gecko/ipc/glue/MessagePump.cpp
Kyle Huey 0a62d4d460 Bug 1179909: Refactor stable state handling. r=smaug
This is motivated by three separate but related problems:

1. Our concept of recursion depth is broken for things that run from AfterProcessNextEvent observers (e.g. Promises). We decrement the recursionDepth counter before firing observers, so a Promise callback running at the lowest event loop depth has a recursion depth of 0 (whereas a regular nsIRunnable would be 1). This is a problem because it's impossible to distinguish a Promise running after a sync XHR's onreadystatechange handler from a top-level event (since the former runs with depth 2 - 1 = 1, and the latter runs with just 1).

2. The nsIThreadObserver mechanism that is used by a lot of code to run "after" the current event is a poor fit for anything that runs script. First, the order the observers fire in is the order they were added, not anything fixed by spec. Additionally, running script can cause the event loop to spin, which is a big source of pain here (bholley has some nasty bug caused by this).

3. We run Promises from different points in the code for workers and main thread. The latter runs from XPConnect's nsIThreadObserver callbacks, while the former runs from a hardcoded call to run Promises in the worker event loop. What workers do is particularly problematic because it means we can't get the right recursion depth no matter what we do to nsThread.

The solve this, this patch does the following:

1. Consolidate some handling of microtasks and all handling of stable state from appshell and WorkerPrivate into CycleCollectedJSRuntime.
2. Make the recursionDepth counter only available to CycleCollectedJSRuntime (and its consumers) and remove it from the nsIThreadInternal and nsIThreadObserver APIs.
3. Adjust the recursionDepth counter so that microtasks run with the recursionDepth of the task they are associated with.
4. Introduce the concept of metastable state to replace appshell's RunBeforeNextEvent. Metastable state is reached after every microtask or task is completed. This provides the semantics that bent and I want for IndexedDB, where transactions autocommit at the end of a microtask and do not "spill" from one microtask into a subsequent microtask. This differs from appshell's RunBeforeNextEvent in two ways:
a) It fires between microtasks, which was the motivation for starting this.
b) It no longer ensures that we're at the same event loop depth in the native event queue. bent decided we don't care about this.
5. Reorder stable state to happen after microtasks such as Promises, per HTML. Right now we call the regular thread observers, including appshell, before the main thread observer (XPConnect), so stable state tasks happen before microtasks.
2015-08-11 06:10:46 -07:00

