this;
-
- // Initialize the other fields:
- quit_received = false;
-#if defined(OS_WIN)
- dispatcher = NULL;
-#endif
-}
-
-MessageLoop::AutoRunState::~AutoRunState() {
- loop_->state_ = previous_state_;
+#endif // defined(OS_WIN)
}
//------------------------------------------------------------------------------
@@ -558,7 +963,6 @@ void MessageLoop::EnableHistogrammer(bool enable) {
void MessageLoop::StartHistogrammer() {
if (enable_histogrammer_ && !message_histogram_.get()
&& StatisticsRecorder::WasStarted()) {
- DCHECK(!thread_name_.empty());
message_histogram_.reset(new LinearHistogram(
ASCIIToWide("MsgLoop:" + thread_name_).c_str(),
kLeastNonZeroMessageId,
@@ -574,6 +978,11 @@ void MessageLoop::HistogramEvent(int event) {
message_histogram_->Add(event);
}
+// Add one undocumented windows message to clean up our display.
+#ifndef WM_SYSTIMER
+#define WM_SYSTIMER 0x118
+#endif
+
// Provide a macro that takes an expression (such as a constant, or macro
// constant) and creates a pair to initalize an array of pairs. In this case,
// our pair consists of the expressions value, and the "stringized" version
@@ -590,13 +999,33 @@ void MessageLoop::HistogramEvent(int event) {
// in the pair (i.e., the quoted string) when printing out a histogram.
#define VALUE_TO_NUMBER_AND_NAME(name) {name, #name},
+
// static
const LinearHistogram::DescriptionPair MessageLoop::event_descriptions_[] = {
+ // Only provide an extensive list in debug mode. In release mode, we have to
+ // read the octal values.... but we save about 450 strings, each of length
+ // 10 from our binary image.
+#ifndef NDEBUG
+ // Prepare to include a list of names provided in a special header file4.
+#define A_NAMED_MESSAGE_FROM_WINUSER_H VALUE_TO_NUMBER_AND_NAME
+#include "base/windows_message_list.h"
+#undef A_NAMED_MESSAGE_FROM_WINUSER_H
+ // Add an undocumented message that appeared in our list :-/.
+ VALUE_TO_NUMBER_AND_NAME(WM_SYSTIMER)
+#endif // NDEBUG
+
// Provide some pretty print capability in our histogram for our internal
// messages.
+ // Values we use for WM_USER+n
+ VALUE_TO_NUMBER_AND_NAME(kMsgPumpATask)
+ VALUE_TO_NUMBER_AND_NAME(kMsgQuit)
+
// A few events we handle (kindred to messages), and used to profile actions.
VALUE_TO_NUMBER_AND_NAME(kTaskRunEvent)
+ VALUE_TO_NUMBER_AND_NAME(kSleepingApcEvent)
+ VALUE_TO_NUMBER_AND_NAME(kSleepingSignalEvent)
+ VALUE_TO_NUMBER_AND_NAME(kPollingSignalEvent)
VALUE_TO_NUMBER_AND_NAME(kTimerEvent)
{-1, NULL} // The list must be null terminated, per API to histogram.