Bug in MultiError.filter: pushing down tracebacks unnecessarily

[njsmith]

[Not sure if this is worth fixing given that we're hopefully going to rip out MultiError.filter soon, see #611. But figured I should write it down. If nothing else this is more evidence for how tricky MultiError.filter is...]

I just noticed this playing with python notes-to-self/trivial-err.py, after #640 was merged. We get:

[... some output elided ...]
Details of embedded exception 2:

  trio.MultiError: NameError('Bob',), KeyError()

  Details of embedded exception 1:

    Traceback (most recent call last):
      File "notes-to-self/trivial-err.py", line 12, in child2
        nursery.start_soon(grandchild2)
      File "/home/njs/trio/trio/_core/_run.py", line 399, in __aexit__
        raise scope_exc
      File "notes-to-self/trivial-err.py", line 18, in grandchild2
        raise NameError("Bob")
    NameError: Bob

  Details of embedded exception 2:

    Traceback (most recent call last):
      File "notes-to-self/trivial-err.py", line 12, in child2
        nursery.start_soon(grandchild2)
      File "/home/njs/trio/trio/_core/_run.py", line 399, in __aexit__
        raise scope_exc
      File "notes-to-self/trivial-err.py", line 15, in grandchild1
        raise KeyError
    KeyError

Notice how the MultiError at the top has no traceback frames, and the NameError and KeyError inside it have identical child2 and __aexit__ frames.

What should happen here is that the child2 and __aexit__ frames are on the MultiError once, instead of being pushed down onto the individual exceptions.

I added some code to CancelScope._close to see what was going on:

            import traceback
            print("pre-filter")
            if exc is not None:
                traceback.print_exception(type(exc), exc, exc.__traceback__)
            filtered_exc = MultiError.filter(self._exc_filter, exc)
            print("post-filter")
            if filtered_exc is not None:
                traceback.print_exception(type(filtered_exc), filtered_exc, filtered_exc.__traceback__)
            return filtered_exc

and it looks like at some point we have an exception:

trio.MultiError: ValueError(), Cancelled(), <MultiError: Cancelled(), Cancelled(), NameError('Bob',), KeyError()>

and that embedded MultiError with the NameError and KeyError inside it has those two frames on it:

Details of embedded exception 3:

  Traceback (most recent call last):
    File "notes-to-self/trivial-err.py", line 12, in child2
      nursery.start_soon(grandchild2)
    File "/home/njs/trio/trio/_core/_run.py", line 399, in __aexit__
      raise scope_exc
  trio.MultiError: Cancelled(), Cancelled(), NameError('Bob',), KeyError()

[...]

  Details of embedded exception 3:

    Traceback (most recent call last):
      File "notes-to-self/trivial-err.py", line 18, in grandchild2
        raise NameError("Bob")
    NameError: Bob

  Details of embedded exception 4:

    Traceback (most recent call last):
      File "notes-to-self/trivial-err.py", line 15, in grandchild1
        raise KeyError
    KeyError

And then MultiError.filter removes the Cancelled exceptions, and somehow in the process it ends up pushing down the tracebacks onto the individual frames:

post-filter
trio.MultiError: ValueError(), <MultiError: NameError('Bob',), KeyError()>

[...]

Details of embedded exception 2:

  trio.MultiError: NameError('Bob',), KeyError()

  Details of embedded exception 1:

    Traceback (most recent call last):
      File "notes-to-self/trivial-err.py", line 12, in child2
        nursery.start_soon(grandchild2)
      File "/home/njs/trio/trio/_core/_run.py", line 399, in __aexit__
        raise scope_exc
      File "notes-to-self/trivial-err.py", line 18, in grandchild2
        raise NameError("Bob")
    NameError: Bob

  Details of embedded exception 2:

    Traceback (most recent call last):
      File "notes-to-self/trivial-err.py", line 12, in child2
        nursery.start_soon(grandchild2)
      File "/home/njs/trio/trio/_core/_run.py", line 399, in __aexit__
        raise scope_exc
      File "notes-to-self/trivial-err.py", line 15, in grandchild1
        raise KeyError
    KeyError

That's not supposed to happen... we're replacing one MultiError with another, so the traceback frames are supposed to be transferred over.

[Zac-HD]

Closing this as ExceptionGroup now exists and is an upstream problem.