449 lines
12 KiB
C++

/* 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/. */
#include "MessagePump.h"
#include "nsIRunnable.h"
#include "nsIThread.h"
#include "nsITimer.h"
#include "nsICancelableRunnable.h"
#include "base/basictypes.h"
#include "base/logging.h"
#include "base/scoped_nsautorelease_pool.h"
#include "mozilla/Assertions.h"
#include "mozilla/DebugOnly.h"
#include "nsComponentManagerUtils.h"
#include "nsDebug.h"
#include "nsServiceManagerUtils.h"
#include "nsString.h"
#include "nsThreadUtils.h"
#include "nsTimerImpl.h"
#include "nsXULAppAPI.h"
#include "prthread.h"
#ifdef MOZ_NUWA_PROCESS
#include "ipc/Nuwa.h"
#endif
using base::TimeTicks;
using namespace mozilla::ipc;
NS_DEFINE_NAMED_CID(NS_TIMER_CID);
static mozilla::DebugOnly<MessagePump::Delegate*> gFirstDelegate;
namespace mozilla {
namespace ipc {
class DoWorkRunnable final : public nsICancelableRunnable,
public nsITimerCallback
{
public:
explicit DoWorkRunnable(MessagePump* aPump)
: mPump(aPump)
{
MOZ_ASSERT(aPump);
}
NS_DECL_THREADSAFE_ISUPPORTS
NS_DECL_NSIRUNNABLE
NS_DECL_NSITIMERCALLBACK
NS_DECL_NSICANCELABLERUNNABLE
private:
~DoWorkRunnable()
{ }
MessagePump* mPump;
// DoWorkRunnable is designed as a stateless singleton. Do not add stateful
// members here!
};
} /* namespace ipc */
} /* namespace mozilla */
MessagePump::MessagePump()
: mThread(nullptr)
{
mDoWorkEvent = new DoWorkRunnable(this);
}
MessagePump::~MessagePump()
{
}
void
MessagePump::Run(MessagePump::Delegate* aDelegate)
{
MOZ_ASSERT(keep_running_);
MOZ_ASSERT(NS_IsMainThread(),
"Use mozilla::ipc::MessagePumpForNonMainThreads instead!");
mThread = NS_GetCurrentThread();
MOZ_ASSERT(mThread);
mDelayedWorkTimer = do_CreateInstance(kNS_TIMER_CID);
MOZ_ASSERT(mDelayedWorkTimer);
base::ScopedNSAutoreleasePool autoReleasePool;
for (;;) {
autoReleasePool.Recycle();
bool did_work = NS_ProcessNextEvent(mThread, false) ? true : false;
if (!keep_running_)
break;
// NB: it is crucial *not* to directly call |aDelegate->DoWork()|
// here. To ensure that MessageLoop tasks and XPCOM events have
// equal priority, we sensitively rely on processing exactly one
// Task per DoWorkRunnable XPCOM event.
did_work |= aDelegate->DoDelayedWork(&delayed_work_time_);
if (did_work && delayed_work_time_.is_null()
#ifdef MOZ_NUWA_PROCESS
&& (!IsNuwaReady() || !IsNuwaProcess())
#endif
)
mDelayedWorkTimer->Cancel();
if (!keep_running_)
break;
if (did_work)
continue;
did_work = aDelegate->DoIdleWork();
if (!keep_running_)
break;
if (did_work)
continue;
// This will either sleep or process an event.
NS_ProcessNextEvent(mThread, true);
}
#ifdef MOZ_NUWA_PROCESS
if (!IsNuwaReady() || !IsNuwaProcess())
#endif
mDelayedWorkTimer->Cancel();
keep_running_ = true;
}
void
MessagePump::ScheduleWork()
{
// Make sure the event loop wakes up.
if (mThread) {
mThread->Dispatch(mDoWorkEvent, NS_DISPATCH_NORMAL);
}
else {
// Some things (like xpcshell) don't use the app shell and so Run hasn't
// been called. We still need to wake up the main thread.
NS_DispatchToMainThread(mDoWorkEvent);
}
event_.Signal();
}
void
MessagePump::ScheduleWorkForNestedLoop()
{
// This method is called when our MessageLoop has just allowed
// nested tasks. In our setup, whenever that happens we know that
// DoWork() will be called "soon", so there's no need to pay the
// cost of what will be a no-op nsThread::Dispatch(mDoWorkEvent).
}
void
MessagePump::ScheduleDelayedWork(const base::TimeTicks& aDelayedTime)
{
#ifdef MOZ_NUWA_PROCESS
if (IsNuwaReady() && IsNuwaProcess())
return;
