GDScript: Fix interrupted coroutines not clearing

[dalexeev]

• Fixes Timer (SceneTreeTimer) leaked when awaiting it while quitting. #79669.
• Fixes Memory leak when calling queue_free() on a node before a SceneTreeTimer times out #84860.
• Fixes When an interrupted coroutine is deleted, local objects are not cleared #116706.
• Supersedes GDScript: Fix SceneTreeTimer leaked when awaiting it while quitting #96441.
• See also GDScript: Fix stack manipulation for await #61003.
• Need to make sure this doesn't reintroduce Memory leak/Stack corruption when using GDScript await in release_debug builds (GCC specific) #57126.

[HolonProduction]

Just to check if I'm understanding this correctly. The purpose of those code snippets:

godot/modules/gdscript/gdscript.cpp

Lines 2033 to 2035 in 2327a82

	if (ObjectDB::get_instance(state_id)) {
	state->_clear_stack();
	}

godot/modules/gdscript/gdscript.cpp

Lines 1498 to 1500 in 2327a82

	if (ObjectDB::get_instance(state_id)) {
	state->_clear_stack();
	}

is to break reference cycles when a function state is referenced from its stack?

[dalexeev]

The purpose of those code snippets <...> is to break reference cycles when a function state is referenced from its stack?

See #65910. I'm not sure, but probably the logic behind this is that during _clear_connections() the above obtained GDScriptFunctionState might be freed, so using the raw pointer state again might be unsafe.

godot/modules/gdscript/gdscript.cpp

Lines 2030 to 2035 in 2327a82

	GDScriptFunctionState *state = E->self();
	ObjectID state_id = state->get_instance_id();
	state->_clear_connections();
	if (ObjectDB::get_instance(state_id)) {
	state->_clear_stack();
	}

• Need to make sure this doesn't reintroduce Memory leak/Stack corruption when using GDScript await in release_debug builds (GCC specific) #57126.

I tested this and looked at the code a bit more. It looks like it works correctly. Perhaps the crashes were caused by the _clear_stack() call not being protected by a mutex, and the memory leaks were caused by #65910 not being implemented yet. Or maybe it was fixed in a later PR.

The current behavior is clearly wrong, and re-adding the _clear_stack() call to the GDScriptFunctionState destructor seems correct to me. If the instance whose signal GDScriptFunctionState is connected to has been freed, the coroutine cannot be resumed under any circumstances, so the saved stack must be cleared regardless of whether the function has completed.

[dalexeev]

See also:

• GDScript: Fix SceneTreeTimer leaked when awaiting it while quitting #96441
• Timer (SceneTreeTimer) leaked when awaiting it while quitting. #79669
• Memory leak when calling queue_free() on a node before a SceneTreeTimer times out #84860

[HolonProduction]

Makes sense to me

[Repiteo]

Thanks!