1 // Copyright (c) 2011 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
5 #include "base/synchronization/condition_variable.h"
10 #include "base/compiler_specific.h"
11 #include "base/logging.h"
12 #include "base/synchronization/lock.h"
13 #include "base/threading/thread_restrictions.h"
14 #include "base/time/time.h"
17 // We can't use the linker supported delay-load for kernel32 so all this
18 // cruft here is to manually late-bind the needed functions.
19 typedef void (WINAPI
*InitializeConditionVariableFn
)(PCONDITION_VARIABLE
);
20 typedef BOOL (WINAPI
*SleepConditionVariableCSFn
)(PCONDITION_VARIABLE
,
21 PCRITICAL_SECTION
, DWORD
);
22 typedef void (WINAPI
*WakeConditionVariableFn
)(PCONDITION_VARIABLE
);
23 typedef void (WINAPI
*WakeAllConditionVariableFn
)(PCONDITION_VARIABLE
);
25 InitializeConditionVariableFn initialize_condition_variable_fn
;
26 SleepConditionVariableCSFn sleep_condition_variable_fn
;
27 WakeConditionVariableFn wake_condition_variable_fn
;
28 WakeAllConditionVariableFn wake_all_condition_variable_fn
;
30 bool BindVistaCondVarFunctions() {
31 HMODULE kernel32
= GetModuleHandleA("kernel32.dll");
32 initialize_condition_variable_fn
=
33 reinterpret_cast<InitializeConditionVariableFn
>(
34 GetProcAddress(kernel32
, "InitializeConditionVariable"));
35 if (!initialize_condition_variable_fn
)
37 sleep_condition_variable_fn
=
38 reinterpret_cast<SleepConditionVariableCSFn
>(
39 GetProcAddress(kernel32
, "SleepConditionVariableCS"));
40 if (!sleep_condition_variable_fn
)
42 wake_condition_variable_fn
=
43 reinterpret_cast<WakeConditionVariableFn
>(
44 GetProcAddress(kernel32
, "WakeConditionVariable"));
45 if (!wake_condition_variable_fn
)
47 wake_all_condition_variable_fn
=
48 reinterpret_cast<WakeAllConditionVariableFn
>(
49 GetProcAddress(kernel32
, "WakeAllConditionVariable"));
50 if (!wake_all_condition_variable_fn
)
58 // Abstract base class of the pimpl idiom.
59 class ConditionVarImpl
{
61 virtual ~ConditionVarImpl() {};
62 virtual void Wait() = 0;
63 virtual void TimedWait(const TimeDelta
& max_time
) = 0;
64 virtual void Broadcast() = 0;
65 virtual void Signal() = 0;
68 ///////////////////////////////////////////////////////////////////////////////
69 // Windows Vista and Win7 implementation.
70 ///////////////////////////////////////////////////////////////////////////////
72 class WinVistaCondVar
: public ConditionVarImpl
{
74 WinVistaCondVar(Lock
* user_lock
);
75 ~WinVistaCondVar() {};
76 // Overridden from ConditionVarImpl.
77 virtual void Wait() OVERRIDE
;
78 virtual void TimedWait(const TimeDelta
& max_time
) OVERRIDE
;
79 virtual void Broadcast() OVERRIDE
;
80 virtual void Signal() OVERRIDE
;
83 base::Lock
& user_lock_
;
84 CONDITION_VARIABLE cv_
;
87 WinVistaCondVar::WinVistaCondVar(Lock
* user_lock
)
88 : user_lock_(*user_lock
) {
89 initialize_condition_variable_fn(&cv_
);
93 void WinVistaCondVar::Wait() {
94 TimedWait(TimeDelta::FromMilliseconds(INFINITE
));
97 void WinVistaCondVar::TimedWait(const TimeDelta
& max_time
) {
98 base::ThreadRestrictions::AssertWaitAllowed();
99 DWORD timeout
= static_cast<DWORD
>(max_time
.InMilliseconds());
100 CRITICAL_SECTION
* cs
= user_lock_
.lock_
.native_handle();
102 #if !defined(NDEBUG) || defined(DCHECK_ALWAYS_ON)
103 user_lock_
.CheckHeldAndUnmark();
106 if (FALSE
== sleep_condition_variable_fn(&cv_
, cs
, timeout
)) {
107 DCHECK(GetLastError() != WAIT_TIMEOUT
);
110 #if !defined(NDEBUG) || defined(DCHECK_ALWAYS_ON)
111 user_lock_
.CheckUnheldAndMark();
115 void WinVistaCondVar::Broadcast() {
116 wake_all_condition_variable_fn(&cv_
);
119 void WinVistaCondVar::Signal() {
120 wake_condition_variable_fn(&cv_
);
123 ///////////////////////////////////////////////////////////////////////////////
124 // Windows XP implementation.
