`ExceptionGroup` repr is misleading if original exception sequence is mutated

[gschaffner]

Bug report

Bug description:

The exception listing in repr(exc_grp) diverges from exc_grp.exceptions when the initial sequence of exceptions passed to the constructor has been mutated.

Expected behavior: The exceptions listed in repr(exc_grp) match exc_grp.exceptions, which is immutable.

Reproducer:

exceptions = list[Exception]()
exceptions.append(ValueError())
exc_grp = ExceptionGroup("", exceptions)
print(f"{exc_grp.exceptions=!r}")
print(f"{exc_grp=!r}")
print()
exceptions.clear()
print(f"{exc_grp.exceptions=!r}")
print(f"{exc_grp=!r}")

Output:

exc_grp.exceptions=(ValueError(),)
exc_grp=ExceptionGroup('', [ValueError()])

exc_grp.exceptions=(ValueError(),)
exc_grp=ExceptionGroup('', [])

Expected output:

exc_grp.exceptions=(ValueError(),)
exc_grp=ExceptionGroup('', [ValueError()])

exc_grp.exceptions=(ValueError(),)
exc_grp=ExceptionGroup('', [ValueError()])

CPython versions tested on:

3.11, 3.12, 3.13, 3.14

Operating systems tested on:

Linux

Linked PRs

• gh-141732: Fix ExceptionGroup repr changing when original exception sequence is mutated #141736
• [3.14] gh-141732: Fix ExceptionGroup repr changing when original exception sequence is mutated (GH-141736) #142388
• [3.13] gh-141732: Fix ExceptionGroup repr changing when original exception sequence is mutated (GH-141736) #142391
• [3.14] gh-141732: Fix ExceptionGroup repr changing when original exception sequence is mutated (GH-141736) #144445

[gschaffner]

I noticed that PyPy behaves the way I expected here; they use self.exceptions in BaseExceptionGroup.__repr__ instead of self.args[1]:

https://github.com/pypy/pypy/blob/release-pypy3.11-v7.3.20/pypy/module/exceptions/app_group.py#L92-L93

On CPython, it looks like BaseExceptionGroup.__repr__ resolves to BaseException.__repr__, so the repr ends up using self.args[1] instead of self.exceptions.

I skimmed the PEP and didn't see anything that indicated that either repr implementation is in violation of the PEP.

[dr-carlos]

I skimmed the PEP and didn't see anything that indicated that either repr implementation is in violation of the PEP.

The PEP does reference the repr() of ExceptionGroups but not explicitly specify anything:

>>> try:
...     raise ExceptionGroup("problem", [BlockingIOError()])
... except* OSError as e:   # Would catch the error
...     print(repr(e))
... except* BlockingIOError: # Would never run
...     print('never')
...
ExceptionGroup('problem', [BlockingIOError()])

But I agree that mutating an argument which is only used to create an immutable tuple shouldn't change the repr.

[serhiy-storchaka]

This may be a part of more complex problem. See #137889.

[hugovk]

This has been merged in main and 3.13 but cannot be merged for 3.14 because of failing CI: #142388.

Please can you check the backport?

[dr-carlos]

This has been merged in main and 3.13 but cannot be merged for 3.14 because of failing CI: #142388.

Please can you check the backport?

Yes, will do!
Opened #144445