tr8dr
commented
2 years ago I am no longer maintaining this (aside from a JVM version). Is this from R or python? I think there are scenarios in R where it can release an object early, where it should not. If you don't mind objects piling up, you could change the code to cause the Release() function in CLRObjectProxy to do nothing.
I don't see any flaw in the C# logic. The object IDs will be unique (and increasing). The only way the object reference disappears is from the release call from R. Have seen this behavior in R myself (back when I was using R - have since switched to python).
Could alternatively create a list or queue of objects pending release, and delete the oldest ones (to avoid the R problem). The logic would be as follows: In the Release function do:
create a member variable called _pending_delete or something like that
_pending_delete.Add (obj);
if (_pending_delete.Length > somethreshold)
<delete 1/2 of that threshold starting from oldest>
<trim list or queue to remove the items no longer pending>Somewhat more complex would be to:
- queue objects for deletion when R "releases" the reference
- if the object is touched again, the object is removed from the deletion queue
- anything left in the queue is periodically deleted up to some threshold, so as to keep newer object in the pending state
daveyboy103
Sometimes the internal cache dictionaries for ObjectId and objects gets corrupted and the program tries to access a key that does not exist. Currently only fixable by restarting the server.
CLRObjectProxy.cs