125 ///////////////////////////////////////////////////////////////////////////////
127 class WinXPCondVar
: public ConditionVarImpl
{
129 WinXPCondVar(Lock
* user_lock
);
131 // Overridden from ConditionVarImpl.
132 virtual void Wait() OVERRIDE
;
133 virtual void TimedWait(const TimeDelta
& max_time
) OVERRIDE
;
134 virtual void Broadcast() OVERRIDE
;
135 virtual void Signal() OVERRIDE
;
137 // Define Event class that is used to form circularly linked lists.
138 // The list container is an element with NULL as its handle_ value.
139 // The actual list elements have a non-zero handle_ value.
140 // All calls to methods MUST be done under protection of a lock so that links
141 // can be validated. Without the lock, some links might asynchronously
142 // change, and the assertions would fail (as would list change operations).
145 // Default constructor with no arguments creates a list container.
149 // InitListElement transitions an instance from a container, to an element.
150 void InitListElement();
152 // Methods for use on lists.
153 bool IsEmpty() const;
154 void PushBack(Event
* other
);
158 // Methods for use on list elements.
160 HANDLE
handle() const;
161 // Pull an element from a list (if it's in one).
164 // Method for use on a list element or on a list.
165 bool IsSingleton() const;
168 // Provide pre/post conditions to validate correct manipulations.
169 bool ValidateAsDistinct(Event
* other
) const;
170 bool ValidateAsItem() const;
171 bool ValidateAsList() const;
172 bool ValidateLinks() const;
177 DISALLOW_COPY_AND_ASSIGN(Event
);
180 // Note that RUNNING is an unlikely number to have in RAM by accident.
181 // This helps with defensive destructor coding in the face of user error.
182 enum RunState
{ SHUTDOWN
= 0, RUNNING
= 64213 };
184 // Internal implementation methods supporting Wait().
185 Event
* GetEventForWaiting();
186 void RecycleEvent(Event
* used_event
);
190 // Private critical section for access to member data.
191 base::Lock internal_lock_
;
193 // Lock that is acquired before calling Wait().
194 base::Lock
& user_lock_
;
196 // Events that threads are blocked on.
199 // Free list for old events.
200 Event recycling_list_
;
201 int recycling_list_size_
;
203 // The number of allocated, but not yet deleted events.
204 int allocation_counter_
;
207 WinXPCondVar::WinXPCondVar(Lock
* user_lock
)
208 : user_lock_(*user_lock
),
210 allocation_counter_(0),
211 recycling_list_size_(0) {
215 WinXPCondVar::~WinXPCondVar() {
216 AutoLock
auto_lock(internal_lock_
);
217 run_state_
= SHUTDOWN
; // Prevent any more waiting.
219 DCHECK_EQ(recycling_list_size_
, allocation_counter_
);
220 if (recycling_list_size_
!= allocation_counter_
) { // Rare shutdown problem.
221 // There are threads of execution still in this->TimedWait() and yet the
222 // caller has instigated the destruction of this instance :-/.
223 // A common reason for such "overly hasty" destruction is that the caller
224 // was not willing to wait for all the threads to terminate. Such hasty
225 // actions are a violation of our usage contract, but we'll give the
226 // waiting thread(s) one last chance to exit gracefully (prior to our
228 // Note: waiting_list_ *might* be empty, but recycling is still pending.
229 AutoUnlock
auto_unlock(internal_lock_
);
230 Broadcast(); // Make sure all waiting threads have been signaled.
231 Sleep(10); // Give threads a chance to grab internal_lock_.
232 // All contained threads should be blocked on user_lock_ by now :-).
233 } // Reacquire internal_lock_.
235 DCHECK_EQ(recycling_list_size_
, allocation_counter_
);
238 void WinXPCondVar::Wait() {
239 // Default to "wait forever" timing, which means have to get a Signal()
240 // or Broadcast() to come out of this wait state.
