[ExceptionGroup] Inconsistent raise inside except*

Bug report

When raising inside except* block and the caught exception wasn't an ExceptionGroup originally, then raised exception doesn't get wrapped in ExceptionGroup:

try:
    try:
        raise TypeError(1)  # prints ValueError(3)
        raise ExceptionGroup('', [TypeError(2)])  # prints ExceptionGroup('', [ValueError(3)])
    except* TypeError:
        raise ValueError(3)
except BaseException as e:
    print(repr(e))

Your environment

CPython versions tested on: 3.11.3
Operating system and architecture: Windows 11 / Ubuntu 22.04

Linked PRs

@iritkatriel

Cc: @iritkatriel

This is fully documented behavior. https://peps.python.org/pep-0654/#raising-new-exceptions

Raising a new instance of a naked exception does not cause this exception to be wrapped by an exception group. Rather, the exception is raised as is, and if it needs to be combined with other propagated exceptions, it becomes a direct child of the new exception group created for that.

>>> try:
...     raise ExceptionGroup("eg", [ValueError('a')])
... except* ValueError:
...     raise KeyError('x')
...
  | ExceptionGroup:  (1 sub-exception)
  +-+---------------- 1 ----------------
    | Exception Group Traceback (most recent call last):
    |   File "<stdin>", line 2, in <module>
    | ExceptionGroup: eg (1 sub-exception)
    +-+---------------- 1 ----------------
      | ValueError: a
      +------------------------------------
    |
    | During handling of the above exception, another exception occurred:
    |
    | Traceback (most recent call last):
    |   File "<stdin>", line 4, in <module>
    | KeyError: 'x'
    +------------------------------------
>>>
>>> try:
...     raise ExceptionGroup("eg", [ValueError('a'), TypeError('b')])
... except* ValueError:
...     raise KeyError('x')
...
  | ExceptionGroup:  (2 sub-exceptions)
  +-+---------------- 1 ----------------
    | Exception Group Traceback (most recent call last):
    |   File "<stdin>", line 2, in <module>
    | ExceptionGroup: eg (1 sub-exception)
    +-+---------------- 1 ----------------
      | ValueError: a
      +------------------------------------
    |
    | During handling of the above exception, another exception occurred:
    |
    | Traceback (most recent call last):
    |   File "<stdin>", line 4, in <module>
    | KeyError: 'x'
    +---------------- 2 ----------------
    | Exception Group Traceback (most recent call last):
    |   File "<stdin>", line 2, in <module>
    | ExceptionGroup: eg (1 sub-exception)
    +-+---------------- 1 ----------------
      | TypeError: b
      +------------------------------------
>>>

@sunmy2019

@sunmy2019 thanks for the docs pointer. It is a bit confusing behavior for me, since I can't always be sure what I'm actually raising. But I guess we can close this then?

@iritkatriel

I don't think that the issue in the OP here clearly matches any example shown in the PEP. The key factor in the PEP example:

>>> try:
...     raise ExceptionGroup("eg", [ValueError('a'), TypeError('b')])
... except* ValueError:
...     raise KeyError('x')

is that the raised exception group includes both a ValueError (will be caught by the except* ValueError) and a TypeError (is not caught and thus must be propagated.) It is the latter TypeError that means the newly raised KeyError "needs to be combined with other propagated exceptions" (the TypeError), and thus a new ExceptionGroup is created for that.

In the OP example:

try:
    raise ExceptionGroup('', [TypeError(2)])
except* TypeError:
    raise ValueError(3)

the original exception group contains only a TypeError, which is caught by except* TypeError. So there are no remaining uncaught exceptions that must propagate along with the newly raised ValueError. So according to the wording in the PEP:

Rather, the exception is raised as is, and if it needs to be combined with other propagated exceptions, it becomes a direct child of the new exception group created for that

it is surprising (to me -- not an expert in the ExceptionGroup feature by any means) that a new ExceptionGroup would be created containing just the newly-raised ValueError.

The documentation says about this case:

Any remaining exceptions that were not handled by any except* clause are re-raised at the end, combined into an exception group along with all exceptions that were raised from within except* clauses.

This wording is more ambiguous than the wording in the PEP, and leaves open the possibility that a new exception group should always be created, even if we are raising just a single exception and there were no "remaining exceptions" to propagate.

So it is possible this is intended behavior, but it doesn't seem at all clear. I think we need input from @iritkatriel here.

pythongh-103590: do not wrap a single exception raised from a try-exc…

@gvanrossum

I agree that the wording of the PEP implies that wrapping in an exception group should not happen unless it is necessary, and it is not necessary when there is only one exception.

I don't remember why I implemented it this way (with unit tests asserting this behaviour, no less). I'm going to assume that it was a simple mistake and make a PR to fix it. Will ask @gvanrossum for a second option.