Ok, I've found a fix for the second crash as well.

My first fix is not perfect as assigning a new instance of the ConditionalWeakTable _weakExtends = new(); still keeps the referenced ExtendNotifier instances alive. They will eventually reach GC finalization and ask NoesisGUI to release itself (which 100% happens during the second reloading which always results in a crash; BTW, I've tried and it's NOT helpful to call GC.Collect() + GC.WaitForPendingFinalizers() to force collecting these references right after _weakExtends = new(); — some entries still remained in heap memory).

Apparently I just need to ensure that no ExtendNotifier instances will ever invoke its finalizer if its parent _weakExtends is replaced with a new instance.
Obviously it would be great if I can simply foreach references stored in _weakExtends and release each of them manually before creating a new instance of _weakExtends table. However ConditionalWeakTable doesn't support enumeration on .NET Framework (this is new API added in .NET Standard 2.1).

The second idea was to store a WeakReference on the current _weakExtends... but apparently it's NOT supported — the WeakReference always replies NOT alive no matter whether _weakExtends is alive or not. (Ha-ha, so there is simply too much weakness for .NET to handle!)

The third idea is simply to store the current version number of the _weakExtends instance:

private static byte _latestWeakStorageId;

if (currentWeakStorageId != _latestWeakStorageId)

{

// storage for this weak reference is destroyed, no need to release

return;

}

And it works! No more crashes.
I think the only downside is that there are still some pointers (ExtendNotifier) stored in NoesisGUI native library which will be never released. (again, we cannot simply enumerate _weakExtends to release them all) But probably it's harmless and will not result in any major leak.

Regards!

Thanks for the detailed info, I'm going to analyze it and see what would be the best option to fix this.