241 TimedWait(TimeDelta::FromMilliseconds(INFINITE
));
244 void WinXPCondVar::TimedWait(const TimeDelta
& max_time
) {
245 base::ThreadRestrictions::AssertWaitAllowed();
246 Event
* waiting_event
;
249 AutoLock
auto_lock(internal_lock_
);
250 if (RUNNING
!= run_state_
) return; // Destruction in progress.
251 waiting_event
= GetEventForWaiting();
252 handle
= waiting_event
->handle();
254 } // Release internal_lock.
257 AutoUnlock
unlock(user_lock_
); // Release caller's lock
258 WaitForSingleObject(handle
, static_cast<DWORD
>(max_time
.InMilliseconds()));
259 // Minimize spurious signal creation window by recycling asap.
260 AutoLock
auto_lock(internal_lock_
);
261 RecycleEvent(waiting_event
);
262 // Release internal_lock_
263 } // Reacquire callers lock to depth at entry.
266 // Broadcast() is guaranteed to signal all threads that were waiting (i.e., had
267 // a cv_event internally allocated for them) before Broadcast() was called.
268 void WinXPCondVar::Broadcast() {
269 std::stack
<HANDLE
> handles
; // See FAQ-question-10.
271 AutoLock
auto_lock(internal_lock_
);
272 if (waiting_list_
.IsEmpty())
274 while (!waiting_list_
.IsEmpty())
275 // This is not a leak from waiting_list_. See FAQ-question 12.
276 handles
.push(waiting_list_
.PopBack()->handle());
277 } // Release internal_lock_.
278 while (!handles
.empty()) {
279 SetEvent(handles
.top());
284 // Signal() will select one of the waiting threads, and signal it (signal its
285 // cv_event). For better performance we signal the thread that went to sleep
286 // most recently (LIFO). If we want fairness, then we wake the thread that has
287 // been sleeping the longest (FIFO).
288 void WinXPCondVar::Signal() {
291 AutoLock
auto_lock(internal_lock_
);
292 if (waiting_list_
.IsEmpty())
293 return; // No one to signal.
294 // Only performance option should be used.
295 // This is not a leak from waiting_list. See FAQ-question 12.
296 handle
= waiting_list_
.PopBack()->handle(); // LIFO.
297 } // Release internal_lock_.
301 // GetEventForWaiting() provides a unique cv_event for any caller that needs to
302 // wait. This means that (worst case) we may over time create as many cv_event
303 // objects as there are threads simultaneously using this instance's Wait()
305 WinXPCondVar::Event
* WinXPCondVar::GetEventForWaiting() {
306 // We hold internal_lock, courtesy of Wait().
308 if (0 == recycling_list_size_
) {
309 DCHECK(recycling_list_
.IsEmpty());
310 cv_event
= new Event();
311 cv_event
->InitListElement();
312 allocation_counter_
++;
313 DCHECK(cv_event
->handle());
315 cv_event
= recycling_list_
.PopFront();
316 recycling_list_size_
--;
318 waiting_list_
.PushBack(cv_event
);
322 // RecycleEvent() takes a cv_event that was previously used for Wait()ing, and
323 // recycles it for use in future Wait() calls for this or other threads.
324 // Note that there is a tiny chance that the cv_event is still signaled when we
325 // obtain it, and that can cause spurious signals (if/when we re-use the
326 // cv_event), but such is quite rare (see FAQ-question-5).
327 void WinXPCondVar::RecycleEvent(Event
* used_event
) {
328 // We hold internal_lock, courtesy of Wait().
329 // If the cv_event timed out, then it is necessary to remove it from
330 // waiting_list_. If it was selected by Broadcast() or Signal(), then it is
332 used_event
->Extract(); // Possibly redundant
333 recycling_list_
.PushBack(used_event
);
334 recycling_list_size_
++;
336 //------------------------------------------------------------------------------
337 // The next section provides the implementation for the private Event class.
338 //------------------------------------------------------------------------------
340 // Event provides a doubly-linked-list of events for use exclusively by the
341 // ConditionVariable class.