#endif
if (!mDelayedWorkTimer) {
mDelayedWorkTimer = do_CreateInstance(kNS_TIMER_CID);
if (!mDelayedWorkTimer) {
// Called before XPCOM has started up? We can't do this correctly.
NS_WARNING("Delayed task might not run!");
delayed_work_time_ = aDelayedTime;
return;
}
}
if (!delayed_work_time_.is_null()) {
mDelayedWorkTimer->Cancel();
}
delayed_work_time_ = aDelayedTime;
// TimeDelta's constructor initializes to 0
base::TimeDelta delay;
if (aDelayedTime > base::TimeTicks::Now())
delay = aDelayedTime - base::TimeTicks::Now();
uint32_t delayMS = uint32_t(delay.InMilliseconds());
mDelayedWorkTimer->InitWithCallback(mDoWorkEvent, delayMS,
nsITimer::TYPE_ONE_SHOT);
}
void
MessagePump::DoDelayedWork(base::MessagePump::Delegate* aDelegate)
{
aDelegate->DoDelayedWork(&delayed_work_time_);
if (!delayed_work_time_.is_null()) {
ScheduleDelayedWork(delayed_work_time_);
}
}
NS_IMPL_ISUPPORTS(DoWorkRunnable, nsIRunnable, nsITimerCallback,
nsICancelableRunnable)
NS_IMETHODIMP
DoWorkRunnable::Run()
{
MessageLoop* loop = MessageLoop::current();
MOZ_ASSERT(loop);
bool nestableTasksAllowed = loop->NestableTasksAllowed();
// MessageLoop::RunTask() disallows nesting, but our Frankenventloop will
// always dispatch DoWork() below from what looks to MessageLoop like a nested
// context. So we unconditionally allow nesting here.
loop->SetNestableTasksAllowed(true);
loop->DoWork();
loop->SetNestableTasksAllowed(nestableTasksAllowed);
return NS_OK;
}
NS_IMETHODIMP
DoWorkRunnable::Notify(nsITimer* aTimer)
{
MessageLoop* loop = MessageLoop::current();
MOZ_ASSERT(loop);
mPump->DoDelayedWork(loop);
return NS_OK;
}
NS_IMETHODIMP
DoWorkRunnable::Cancel()
{
// Workers require cancelable runnables, but we can't really cancel cleanly
// here. If we don't process this runnable then we will leave something
// unprocessed in the message_loop. Therefore, eagerly complete our work
// instead by immediately calling Run(). Run() should be called separately
// after this. Unfortunately we cannot use flags to verify this because
// DoWorkRunnable is a stateless singleton that can be in the event queue
// multiple times simultaneously.
MOZ_ALWAYS_TRUE(NS_SUCCEEDED(Run()));
return NS_OK;
}
void
MessagePumpForChildProcess::Run(base::MessagePump::Delegate* aDelegate)
{
if (mFirstRun) {
MOZ_ASSERT(aDelegate && !gFirstDelegate);
gFirstDelegate = aDelegate;
mFirstRun = false;
if (NS_FAILED(XRE_RunAppShell())) {
NS_WARNING("Failed to run app shell?!");
}
MOZ_ASSERT(aDelegate && aDelegate == gFirstDelegate);
gFirstDelegate = nullptr;
return;
}
MOZ_ASSERT(aDelegate && aDelegate == gFirstDelegate);
// We can get to this point in startup with Tasks in our loop's
// incoming_queue_ or pending_queue_, but without a matching
// DoWorkRunnable(). In MessagePump::Run() above, we sensitively
// depend on *not* directly calling delegate->DoWork(), because that
// prioritizes Tasks above XPCOM events. However, from this point
// forward, any Task posted to our loop is guaranteed to have a
// DoWorkRunnable enqueued for it.
//
// So we just flush the pending work here and move on.
MessageLoop* loop = MessageLoop::current();
bool nestableTasksAllowed = loop->NestableTasksAllowed();
loop->SetNestableTasksAllowed(true);
while (aDelegate->DoWork());
loop->SetNestableTasksAllowed(nestableTasksAllowed);
// Really run.
mozilla::ipc::MessagePump::Run(aDelegate);
}
void
MessagePumpForNonMainThreads::Run(base::MessagePump::Delegate* aDelegate)
{
MOZ_ASSERT(keep_running_);
MOZ_ASSERT(!NS_IsMainThread(), "Use mozilla::ipc::MessagePump instead!");
