…4349)

* Improve error messages for #[pyfunction] defined inside #[pymethods]

Make error message more specific when `#[pyfunction]` is used in
`#[pymethods]`.

Effectively, this replaces the error message:

```
error: static method needs #[staticmethod] attribute
```

To:
```
functions inside #[pymethods] do not need to be annotated with #[pyfunction]
```

...and also removes the other misleading error messages to the function in
question.

Fixes #4340

Co-authored-by: László Vaskó <1771332+vlaci@users.noreply.github.com>

* review fixes

---------

Co-authored-by: László Vaskó <1771332+vlaci@users.noreply.github.com>

* Update dict.get_item binding to use PyDict_GetItemRef

Refs #4265

* test: add test for dict.get_item error path

* test: add test for dict.get_item error path

* test: add test for dict.get_item error path

* fix: fix logic error in dict.get_item bindings

* update: apply david's review suggestions for dict.get_item bindings

* update: create ffi::compat to store compatibility shims

* update: move PyDict_GetItemRef bindings to spot in order from dictobject.h

* build: fix build warning with --no-default-features

* doc: expand release note fragments

* fix: fix clippy warnings

* respond to review comments

* Apply suggestion from @mejrs

* refactor so cfg is applied to functions

* properly set cfgs

* fix clippy lints

* Apply @davidhewitt's suggestion

* deal with upstream deprecation of new_bound

* fix: updated pyclass heighten check to validate for eq and ord, fixing Ok issue in eq implementation.

identified similar issues in Complex enum and tuple enum,  resolved serveral cases, but working on current error coming from traits not being in scope

* fix: fully qualified clone and from calls for enums.

* update: added changelog entry

---------

Co-authored-by: MG <mg@rookie.ninja>

#4382)

* fix incorrect spans on `ret` and `py` local varianbles in emitted code

* add newsfragment

Fixes #4394.

Fixes #4400

As the return value is ultimately communicated back via a StopIteration
exception instance, a peculiar behavior of `PyErr::new` is encountered
here: when the argument is a tuple `arg`, it is used to construct the
exception as if calling from Python `Exception(*arg)` and not
`Exception(arg)` like for every other type of argument.

This comes from from CPython's `PyErr_SetObject` which ultimately calls
`_PyErr_CreateException` where the "culprit" is found here:
https://github.com/python/cpython/blob/main/Python/errors.c#L33

We can fix this particular bug in the invocation of `PyErr::new` but it
is a more general question if we want to keep reflecting this somewhat
surprising CPython behavior, or create a better API, introducing a
breaking change.

* Add PyList_GetItemRef bindings and compat shim

* Use PyList_GetItemRef in list.get_item()

* add release notes

* apply code review comments

* Update newsfragments/4410.added.md

Co-authored-by: David Hewitt <mail@davidhewitt.dev>

* apply `all()` doc cfg hints

---------

Co-authored-by: David Hewitt <mail@davidhewitt.dev>

From now you cannot initialize a `PyClassInitializer<SubClass>` with `PyClassInitializer::from(Py<BaseClass>).add_subclass(SubClass)`.

This was out of bounds write. Now it panics. See details at #4452.

* ffi: define compat for `Py_NewRef` and `Py_XNewRef`

* add missing inline hint

Co-authored-by: Nathan Goldbaum <nathan.goldbaum@gmail.com>

* don't use std::ffi::c_int (requires MSRV 1.64)

* add test to guard against ambiguity

* fix `Py_NewRef` cfg on PyPy

---------

Co-authored-by: Nathan Goldbaum <nathan.goldbaum@gmail.com>

* Use vectorcall (where possible) when calling Python functions

This works without any changes to user code.

The way it works is by creating a methods on `IntoPy` to call functions, and specializing them for tuples.

This currently supports only non-kwargs for methods, and kwargs with somewhat slow approach (converting from PyDict) for functions. This can be improved, but that will require additional API.

We may consider adding more impls IntoPy<Py<PyTuple>> that specialize (for example, for arrays and `Vec`), but this i a good start.

* Add vectorcall benchmarks

* Fix Clippy (elide a lifetime)

---------

Co-authored-by: David Hewitt <mail@davidhewitt.dev>

fixes #4470

…4462)

* ci: swap .python-version file for explicit versions in actions yml

* pin valgrind job to 3.12.4

)

* type_object: fix new clippy complaint about length of doc comment

* all: replace minor version specific links to CPython docs

* ci: fix nightly warning about unreachable pattern

* add comment