343 // This custom container was crafted because no simple combination of STL
344 // classes appeared to support the functionality required. The specific
345 // unusual requirement for a linked-list-class is support for the Extract()
346 // method, which can remove an element from a list, potentially for insertion
347 // into a second list. Most critically, the Extract() method is idempotent,
348 // turning the indicated element into an extracted singleton whether it was
349 // contained in a list or not. This functionality allows one (or more) of
350 // threads to do the extraction. The iterator that identifies this extractable
351 // element (in this case, a pointer to the list element) can be used after
352 // arbitrary manipulation of the (possibly) enclosing list container. In
353 // general, STL containers do not provide iterators that can be used across
354 // modifications (insertions/extractions) of the enclosing containers, and
355 // certainly don't provide iterators that can be used if the identified
356 // element is *deleted* (removed) from the container.
358 // It is possible to use multiple redundant containers, such as an STL list,
359 // and an STL map, to achieve similar container semantics. This container has
360 // only O(1) methods, while the corresponding (multiple) STL container approach
361 // would have more complex O(log(N)) methods (yeah... N isn't that large).
362 // Multiple containers also makes correctness more difficult to assert, as
363 // data is redundantly stored and maintained, which is generally evil.
365 WinXPCondVar::Event::Event() : handle_(0) {
366 next_
= prev_
= this; // Self referencing circular.
369 WinXPCondVar::Event::~Event() {
371 // This is the list holder
373 Event
* cv_event
= PopFront();
374 DCHECK(cv_event
->ValidateAsItem());
378 DCHECK(IsSingleton());
380 int ret_val
= CloseHandle(handle_
);
385 // Change a container instance permanently into an element of a list.
386 void WinXPCondVar::Event::InitListElement() {
388 handle_
= CreateEvent(NULL
, false, false, NULL
);
392 // Methods for use on lists.
393 bool WinXPCondVar::Event::IsEmpty() const {
394 DCHECK(ValidateAsList());
395 return IsSingleton();
398 void WinXPCondVar::Event::PushBack(Event
* other
) {
399 DCHECK(ValidateAsList());
400 DCHECK(other
->ValidateAsItem());
401 DCHECK(other
->IsSingleton());
402 // Prepare other for insertion.
403 other
->prev_
= prev_
;
406 prev_
->next_
= other
;
408 DCHECK(ValidateAsDistinct(other
));
411 WinXPCondVar::Event
* WinXPCondVar::Event::PopFront() {
412 DCHECK(ValidateAsList());
413 DCHECK(!IsSingleton());
414 return next_
->Extract();
417 WinXPCondVar::Event
* WinXPCondVar::Event::PopBack() {
418 DCHECK(ValidateAsList());
419 DCHECK(!IsSingleton());
420 return prev_
->Extract();
423 // Methods for use on list elements.
425 HANDLE
WinXPCondVar::Event::handle() const {
426 DCHECK(ValidateAsItem());
430 // Pull an element from a list (if it's in one).
431 WinXPCondVar::Event
* WinXPCondVar::Event::Extract() {
432 DCHECK(ValidateAsItem());
433 if (!IsSingleton()) {
434 // Stitch neighbors together.
435 next_
->prev_
= prev_
;
436 prev_
->next_
= next_
;
437 // Make extractee into a singleton.
438 prev_
= next_
= this;
440 DCHECK(IsSingleton());
444 // Method for use on a list element or on a list.
445 bool WinXPCondVar::Event::IsSingleton() const {
446 DCHECK(ValidateLinks());
447 return next_
== this;
450 // Provide pre/post conditions to validate correct manipulations.
451 bool WinXPCondVar::Event::ValidateAsDistinct(Event
* other
) const {
452 return ValidateLinks() && other
->ValidateLinks() && (this != other
);
455 bool WinXPCondVar::Event::ValidateAsItem() const {
456 return (0 != handle_
) && ValidateLinks();
459 bool WinXPCondVar::Event::ValidateAsList() const {
460 return (0 == handle_
) && ValidateLinks();
463 bool WinXPCondVar::Event::ValidateLinks() const {
464 // Make sure both of our neighbors have links that point back to us.
465 // We don't do the O(n) check and traverse the whole loop, and instead only
466 // do a local check to (and returning from) our immediate neighbors.