mThread = NS_GetCurrentThread();
MOZ_ASSERT(mThread);
mDelayedWorkTimer = do_CreateInstance(kNS_TIMER_CID);
MOZ_ASSERT(mDelayedWorkTimer);
if (NS_FAILED(mDelayedWorkTimer->SetTarget(mThread))) {
MOZ_CRASH("Failed to set timer target!");
}
// Chromium event notifications to be processed will be received by this
// event loop as a DoWorkRunnables via ScheduleWork. Chromium events that
// were received before our mThread is valid, however, will not generate
// runnable wrappers. We must process any of these before we enter this
// loop, or we will forever have unprocessed chromium messages in our queue.
//
// Note we would like to request a flush of the chromium event queue
// using a runnable on the xpcom side, but some thread implementations
// (dom workers) get cranky if we call ScheduleWork here (ScheduleWork
// calls dispatch on mThread) before the thread processes an event. As
// such, clear the queue manually.
while (aDelegate->DoWork()) {
}
base::ScopedNSAutoreleasePool autoReleasePool;
for (;;) {
autoReleasePool.Recycle();
bool didWork = NS_ProcessNextEvent(mThread, false) ? true : false;
if (!keep_running_) {
break;
}
didWork |= aDelegate->DoDelayedWork(&delayed_work_time_);
if (didWork && delayed_work_time_.is_null()) {
mDelayedWorkTimer->Cancel();
}
if (!keep_running_) {
break;
}
if (didWork) {
continue;
}
didWork = aDelegate->DoIdleWork();
if (!keep_running_) {
break;
}
if (didWork) {
continue;
}
// This will either sleep or process an event.
NS_ProcessNextEvent(mThread, true);
}
mDelayedWorkTimer->Cancel();
keep_running_ = true;
}
#if defined(XP_WIN)
NS_IMPL_QUERY_INTERFACE(MessagePumpForNonMainUIThreads, nsIThreadObserver)
#define CHECK_QUIT_STATE { if (state_->should_quit) { break; } }
void MessagePumpForNonMainUIThreads::DoRunLoop()
{
// If this is a chromium thread and no nsThread is associated
// with it, this call will create a new nsThread.
mThread = NS_GetCurrentThread();
MOZ_ASSERT(mThread);
// Set the main thread observer so we can wake up when
// xpcom events need to get processed.
nsCOMPtr<nsIThreadInternal> ti(do_QueryInterface(mThread));
MOZ_ASSERT(ti);
ti->SetObserver(this);
base::ScopedNSAutoreleasePool autoReleasePool;
for (;;) {
autoReleasePool.Recycle();
bool didWork = NS_ProcessNextEvent(mThread, false);
didWork |= ProcessNextWindowsMessage();
CHECK_QUIT_STATE
didWork |= state_->delegate->DoWork();
CHECK_QUIT_STATE
didWork |= state_->delegate->DoDelayedWork(&delayed_work_time_);
if (didWork && delayed_work_time_.is_null()) {
KillTimer(message_hwnd_, reinterpret_cast<UINT_PTR>(this));
}
CHECK_QUIT_STATE
if (didWork) {
continue;
}
didWork = state_->delegate->DoIdleWork();
CHECK_QUIT_STATE
SetInWait();
bool hasWork = NS_HasPendingEvents(mThread);
if (didWork || hasWork) {
ClearInWait();
continue;
}
WaitForWork(); // Calls MsgWaitForMultipleObjectsEx(QS_ALLINPUT)
ClearInWait();
}
ClearInWait();
ti->SetObserver(nullptr);
}
NS_IMETHODIMP
MessagePumpForNonMainUIThreads::OnDispatchedEvent(nsIThreadInternal *thread)
{
// If our thread is sleeping in DoRunLoop's call to WaitForWork() and an
// event posts to the nsIThread event queue - break our thread out of
// chromium's WaitForWork.
if (GetInWait()) {
ScheduleWork();
}
return NS_OK;
}
NS_IMETHODIMP
MessagePumpForNonMainUIThreads::OnProcessNextEvent(nsIThreadInternal *thread,
bool mayWait)
{
return NS_OK;
}
NS_IMETHODIMP
MessagePumpForNonMainUIThreads::AfterProcessNextEvent(nsIThreadInternal *thread,
bool eventWasProcessed)
{
return NS_OK;
}
#endif // XP_WIN