Use LongLivedObjects for TurboModule callbacks

[vmoroz]

Description

Type of Change

• Bug fix (non-breaking change which fixes an issue)

Why

@acoates-ms has reported an issue with RNW Turbo Module implementation where we see a crash after a RN instance is unloaded. The crash is caused by a JSI Function that captured in a lambda used for asynchronous callbacks or in a Promise.
We destroy JS engine instance when unloading the RN instance, and the JSI Function cannot be safely released anymore.

What

RN Turbo Modules that wrap up CxxModules use LongLivedObject and LongLivedObjectCollection as a mechanism to address similar issues. The callbacks are wrapped up into CallbackWrapper class inherited from the LongLivedObject. The LongLivedObjectCollection is cleared when JS engine instance is shut down and it releases all associated std::shared_ptr<LongLivedObject> instances, which in turn release all associated JSI values. The deferred callback and Promise lambdas capture std::weak_ptr<LongLivedObject>, and do nothing if the weak pointer cannot be "locked" anymore.

The PR uses the same LongLivedObjectCollection for RNW Turbo Modules, but it introduces new LongLivedJsiRuntime and LongLivedJsiValue classes inherited from LongLivedObject. They have smaller size and do not use global LongLivedObjectCollection instance.

The solution consists of the following parts:

• JSDispatcherWriter keeps std::weak_ptr<LongLivedJsiRuntime> instead of JSI Runtime and does nothing while handling results if the JSI Runtime is not available anymore.
• TurboModulesProvider is wrapping up JSI Function instances into std::weak_ptr<LongLivedJsiFunction>. (The LongLivedJsiFunction is an alias for LongLivedJsiValue<jsi::Function>.) It helps to release all JSI Function instances while the JS engine instance is shutting down.
• ReactNativeHost creates a new instance of LongLivedObjectCollection and passes it to the TurboModulesProvider.
• The InstanceImpl in OInstance.h passes the LongLivedObjectCollection instance from TurboModulesProvider to the OJSIExecutorFactory. The OJSIExecutorFactory associates the LongLivedObjectCollection instance with a JS engine instance by passing it to the TurboModuleBinding::install call.

Testing

We added new unit tests in TurboModuleTests.cpp to check behavior when callbacks and Promises are released after RN instance unload. They are all passing. These tests cause a crash without changes to the TurboModulesProvider.

Also, we added a new test utility class TestNotificationService that helps to coordinate setting and waiting for synchronization events. Unlike the TestEventService that expects synchronization events to appear in a specified order, the new TestNotificationService keeps a set of events and lets the Wait function to continue if the set contains an event that it waits for.

Microsoft Reviewers: Open in CodeFlow

[ghost]

Hello @vmoroz!

Because this pull request has the AutoMerge label, I will be glad to assist with helping to merge this pull request once all check-in policies pass.

p.s. you can customize the way I help with merging this pull request, such as holding this pull request until a specific person approves. Simply @mention me (@msftbot) and give me an instruction to get started! Learn more here.

[jonthysell]

@vmoroz Is this important enough to get backported into 0.70?

[vmoroz]

@vmoroz Is this important enough to get backported into 0.70?

Yes, it is important. Without this fix we have a risk of crashing every time when a RN instance is unloaded.
This is when we use JSI-based TurboModules.