Incorrect implementation of StaticClass for Blueprint classes about dumper-7 HOT 5 CLOSED

Yea, I've seen that issue before and you're right, BlueprintGeneratedClasses need some fixes for StaticClass.

I don't think finding the class every time without any caching is feasible, as most people use StaticClass per-tick to check object-types.

A solutions I am considering is to store the index rather than the pointer and verify the validity of the index of the class every time StaticClass is called.

Pseudo-cpp:

class UClass* ABP_CharacterLightController_C::StaticClass()
{
	static int32 ClassIdx = 0x0;

	/* Could be external function, not really unique to this StaticClass functon */
	static auto SetClassIndex = [](UClass* Object, int32& Index) -> UClass*
	{
		if (Object)
			Index = Object->Index;

		return Object;
	};

	if (ClassIdx == 0x0)
		return SetClassIndex(UObject::FindClassFast("BP_CharacterLightController_C"), ClassIdx);

	UClass* ClassObj = static_cast<UClass*>(UObject::GObjects->GetByIndex(ClassIdx));

	/* Could use cast flags too to save some string comparisons */
	if (!ClassObj || ClassObj->GetName() != "BP_CharacterLightController_C")
		return SetClassIndex(UObject::FindClassFast("BP_CharacterLightController_C"), ClassIdx);

	return ClassObj;
}

from dumper-7.

A solutions I am considering is to store the index rather than the pointer and verify the validity of the index of the class every time StaticClass is called.

class UClass* ABP_CharacterLightController_C::StaticClass()
{
	static int32 ClassIdx = 0x0;

	/* Could be external function, not really unique to this StaticClass functon */
	static auto SetClassIndex = [](UClass* Object, int32& Index) -> UClass*
	{
		if (Object)
			Index = Object->Index;

		return Object;
	};

The problem with this is the class object can get destroyed at any point in time, and then once you try to access it, you might be accessing not accessible memory or just garbage memory.

When I was trying to figure out the bugs in the code, once the blueprint class actually gets loaded, the old class object would point to a different object, it wasn't even a class object. And I was just lucky that another object was put at the same memory location as the old class object, so I could get its' name without crashing.
Other times might not be that lucky and once you try to access Object->Index, you might be accessing either inaccessible memory or just garbage reallocated memory, so you can't just check if memory is accessible with IsBadReadPtr(Object): it will be accessible but something else is there now.

I don't think finding the class every time without any caching is feasible, as most people use StaticClass per-tick to check object-types.

Well, that will be on programmers to cache and reset the pointers. I get the class only once at match start and then invalidate on match end, because that's when levels get loaded and unloaded, so it's always correct.

To actually be able to cache these class pointers, you'd need to hook into the object destruction function, and then have like a vector of UClass**(to the fields in each class, so you'd make a field static UClass* StaticClass in each class), and then invalidate those pointers once UClass object gets destroyed ABP_CharacterLightController_C::StaticClass = nullptr, but using the vector of those pointers to the UClass pointers, because you'd be doing it from inside of the object destruction hook. That would fix it, there would be no way to get invalid UClass pointer this way at any point in time, as I understand it.

from dumper-7.

The problem with this is the class object can get destroyed at any point in time,

I am aware of this. In the proposed code above I'm explicitly not storing a UClass*, but rather the index of the class from the previous (re-)initialization. Re-fetching the Class from the index will return nullptr if the index was invalidated. If the index is still valid, but the object at said index was changed the check ClassObj->GetName() != "BP_CharacterLightController_C") fails and the object will be re-initialized.

from dumper-7.

A fix for the issue has been implemented, but it is mostly untested. It would be nice if you could test the latest commit on main and tell me if StaticClass is still crashing with BP geneated classes.

I'll leave this issue ope until you can confirm it's been fixed.

from dumper-7.

As you seemingly don't care about this issues anymore and it has been fixed, I'm going to close this.

from dumper-7.