467 return (next_
->prev_
== this) && (prev_
->next_
== this);
472 FAQ On WinXPCondVar subtle implementation details:
474 1) What makes this problem subtle? Please take a look at "Strategies
475 for Implementing POSIX Condition Variables on Win32" by Douglas
476 C. Schmidt and Irfan Pyarali.
477 http://www.cs.wustl.edu/~schmidt/win32-cv-1.html It includes
478 discussions of numerous flawed strategies for implementing this
479 functionality. I'm not convinced that even the final proposed
480 implementation has semantics that are as nice as this implementation
481 (especially with regard to Broadcast() and the impact on threads that
482 try to Wait() after a Broadcast() has been called, but before all the
483 original waiting threads have been signaled).
485 2) Why can't you use a single wait_event for all threads that call
486 Wait()? See FAQ-question-1, or consider the following: If a single
487 event were used, then numerous threads calling Wait() could release
488 their cs locks, and be preempted just before calling
489 WaitForSingleObject(). If a call to Broadcast() was then presented on
490 a second thread, it would be impossible to actually signal all
491 waiting(?) threads. Some number of SetEvent() calls *could* be made,
492 but there could be no guarantee that those led to to more than one
493 signaled thread (SetEvent()'s may be discarded after the first!), and
494 there could be no guarantee that the SetEvent() calls didn't just
495 awaken "other" threads that hadn't even started waiting yet (oops).
496 Without any limit on the number of requisite SetEvent() calls, the
497 system would be forced to do many such calls, allowing many new waits
498 to receive spurious signals.
500 3) How does this implementation cause spurious signal events? The
501 cause in this implementation involves a race between a signal via
502 time-out and a signal via Signal() or Broadcast(). The series of
503 actions leading to this are:
505 a) Timer fires, and a waiting thread exits the line of code:
507 WaitForSingleObject(waiting_event, max_time.InMilliseconds());
509 b) That thread (in (a)) is randomly pre-empted after the above line,
510 leaving the waiting_event reset (unsignaled) and still in the
513 c) A call to Signal() (or Broadcast()) on a second thread proceeds, and
514 selects the waiting cv_event (identified in step (b)) as the event to revive
515 via a call to SetEvent().
517 d) The Signal() method (step c) calls SetEvent() on waiting_event (step b).
519 e) The waiting cv_event (step b) is now signaled, but no thread is
522 f) When that waiting_event (step b) is reused, it will immediately
523 be signaled (spuriously).
526 4) Why do you recycle events, and cause spurious signals? First off,
527 the spurious events are very rare. They can only (I think) appear
528 when the race described in FAQ-question-3 takes place. This should be
529 very rare. Most(?) uses will involve only timer expiration, or only
530 Signal/Broadcast() actions. When both are used, it will be rare that
531 the race will appear, and it would require MANY Wait() and signaling
532 activities. If this implementation did not recycle events, then it
533 would have to create and destroy events for every call to Wait().
534 That allocation/deallocation and associated construction/destruction
535 would be costly (per wait), and would only be a rare benefit (when the
536 race was "lost" and a spurious signal took place). That would be bad
537 (IMO) optimization trade-off. Finally, such spurious events are
538 allowed by the specification of condition variables (such as
539 implemented in Vista), and hence it is better if any user accommodates
540 such spurious events (see usage note in condition_variable.h).
542 5) Why don't you reset events when you are about to recycle them, or
543 about to reuse them, so that the spurious signals don't take place?
544 The thread described in FAQ-question-3 step c may be pre-empted for an
545 arbitrary length of time before proceeding to step d. As a result,
546 the wait_event may actually be re-used *before* step (e) is reached.
547 As a result, calling reset would not help significantly.
549 6) How is it that the callers lock is released atomically with the
550 entry into a wait state? We commit to the wait activity when we
551 allocate the wait_event for use in a given call to Wait(). This
552 allocation takes place before the caller's lock is released (and
553 actually before our internal_lock_ is released). That allocation is
554 the defining moment when "the wait state has been entered," as that
555 thread *can* now be signaled by a call to Broadcast() or Signal().
556 Hence we actually "commit to wait" before releasing the lock, making
557 the pair effectively atomic.
