PySet::empty could be infallible

[anuraaga]

Currently PySet::empty() returns PyResult<Bound<'_, PySet>>, though AFAIK, this operation can't fail. Or at least it can't fail any more likely than PyDict::new(py) which does not return PyResult. I was initially confused since I was populating an empty set and dict to the same struct.

On the flip side, I understand that PySet::empty is being consistent with PySet::new, which needs to return error for unhashable items. So there is a consistency argument to be made, but in general I guess infallible methods should not return PyResult?

[davidhewitt]

Actually PySet::empty() could fail with a MemoryError.

PyO3 has been inconsistent about whether to panic or return a result when collection construction fails with a MemoryError.

It's potentially nice to have empty() which panics and some API like try_empty() which fails with MemoryError. But at the same time, given almost everything in Python is fallible, is it better to avoid the panics?

[anuraaga]

For me, I prefer consistency on the empty versions of PySet and PyDict. I guess the latter must be panicing for MemoryError then - if it instead returned an error, then this wouldn't have been surprising. I would personally though feel panicing on MemoryError for an empty collection to be reasonable since that's getting to the point where the runtime will likely crash soon anyways.