WEAK_SCOPED_MUTEX_LOCK/WEAK_MUTEX_TRY_LOCK revisited

[masaori335]

WEAK_SCOPED_MUTEX_LOCK and WEAK_MUTEX_TRY_LOCK are introduced to allow the mutex to be a nullptr.

trafficserver/include/iocore/eventsystem/Lock.h

Lines 52 to 57 in 31c1592

	// A weak version of the SCOPED_MUTEX_LOCK macro, allows the mutex to be a nullptr.
	#ifdef DEBUG
	#define WEAK_SCOPED_MUTEX_LOCK(_l, _m, _t) WeakMutexLock _l(MakeSourceLocation(), (char *)nullptr, _m, _t);
	#else // DEBUG
	#define WEAK_SCOPED_MUTEX_LOCK(_l, _m, _t) WeakMutexLock _l(_m, _t);
	#endif // DEBUG

Typical usage is below.

trafficserver/src/api/APIHook.cc

Lines 73 to 75 in 31c1592

	WEAK_SCOPED_MUTEX_LOCK(lock, m_cont->mutex, this_ethread());
	return m_cont->handleEvent(event, edata);

However, if we take a look at the Continuation::handleEvent in this case, it has a release assert of the mutex.

trafficserver/include/iocore/eventsystem/Continuation.h

Lines 223 to 229 in 31c1592

	TS_INLINE int
	handleEvent(int event = CONTINUATION_EVENT_NONE, void *data = nullptr)
	{
	// If there is a lock, we must be holding it on entry
	ink_release_assert(!mutex || mutex->thread_holding == this_ethread());
	return (this->*handler)(event, data);
	}

It looks like we don't have reason to use WEAK_ version to allow nullptr in many cases.

[duke8253]

Hmm, I need to think this through a bit. I'm referencing #5879 to keep track of related PRs.