Make TracebackException work with Python 3.10.

[nascheme]

Pass '_seen' argument as None if empty (on first call). Pass 'compact'
kwarg as needed.

[nascheme]

This might fix #1899.

[codecov]

Codecov Report

Merging #2030 (5b33786) into master (2195763) will decrease coverage by 6.43%.
The diff coverage is 60.00%.

@@            Coverage Diff             @@
##           master    #2030      +/-   ##
==========================================
- Coverage   99.56%   93.12%   -6.44%     
==========================================
  Files         114      114              
  Lines       14618    14584      -34     
  Branches     1117     1117              
==========================================
- Hits        14554    13582     -972     
- Misses         43      937     +894     
- Partials       21       65      +44     

Impacted Files	Coverage Δ
trio/_core/_multierror.py	88.88% <60.00%> (-9.99%)	⬇️
trio/_windows_pipes.py	0.00% <0.00%> (-100.00%)	⬇️
trio/_wait_for_object.py	0.00% <0.00%> (-100.00%)	⬇️
trio/_core/_windows_cffi.py	0.00% <0.00%> (-100.00%)	⬇️
trio/_subprocess_platform/windows.py	0.00% <0.00%> (-100.00%)	⬇️
...tests/test_multierror_scripts/apport_excepthook.py	0.00% <0.00%> (-100.00%)	⬇️
trio/_core/_io_windows.py	0.00% <0.00%> (-98.77%)	⬇️
trio/tests/test_wait_for_object.py	10.37% <0.00%> (-89.63%)	⬇️
trio/_core/tests/test_windows.py	17.82% <0.00%> (-82.18%)	⬇️
trio/tests/test_windows_pipes.py	25.33% <0.00%> (-74.67%)	⬇️
... and 37 more

[altendky]

Is this covered in #1921?

[nascheme]

Is this covered in #1921?

I don't think so. It fixes the issue with the keyword argument. However there is a more subtle problem with Python 3.10. I believe the key commit is 6dfd1734f5b230bb8fbd2a9df806c1333b6652a8 in the cpython repo:
bpo-42848: remove recursion from TracebackException (GH-24158)

The meaning of _seen = None is now different than from _seen = set(). If you pass _seen=set(), the __init__ method will not correctly initialize the TracebackException object.

[altendky]

I believe we have just implemented different means of passing None or set(), but that they are the same result?

https://github.com/altendky/trio/blob/c079d3cfdf4b91c76cd9aa376d1ddb76db4d40d7/trio/_core/_multierror.py#L405-L423

I tried a GitHub compare but it isn't showing what I expect.
altendky/trio@c079d3c...5b33786 So, here is a screenshot of my local diff. This PR on the left and #1921 on the right.

Aside from the _seen bit, it seems like the kwargs in this PR are unused.

[graingert]

fixed in #1921 ?

[graingert]

ah no this is different

[Zac-HD]

I think this has been fixed by the recent change to exceptiongroup, @agronholm can reopen if I'm wrong.