559 8) Why do you need to lock your data structures during waiting, as the
560 caller is already in possession of a lock? We need to Acquire() and
561 Release() our internal lock during Signal() and Broadcast(). If we tried
562 to use a callers lock for this purpose, we might conflict with their
563 external use of the lock. For example, the caller may use to consistently
564 hold a lock on one thread while calling Signal() on another, and that would
567 9) Couldn't a more efficient implementation be provided if you
568 preclude using more than one external lock in conjunction with a
569 single ConditionVariable instance? Yes, at least it could be viewed
570 as a simpler API (since you don't have to reiterate the lock argument
571 in each Wait() call). One of the constructors now takes a specific
572 lock as an argument, and a there are corresponding Wait() calls that
573 don't specify a lock now. It turns that the resulting implmentation
574 can't be made more efficient, as the internal lock needs to be used by
575 Signal() and Broadcast(), to access internal data structures. As a
576 result, I was not able to utilize the user supplied lock (which is
577 being used by the user elsewhere presumably) to protect the private
580 9) Since you have a second lock, how can be be sure that there is no
581 possible deadlock scenario? Our internal_lock_ is always the last
582 lock acquired, and the first one released, and hence a deadlock (due
583 to critical section problems) is impossible as a consequence of our
586 10) When doing a Broadcast(), why did you copy all the events into
587 an STL queue, rather than making a linked-loop, and iterating over it?
588 The iterating during Broadcast() is done so outside the protection
589 of the internal lock. As a result, other threads, such as the thread
590 wherein a related event is waiting, could asynchronously manipulate
591 the links around a cv_event. As a result, the link structure cannot
592 be used outside a lock. Broadcast() could iterate over waiting
593 events by cycling in-and-out of the protection of the internal_lock,
594 but that appears more expensive than copying the list into an STL
597 11) Why did the lock.h file need to be modified so much for this
598 change? Central to a Condition Variable is the atomic release of a
599 lock during a Wait(). This places Wait() functionality exactly
600 mid-way between the two classes, Lock and Condition Variable. Given
601 that there can be nested Acquire()'s of locks, and Wait() had to
602 Release() completely a held lock, it was necessary to augment the Lock
603 class with a recursion counter. Even more subtle is the fact that the
604 recursion counter (in a Lock) must be protected, as many threads can
605 access it asynchronously. As a positive fallout of this, there are
606 now some DCHECKS to be sure no one Release()s a Lock more than they
607 Acquire()ed it, and there is ifdef'ed functionality that can detect
608 nested locks (legal under windows, but not under Posix).
610 12) Why is it that the cv_events removed from list in Broadcast() and Signal()
611 are not leaked? How are they recovered?? The cv_events that appear to leak are
612 taken from the waiting_list_. For each element in that list, there is currently
613 a thread in or around the WaitForSingleObject() call of Wait(), and those
614 threads have references to these otherwise leaked events. They are passed as
615 arguments to be recycled just aftre returning from WaitForSingleObject().
617 13) Why did you use a custom container class (the linked list), when STL has
618 perfectly good containers, such as an STL list? The STL list, as with any
619 container, does not guarantee the utility of an iterator across manipulation
620 (such as insertions and deletions) of the underlying container. The custom
621 double-linked-list container provided that assurance. I don't believe any
622 combination of STL containers provided the services that were needed at the same
623 O(1) efficiency as the custom linked list. The unusual requirement
624 for the container class is that a reference to an item within a container (an
625 iterator) needed to be maintained across an arbitrary manipulation of the
626 container. This requirement exposes itself in the Wait() method, where a
627 waiting_event must be selected prior to the WaitForSingleObject(), and then it
628 must be used as part of recycling to remove the related instance from the
629 waiting_list. A hash table (STL map) could be used, but I was embarrased to
630 use a complex and relatively low efficiency container when a doubly linked list
631 provided O(1) performance in all required operations. Since other operations
632 to provide performance-and/or-fairness required queue (FIFO) and list (LIFO)
633 containers, I would also have needed to use an STL list/queue as well as an STL
634 map. In the end I decided it would be "fun" to just do it right, and I
635 put so many assertions (DCHECKs) into the container class that it is trivial to
636 code review and validate its correctness.
640 ConditionVariable::ConditionVariable(Lock
* user_lock
)
642 static bool use_vista_native_cv
= BindVistaCondVarFunctions();
643 if (use_vista_native_cv
)
644 impl_
= new WinVistaCondVar(user_lock
);
646 impl_
= new WinXPCondVar(user_lock
);
649 ConditionVariable::~ConditionVariable() {
653 void ConditionVariable::Wait() {
657 void ConditionVariable::TimedWait(const TimeDelta
& max_time
) {
658 impl_
->TimedWait(max_time
);
661 void ConditionVariable::Broadcast() {
665 void ConditionVariable::Signal() {