Reordered operations in Map.add_vertex

[seventil]

Fix is made to avoid hitting a race condition due to weakref.finalize() being called. It seems that .finalize temporarily deletes references of the object passed to it from dicts, therefore trying to get it from __type_dict or _store might randomly raise a KeyError.

This PR fixes failing tests from easyscience/dynamics-lib#96

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 81.16%. Comparing base (bd10653) to head (787222d).
⚠️ Report is 5 commits behind head on more_weakref_fixes.

Additional details and impacted files

@@                 Coverage Diff                 @@
##           more_weakref_fixes     #197   +/-   ##
===================================================
  Coverage               81.15%   81.16%           
===================================================
  Files                      52       52           
  Lines                    4267     4268    +1     
  Branches                  740      740           
===================================================
+ Hits                     3463     3464    +1     
  Misses                    624      624           
  Partials                  180      180           

Flag	Coverage Δ
unittests	81.16% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/easyscience/global_object/map.py	96.93% <100.00%> (+0.01%)	⬆️

[rozyczko]

How is this fixing the issue? Can you explain what those lines are changing?

[seventil]

How is this fixing the issue? Can you explain what those lines are changing?

Just for reference, the traceback that lead to the creation of this PR

        self.__type_dict[name] = _EntryList()  # Add objects type to the list of types
>       self.__type_dict[name].finalizer = weakref.finalize(self._store[name], self.prune, name)
        ^^^^^^^^^^^^^^^^^^^^^^
E       KeyError: 'Parameter_338'

Weird thing is that here we see sequential __setitem__ call and __getitem__ call on the same dict resulting in KeyError, which just doesn't make any sense.

Here is my assumption of what is happening:
when wekref.finalize() is being called with some object passed as an argument, the dicts that hold a reference to such object are being renewed -> key-val pairs dissapear and then return back to the dicts. In some cases, trying to get an item from a dict while the dicts are being 'renewed' results in a KeyError. This is just my observation from try runs, I don't know how finalize actually works underneath the python API.

Basically, the line self.__type_dict[name].finalizer = weakref.finalize(self._store[name], self.prune, name) makes unnecesary calls to self.__type_dict.__getitem__ and to self._store.__getitem__.
If removing unnecessary calls also happens to solve the issue with "disappearing" keys, I think it's a worthwile change.
It doesn't change the logic of the method, but rather just semantics, so why not try?

I tested this change in EasyDynamics, which fixed the failing tests from easyscience/dynamics-lib#96

I didn't investigate why tests fail in this branch, but I expect that more_weakref_fixes hold the fixes for the issue I have here.

[seventil]

@rozyczko could you please merge this into your branch if it's ok?