Improvements to property macros #760
Conversation
|
The CI failures are from new clippy lints from the newer rust version. I've added a PR #761 to fix those separately. |
| T: IntoPy<PyObject> + Clone, | ||
| { | ||
| fn get_property_value(&self, py: Python) -> PyObject { | ||
| (*self).clone().into_py(py) |
There was a problem hiding this comment.
&self is &&T, so self.clone() makes &T. We need to deref like (*self).clone() to get a T.
|
Thank you! |
|
|
||
| impl<T> GetPropertyValue for &T | ||
| where | ||
| T: IntoPy<PyObject> + Clone, |
There was a problem hiding this comment.
I was wondering whether the Clone bound should really be Copy, so that we don't let people accidentally do really expensive properties with #[pyo3(get)] (such as Vec<T>).
Users can always write their own #[getter] functions for types which aren't Copy.
|
Thanks, I've made some changes to address your comments. I suggest we wait for #761 to merge first. After that I'll fix up the merge conflicts in this PR and then we can potentially merge this. |
|
Thank you! |
|
Thanks for fixing the merge conflicts! |
There are two main fixes that I've added to the property macro code:
#[pyo3(get)]now works withPyObjectfields#[setter]functions can now take aPythonargumentAfter implementing the first I'd changed enough of the code such that cleaning it up caused the second to be fixed trivially.
I also bumped the minimum Rust version to
1.42.0-nightly 2020-01-21, to make use of the recently stabilized#![feature(slice_patterns)].