@Flamefire This looks ok to me since the clear() and update() are only being done in the tests, so we shouldn't run into these runtime issues any more. I don't trust that I fully understand the issue though, can you review this PR?

I have found https://stackoverflow.com/questions/43181440/what-does-del-sys-modulesmodule-actually-do which references the python docs:

[sys.modules] is a dictionary that maps module names to modules which have already been loaded. This can be manipulated to force reloading of modules and other tricks. However, replacing the dictionary will not necessarily work as expected and deleting essential items from the dictionary may cause Python to fail.

So IMO even though it is test code this shouldn't be done as it seems to be unreliable (as shown by the issue: adding an extra import seems to have triggered the GC causing the issue to become a bug).
Why was the test code changed? Wasn't the old stuff (del and reload) enough?

I assume the del to be fine, so the module code change is good as far as I can tell, thanks!

I added the sys.modules.clear() in the test at some point because it seemed necessary, but it actually isn't, the del sys.modules[*] statements indeed work just as well; fixed in 482860a .