Great, thanks for your continuous efforts to backport fixes, it's highly appreciated. I'll look into merging this before the 2.4 release :)

Codecov Report

Merging #852 into master will decrease coverage by 11.57%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #852       +/-   ##
==========================================
- Coverage   76.88%   65.3%   -11.58%     
==========================================
  Files          64       1       -63     
  Lines        5935     294     -5641     
  Branches      976       0      -976     
==========================================
- Hits         4563     192     -4371     
+ Misses       1041     102      -939     
+ Partials      331       0      -331

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6034640...cf17b3c. Read the comment docs.

The Linux build seems to be fine apart from an unrelated error (happens on master as well). On Windows, however, for some reason this seems to stall, the builds time out during the unit tests.

It trips up the DomainReloadAndGC test, it looks like there is a deadlock in DisposeAll.

If I remove the GIL call from DisposeAll everything works fine as far as I can tell right now.

Great, this looks a lot better, thank you very much :)

@Martin-Molinero Using Py_AddPendingCall is aiming for no Python thread, although we called PyEval_InitThreads at Initialize, but that is a optimize point that we can prevent create a GIL on single thread environment. Also using Py.GIL may create a temporary Python thread state which takes overhead.
I think we can make a better user experience for the finalizer, we can setup a timer for finalizing queue, if the time past threshold value then create a Python thread to dispose them. That may also help for the queue size don't reach count threshold for a long time but keep the object reference.