Also relax the PySequence check when extracting fixed-sized arrays. #2675
Conversation
adamreichold
force-pushed
the
relax-extract-array
branch
3 times, most recently
from
7863dc6
to
4c48977
Compare
| if ffi::PySequence_Check(obj.as_ptr()) != 0 { | ||
| <PySequence as PyTryFrom>::try_from_unchecked(obj) | ||
| } else { | ||
| return Err(PyDowncastError::new(obj, "Sequence").into()); |
There was a problem hiding this comment.
Willing to add a test for this error branch (ie. pass a non-iterable) to get the coverage to 100%? 😃
There was a problem hiding this comment.
Added a test case triggering this error path.
newsfragments/2675.fixed.md
Outdated
| @@ -0,0 +1 @@ | |||
| `impl FromPyObject for [T; N]` will accept anything passing `PySequence_Check`, e.g. NumPy arrays, in the same way that `impl FromPyObject for Vec<T>` was modified after [#2477](https://github.com/PyO3/pyo3/pull/2477). | |||
There was a problem hiding this comment.
Given this is a revert, I think it'd be quite handy to state what version the behaviour is reverting to rather than the offending PR. Users who want to see the PR can start with the link from the CHANGELOG to this PR and follow the discussion thread.
E.g. "Fix regression of impl FromPyObject for [T; N] no longer accepting types passing PySequence_Check, e.g. NumPy arrays, since 0.17.0."
There was a problem hiding this comment.
Adjusted the wording, please re-review.
adamreichold
force-pushed
the
relax-extract-array
branch
from
4c48977
to
f456ed7
Compare
I don't think we have any other bug fixes ready that are blocked only on review. Everything else seems to be enhancements. We do have two undiagnosed crash bugs in #2611 and https://github.com/PyO3/pyo3/discussions/26778 though. Personally, I am not sure if this fix here alone warrants a point release but then again it is a regression. (Turning NumPy arrays into fixed-size Rust arrays seems a bit fringe as NumPy arrays are relatively inefficient for small arrays which is where fixed-size Rust arrays are usually applied.) |
|
Ok. Let's hold off for the moment; Python 3.11.0 is due at the end of the month so maybe we can target 0.17.3 for then once we can confirm PyO3 officially supports it. (Unless we can land some bigger features worthy of calling it 0.18!) |
Closes #2674